Prospective evaluation of deep learning image reconstruction for Lung-RADS and automatic nodule volumetry on ultralow-dose chest CT

Seung-Jin Yoo; Young Sik Park; Hyewon Choi; Da Som Kim; Jin Mo Goo; Soon Ho Yoon

doi:10.1371/journal.pone.0297390

. 2024 Feb 22;19(2):e0297390. doi: 10.1371/journal.pone.0297390

Prospective evaluation of deep learning image reconstruction for Lung-RADS and automatic nodule volumetry on ultralow-dose chest CT

Seung-Jin Yoo ¹, Young Sik Park ², Hyewon Choi ³, Da Som Kim ⁴, Jin Mo Goo ⁵, Soon Ho Yoon ^5,^*

Editor: Gayle E Woloschak⁶

PMCID: PMC10883577 PMID: 38386632

Abstract

Purpose

To prospectively evaluate whether Lung-RADS classification and volumetric nodule assessment were feasible with ultralow-dose (ULD) chest CT scans with deep learning image reconstruction (DLIR).

Methods

The institutional review board approved this prospective study. This study included 40 patients (mean age, 66±12 years; 21 women). Participants sequentially underwent LDCT and ULDCT (CTDIvol, 0.96±0.15 mGy and 0.12±0.01 mGy) scans reconstructed with the adaptive statistical iterative reconstruction-V 50% (ASIR-V₅₀) and DLIR. CT image quality was compared subjectively and objectively. The pulmonary nodules were assessed visually by two readers using the Lung-RADS 1.1 and automatically using a computerized assisted tool.

Results

DLIR provided a significantly higher signal-to-noise ratio for LDCT and ULDCT images than ASIR-V₅₀ (all P < .001). In general, DLIR showed superior subjective image quality for ULDCT images (P < .001) and comparable quality for LDCT images compared to ASIR-V₅₀ (P = .01–1). The per-nodule sensitivities of observers for Lung-RADS category 3–4 nodules were 70.6–88.2% and 64.7–82.4% for DLIR-LDCT and DLIR-ULDCT images (P = 1) and categories were mostly concordant within observers. The per-nodule sensitivities of the computer-assisted detection for nodules ≥4 mm were 72.1% and 67.4% on DLIR-LDCT and ULDCT images (P = .50). The 95% limits of agreement for nodule volume differences between DLIR-LDCT and ULDCT images (-85.6 to 78.7 mm³) was similar to the within-scan nodule volume differences between DLIR- and ASIR-V₅₀-LDCT images (-63.9 to 78.5 mm³), with volume differences smaller than 25% in 88.5% and 92.3% of nodules, respectively (P = .65).

Conclusion

DLIR enabled comparable Lung-RADS and volumetric nodule assessments on ULDCT images to LDCT images.

Introduction

Large-scale randomized controlled trials have proven that lung cancer screening using low-dose chest CT (LDCT) scans reduces lung cancer mortality by 20–33% in high-risk groups [1,2]. Several countries have implemented LDCT-based lung cancer screening programs, increasing the use of LDCT scans, which are assessed in terms of the Lung Imaging Reporting and Data System (Lung-RADS) categorization or nodule volumetry [3,4]. LDCT scans are also widely used to evaluate various lung diseases since LDCT is easily accessible while providing similar diagnostic performance and less radiation exposure compared to standard-dose CT scans [5]. A recent study by Sakane et al. reported no damage of human DNA from a single low-dose CT scan [6]. However, lung cancer CT screening and pulmonary disease evaluations may often involve the frequent use of follow-up CT scans [7], and the safety of the cumulative radiation exposure remains underexplored.

Ultra-low-dose chest CT (ULDCT) scans are a potential option to reduce radiation exposure. ULDCT scans of the thorax deliver a lower effective dose (0.20–0.49 mSv) than LDCT scans (about 3 mSv) [8–11]. However, ULDCT scans are inevitably accompanied by higher image noise than LDCT scans, hampering image interpretation. Various image reconstruction methods, including model-based iterative reconstruction, have been attempted for ULDCT scans to mitigate the image noise compromise. Recently, a deep learning image reconstruction (DLIR) system was commercialized by GE (TrueFidelity, GE Healthcare), based on a vast amount of training data of standard-dose phantom and patient CT images reconstructed by filtered back projection [12]. Indeed, a few studies have reported that DLIR could provide better image quality of thoracic LDCT and ULDCT scans than the pre-existing iterative reconstruction method [13–15]. Nevertheless, whether DLIR enables Lung-RADS classification and volumetric evaluation of pulmonary nodules in ULDCT scans remains unknown.

Our study aimed to prospectively evaluate whether Lung-RADS and volumetric nodule assessments were feasible in DLIR-ULD CT scans.

Materials and methods

The Institutional Review Board of Seoul National University Hospital approved this prospective study, and medical staff obtained written informed consent from all participants. The study protocol was registered at the Clinical Research Information Service (CRIS, Registration Number: KCT0004692).

Study population

The inclusion criteria were adults who visited the outpatient respiratory clinic for the first time and planned to undergo a chest CT scan at a single tertiary hospital from February 2020 through July 2020. We excluded patients with a) a body mass index over 30 kg/m², b) an inability to hold their breath sufficiently for chest CT scanning, c) a previous history of procedures or operations that may affect the image quality of CT scans (e.g., central venous catheter, defibrillator implantation, valve replacement) and d) contraindications for CT scans (e.g., pregnancy). 40 patients (mean age ± standard deviation, 66±12; 21 women) were included in this study.

CT acquisition

Consecutive full-inspiratory thoracic LDCT scans at 120 kVp and ULDCT at 100 kVp were acquired in all patients using a single CT machine (Revolution CT; GE Healthcare, Waukesha, WI, USA) and the scanning parameters were as follows: tube voltage, 120 kVp with a mAs of 25 for LDCT, 100 kVp with a mAs of 5 for ULDCT; automatic tube current modulation; gantry rotation time, 280 ms; detector configuration, 128 × 0.625 mm; beam pitch, 1.53; matrix, 512x512; reconstruction increment and section thickness, 1.25 mm; noise index for LDCT, 28 and noise index for ULDCT, 33. Each CT scan was reconstructed with adaptive statistical iterative reconstruction V with 50% blending with FBP (ASIR-V₅₀), considering the results of previous studies with various ASIR-V blending levels [16,17], and DLIR-high level (DLIR-H, TrueFidelity). Thus, four CT image series were generated per patient, resulting in 160 CT series.

Objective and subjective image quality assessment

The objective CT image quality was calculated using the signal-to-noise ratio (SNR) on a picture archiving and communication system (PACS). Two thoracic radiologists (S.H.Y and S.J.Y.; 16 and 7 years of experience in chest CT interpretation) independently drew six circular regions of interest (ROIs) in the lung parenchyma while avoiding vessels or visible pathology on DLIR-LDCT and DLIR-ULDCT images. The ROIs were placed in bilateral upper, mid, and lower lungs with similar sizes of 50–100 mm². Then, the ROIs were copied and pasted into the LDCT and ULDCT with ASIR-V50 images. The mean and standard deviation of the ROIs in Hounsfield units (HU) were recorded and represented the signal and noise, respectively. The SNR was calculated as follows: SNR = |Mean HU_ROI| / SD_ROI.

Two board-certified thoracic radiologists (H.C. and D.S.K.; 7 years of experience in chest CT interpretation) visually examined the paired LDCT and ULDCT images reconstructed with ASIR-V₅₀ and DLIR-H in a random order. The captured CT images were prepared at the aortic arch level, right middle lobar bronchus level, and right diaphragm level. The pulmonary lesion images were also provided if they existed. The CT images were captured in the lung window setting (window level, -550 HU; window width, 1700 HU), and great vessel–level images were also provided in the mediastinal window setting (window level, 45 HU; window width, 450 HU). The radiologists subjectively scored the image quality regarding image noise, visibility of small structures, lesion conspicuity, and diagnostic acceptability (S1 Table). They were blinded to the radiation dose information and reconstruction algorithm.

Nodule detection and Lung-RADS evaluation by readers

One board-certified thoracic radiologist (S.J.Y.) reviewed 40 DLIR-LDCT scans thoroughly and detected nodules of Lung-RADS 1.1 category 3 or over for gold standard. The latter two thoracic radiologists (H.C. and D.S.K.) independently reviewed pairs of 40 DLIR-LDCT and 40 DLIR-ULDCT series on the PACS using Lung-RADS 1.1 in random order. The radiologists detected nodules of Lung-RADS 1.1 category 3 or over as follows: solid nodules with a mean diameter ≥6 mm, part-solid nodules with a total diameter ≥6 mm with a solid component <6 mm, nonsolid nodules with a diameter ≥30 mm.

Volumetric nodule evaluation

Nodule volumetry was conducted using the vendor-provided Lung VCAR (volume computerized assisted reporting, version 14.0–8.11, GE Healthcare), which is an automatic CT lung nodule detection tool designed to detect and quantify nodules to assist the readers. Two radiologists (S.H.Y. and S.J.Y.) prepared gold-standard nodules in consensus with a maximal axial diameter ≥4 mm and recorded the CT characteristics of the nodules (solid, subsolid, calcified). Twelve patients’ CT scans were excluded due to multiple lung nodules (more than 20) due to pulmonary tuberculosis (n = 7), non-tuberculous lung disease (n = 4), and hematolymphangitic metastasis (n = 1). Of the remaining 28 patients, eight patients had no lung nodules ≥4 mm in diameter. As a result, 43 lung nodules (solid, 18; subsolid, 7; calcified, 18) from 20 patients were included in the VCAR evaluation (Fig 1). VCAR produced the maximum x-axis diameter and volume of the nodules (Fig 2). The Lung-RADS 1.1 category by volume of each nodule was also recorded.

Fig 1 — VCAR = volume computerized assisted reporting.

Fig 2 — Ultralow-dose chest CT scan with deep learning image reconstruction-high level of a 60-year-old female patient with a 1.7-cm solid nodule with a spiculated margin at the right upper lobe (A). Lung volume computerized assisted reporting (VCAR) software automatically detected the right upper lobe nodule and depicted it as a red dot on three-dimensional air maximal intensity projection view (B) and axial maximum intensity projection view (C). The nodule was segmented (D), reconstructed as three-dimensional volume rendering image (E), and the x-, y-, and z-axis diameter, volume, and mean attenuation of the nodule were calculated.

To evaluate the reliability of Lung VCAR’s nodule detection and volume measurement ability, we additionally applied LuCAS (Monitor Corporation, Seoul, Korea), a deep-learning based CAD for nodule detection and segmentation, on LDCT scans with ASIR-V₅₀ reconstruction and compared the nodule detection sensitivity and measured volume of the 43 lung nodules.

Statistical analysis

The objective data are expressed as means ± standard deviations. For objective and subjective image quality assessment, the LDCT and ULDCT scans—each with two reconstruction algorithms—were compared using repeated-measures analysis of variance and post-hoc pairwise comparisons after Bonferroni correction. A Bonferroni-corrected p value of <0.008 (0.05/6) was considered to indicate statistical significance. Additionally, for the subjective image quality assessment, interobserver agreement was calculated using intraclass correlation coefficients (ICCs) with a two-way model on each quality assessment subject. An ICC of 0.76–1.0 indicates excellent agreement; 0.40–0.75, fair to good agreement; < 0.40, poor agreement [18]. Reader agreement between LDCT and ULDCT for the Lung-RADS 1.1 nodule classification was evaluated using Cohen’s kappa: < 0.21, slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; 0.81–1.00, almost perfect agreement [19,20]. The McNemar test was conducted to compare the nodule detection sensitivities of each radiologist and VCAR between DLIR-LDCT and DLIR-ULDCT and to compare the nodule detection sensitivity of VCAR and LuCAS. Bland-Altman plots were used to evaluate differences in nodule volume and diameters among ASIR-V₅₀-LDCT, DLIR-LDCT, and DLIR-ULDCT, and the chi-square statistic was used to compare the proportion of nodules with changes of volume more than 25% or diameter more than 1.5 mm. All statistical analyses were performed using MedCalc software version 20.011 (MedCalc Software Ltd, Belgium).

Results

Demographics

Table 1 shows the characteristics of the study participants. The mean and standard deviation of body mass index was 23.7±3.3 kg/m² (Table 1). The mean and standard deviation of the CTDIvol and DLP values were 0.96±0.15 mGy and 40.53±6.04 mGy·cm for the LDCT scans and 0.12±0.01 mGy and 5.09±0.40 mGy·cm for the ULDCT scans, respectively. When a conversion factor of 0.014 mSv/mGy·cm was applied, the mean effective doses were 0.57±0.08 mSv and 0.07±0.01 mSv for LDCT and ULDCT, respectively [21].

Table 1. Characteristics of study participants.

Variables	All Participants (n = 40)
Age (y) ^*	66±12
Sex ^†
Women	21 (53)
Men	19 (47)
Height (cm)^*	161.5±8.0
Weight (kg)^*	62.0±11.2
Body mass index (kg/m2)^*	23.7±3.3

Open in a new tab

* Data are means ± standard deviations.

^† Data are number of patients with percentage in parentheses.

Objective and subjective image quality assessment

The results of the objective image quality assessment are presented in Table 2. DLIR-H provided significantly lower image noise and higher SNR in both LDCT and ULDCT scans than ASIR-V₅₀ (all P < .001). DLIR-LDCT showed significantly lower noise and significantly higher SNR than DLIR-ULDCT (both P < .001).

Table 2. Results of objective and subjective image quality assessment.

	LDCT		ULDCT			P-value of paired comparisons^*
	ASIR_V50	DLIR-H	ASIR_V50	DLIR-H	p-value	ASIR_V50-LDCT vs. DLIR-LDCT	ASIR_V50-ULDCT vs. DLIR-ULDCT	DLIR-LDCT vs. DLIR-ULDCT
Objective analysis
Signal (HU)	-887.9±38.6	-885.11±38.4	-860.3±93.1	-872.12±47.7	< .001	< .001	.01	< .001
Noise (HU)	26.9±12.5	18.9±12.7	49.1±42.3	30.8±4.8	< .001	< .001	< .001	< .001
Signal-to-noise ratio	34.8±7.1	51.2±13.1	21.5±6.4	29.1±5.4	< .001	< .001	< .001	< .001
Subjective analysis
Image noise	4.6±0.5	5.0±0.2	3.0±0.9	3.7±0.8	< .001	.01	< .001	< .001
Visibility of small structures	4.6±0.9	4.7±0.9	2.4±1.1	3.3±0.9	< .001	1	< .001	< .001
Lesion conspicuity	4.7±0.7	4.9±0.2	2.8±1.3	3.3±1.2	< .001	1	.14	< .001
Diagnostic acceptability	4.8±0.4	5.0±0.0	3.3±0.74	3.8±0.8	< .001	.18	< .001	< .001

Open in a new tab

Data are mean ± standard deviation.

* Bonferroni corrected p value < .008 indicated a statistically significant difference.

LDCT = low-dose chest computed tomography, ULDCT = ultralow-dose chest computed tomography, ASIR-V = adaptive statistical iterative reconstruction-V, DLIR-H = deep-learning image reconstruction-high level, HU = Hounsfield units.

Table 2 also summarizes the results of the subjective image quality assessment. In both LDCT and ULDCT, DLIR-H yielded better results in terms of image noise, small structure visibility, lesion conspicuity, and diagnostic acceptability than ASIR-V₅₀ (Figs 3 and 4). Nevertheless, regarding LDCT, image noise (P = .01), visibility of small structures (P = 1), lesion conspicuity (P = 1), and diagnostic acceptability (P = .18) did not significantly differ between ASIR-V₅₀ and DLIR-H. In the ULDCT scans, the image noise, visibility of small structures, and diagnostic acceptability were significantly superior in DLIR-H compared to ASIR-V₅₀ (all P < .001). When comparing LDCT and ULDCT scans reconstructed by DLIR-H, LDCT showed significantly superior results in all categories (all P < .001). The ICC for interobserver agreement was excellent for image noise and diagnostic acceptability (0.82 and 0.86, respectively) and fair to good for visibility of small structures and lesion conspicuity (0.67 and 0.73, respectively).

Fig 3 — A 72-year-old female patient with a 1.2-cm subsolid nodule in the left upper lobe underwent low-dose and ultralow-dose chest CT scans. Each CT scan was reconstructed with adaptive statistical iterative reconstruction-V 50% (ASIR-V₅₀) and deep-learning image reconstruction (DLIR). DLIR showed superior lesion conspicuity and small vessel visibility compared to the ASIR-V₅₀ reconstruction algorithm in both low-dose and ultralow-dose chest CT scans.

Fig 4 — A 80-year-old male patient with small amount of pericardial effusion at anterior superior aortic recess (arrow). The boundaries of anterior superior aortic recess is well defined in the DLIR compared to ASIR-V₅₀ reconstructed images in both low-dose and ultralow-dose chest CT scans. The boundaries of anterior superior aortic recess is indistinguishable with the aorta and main pulmonary artery in ULDCT scan with ASIR-V₅₀.

Nodule detectability and Lung-RADS categorization by readers

The radiologists’ sensitivity for nodule detection is summarized in Table 3. Of the 80 CT scans, which included 40 patients’ LDCT and ULDCT scans with DLIR-H reconstruction, 17 nodules belonged to Lung-RADS 1.1 category 3 or over: 10 category 3 nodules, five category 4A nodules, one category 4B nodule, and one category 4X nodule (Fig 1). Radiologists’ nodule detection sensitivities did not significantly differ between DLIR-LDCT and DLIR-ULDCT (Reader 1, P = 1; Reader 2, P = 1).

Table 3. Nodule detection sensitivity of radiologists.

	DLIR-LDCT	DLIR-ULDCT
Per-patient sensitivity
Reader 1	69.2% (9/13)	61.5% (8/13)
Reader 2	100% (13/13)	100% (13/13)
Per- patient specificity
Reader 1	96.3% (26/27)	96.3% (26/27)
Reader 2	55.6% (15/27)	66.7% (18/27)
Per-nodule sensitivity
Reader 1	70.6% (12/17)	64.7% (11/17)
Reader 2	88.2% (15/17)	82.4% (14/17)
Per-nodule sensitivity for subsolid nodules
Reader 1	100% (7/7)	85.7% (6/7)
Reader 2	71.4% (5/7)	71.4% (5/7)

Open in a new tab

Data in parentheses are the number of patients or nodules.

LDCT = low-dose chest computed tomography, ULDCT = ultralow-dose chest computed tomography, DLIR-H = deep-learning image reconstruction-high level.

In terms of Lung-RADS 1.1 categories, the two radiologists did not detect or down-categorized three and two nodules each in DLIR-ULDCT compared to in DLIR-LDCT (S2 Table). Two nodules were up categorized from category 4A to 4B in DLIR-ULDCT compared to DLIR-LDCT by reader 2 (S2 Table). The reader agreement between LDCT and ULDCT was 0.73–0.76 (substantial agreement).

Nodule detectability and Lung-RADS categorization by VCAR

The sensitivity of nodule detection of VCAR is summarized in Table 4. The nodule detection sensitivity by VCAR was equivalent for DLIR-H and ASIR-V₅₀ reconstructions in LDCT (ASIR-V₅₀ vs. DLIR-H, 72.1% vs. 72.1%, P = 1). In the ULDCT scans, the nodule detection sensitivity of DLIR-H was comparable to that of ASIR-V₅₀, (ASIR-V₅₀ vs. DLIR-H, 65.1% vs. 67.4%, P = 1). There was no significant difference in the nodule detection sensitivity of VCAR between the DLIR-LDCT and DLIR-ULDCT images (72.1% and 67.4%, P = .50).

Table 4. Nodule detection sensitivity of Lung VCAR.

	LDCT		ULDCT
Nodules	ASIR-V₅₀	DLIR-H	ASIR-V₅₀	DLIR-H
Total	72.1% (31/43)	72.1% (31/43)	65.1% (28/43)	67.4% (29/43)
Solid	77.8% (14/18)	77.8% (14/18)	72.2% (13/18)	72.2% (13/18)
Subsolid	14.3% (1/7)	14.3% (1/7)	0% (0/7)	14.3% (1/7)
Calcified	88.9% (16/18)	88.9% (16/18)	83.3% (15/18)	83.3% (15/18)

Open in a new tab

Data in parentheses in number of nodules.

VCAR = volume computerized assisted reporting, LDCT = low-dose chest computed tomography, ULDCT = ultralow-dose chest computed tomography, ASIR-V = adaptive statistical iterative reconstruction-V, DLIR-H = deep-learning image reconstruction-high level.

There was no significant difference of nodule detection sensitivity of VCAR and LuCAS (VCAR vs. LuCAS, 72.1% vs. 86.0%, P = .18) (S3 Table). However, LuCAS showed better performance than VCAR in detection of subsolid nodules (VCAR vs. LuCAS, 14.3% vs. 100%, P = .03) (S3 Table).

In total, 26 lung nodules (solid, 11; subsolid, 1; calcified, 14) were detected and volume measured by VCAR in all ASIR-V₅₀-LDCT, DLIR-LDCT, and DLIR-ULDCT reconstruction images. When applying the Lung-RADS 1.1 categorization of noncalcified nodules based on nodule volume, there were 10 category 2 nodules, one category 3 nodule, and one category 4B nodule in the ASIR-V₅₀-LDCT images (Fig 1). The Lung-RADS 1.1 categories of the 12 nodules were exactly the same among ASIR-V₅₀-LDCT, DLIR-LDCT, and DLIR-ULDCT.

Nodule volume and diameter difference by VCAR

Bland-Altman plots between the measurements of the 26 nodules which were detected and volume measured in all ASIR-V50-LDCT, DLIR-LDCT, and DLIR-ULDCT reconstruction images are shown in Figs 5 and S1. The mean measurement differences and 95% limits of agreement of nodule volumetry were 7.3 mm³ (-63.9, 78.5 mm³) between DLIR- and ASIR-V₅₀-LDCT, 3.9 mm³ (-33.9, 41.6 mm³) between DLIR-ULDCT and ASIR-V₅₀-LDCT, and -3.5 mm³ (-85.6, 78.7 mm³) between DLIR-LDCT and DLIR-ULDCT (Fig 5). The distribution of nodule volume and x-axis diameter differences are shown in Figs 6 and S2.

Fig 5 — Compared to low-dose chest CT with adaptive statistical iterative reconstruction-V 50% (ASIR-V₅₀), low-dose chest CT with deep-learning image reconstruction (DLIR) and ultralow-dose chest CT with DLIR showed a mean nodule volume measurement difference of 7.3 mm³ (-63.9, 78.5) (A) and 3.9 mm³ (-33.9, 41.6), respectively (B). In terms of deep learning image reconstruction, low-dose chest CT and ultralow-dose chest CT showed mean nodule volume measurement difference of -3.5 mm³ (-85.6, 78.7) (C). LDCT = low-dose chest computed tomography, ULDCT = ultralow-dose chest computed tomography, ASIR-V₅₀ = adaptive statistical iterative reconstruction-V 50%, DLIR = deep-learning image reconstruction, VCAR = volume computerized assisted reporting.

Fig 6 — Distribution of measurement differences between low-dose chest CT with adaptive statistical iterative reconstruction-V 50% (ASIR-V₅₀) and low-dose chest CT with deep-learning image reconstruction (DLIR) (A, B), low-dose chest CT with ASIR-V₅₀ and ultralow-dose chest CT with DLIR (C, D), and low-dose chest CT and ultralow-dose chest CT with DLIR (E, F). The majority (84.6–92.3%) of the nodules showed measurement differences of volume within 25%. LDCT = low-dose chest computed tomography, ULDCT = ultralow-dose chest computed tomography, ASIR-V50 = adaptive statistical iterative reconstruction-V 50%, DLIR = deep-learning image reconstruction.

17 nodules were detected and volume measured in both Lung VCAR and LuCAS. The mean measurement difference and 95% limits of agreement of nodule volume was 8.0 mm³ (-146.6, 162.5 mm³). Bland-Altman plots between the volume measurements of the nodules is in S3 Fig.

A volume difference greater than 25% occurred in 11.5% of comparisons between DLIR-LDCT and DLIR-ULDCT scans and in 7.7% of comparisons between DLIR- and ASIR-V₅₀-LDCT scans; this proportion was not significantly different (P = .65) (number of nodules with measurement difference of volume within 25% between ASIR-V₅₀-LDCT and DLIR-LDCT, 24/26; between ASIR-V₅₀-LDCT and DLIR-ULDCT, 22/26; and between DLIR-LDCT and DLIR-ULDCT reconstruction, 23/26; P = .39-.68).

Furthermore, 76.9–80.7% of the nodules showed diameter differences equal to or less than 1.5 mm, and the proportions were not significantly different (number of nodules with a measurement difference of the diameter in the x-axis within 1.5 mm between ASIR-V₅₀-LDCT and DLIR-LDCT, 21/26; between ASIR-V₅₀-LDCT and DLIR-ULDCT, 21/26; and between DLIR-LDCT and DLIR-ULDCT reconstruction, 20/26; P = .74–1).

Discussion

DLIR provided better SNR and subjective image scores in both LDCT and ULDCT, than ASIR-V_50. In the reader study, nodule detection sensitivities did not significantly differ between DLIR-LDCT and DLIR-ULDCT (70.6%-88.2% versus 64.7%-82.4%, P = 1). The Lung-RADS 1.1 categories showed substantial agreement between DLIR-LDCT and DLIR-ULDCT for each reader (κ = 0.73–0.76). In terms of automated nodule detection by VCAR, there was no significant difference in nodule detection sensitivities between the DLIR-LDCT and DLIR-ULDCT images (72.1% and 67.4%, P = .50). The frequencies of volume change >25% (11.5%, 3/26) or diameter change >1.5mm (23.1%, 6/26) between DLIR-LDCT and DLIR-ULDCT were similar to those (7.7%, 2/26 and 19.2%, 5/26) that occurred between the two different reconstruction algorithms of LDCT (P = .65 and P = .74). One nodule showed a 112% nodule volume difference between ASiR-V₅₀-LDCT and DLIR-LDCT and a 140% nodule volume difference between ASiR-V₅₀-LDCT and DLIR-ULDCT, but a 13% nodule volume difference between DLIR-LDCT and DLIR-ULDCT. The nodule was the smallest among the 26 nodules. The Lung VCAR measured volume of the nodule was 25 mm³ in ASIR-V₅₀-LD CT, 53 mm³ in DLIR-LDCT, and 60 mm³ in DLIR-ULDCT. A larger measurement error in small-sized nodules by CAD was found in previous studies [16,22,23]. However, the Lung-RADS 1.1 category of this small-sized solid nodule was 2, and despite the large percentage volume difference among ASIR-V₅₀-LD CT, DLIR-LDCT, DLIR-ULDCT, the Lung-RADS 1.1 category did not change. Our results indicate that DLIR-ULDCT provided a comparable nodule detection sensitivity and Lung-RADS categorization to DLIR-LDCT in both the reader study and VCAR analysis.

The Lung-RADS 1.1 categories based on the volume of 12 nodules detected by VCAR were consistent among ASIR-V₅₀-LDCT, DLIR-LDCT, and DLIR-ULDCT. In the observer performance study, of the 17 nodules with Lung-RADS 1.1 category 3 or 4, three and two nodules were down-categorized or not detected in DLIR-ULDCT compared to DLIR-LDCT by readers 1 and 2, respectively. Two nodules were up-categorized in DLIR-ULDCT compared to DLIR-LDCT by reader 2. Therefore, the nodule management was unchanged in 76–82% of cases. The inter- and intra-observer variability of Lung-RADS categorization is a challenge due to the closeness of categories that are subdivided into units of millimeters or millimeters cubed. In real clinical practice, measurement variability is inevitable. In a previous study, interobserver disagreement of Lung-RADS categorization was seen in 29% of cases, and 8% resulted in different patient management [24]. Considering this variability, the agreement of Lung-RADS categories in our study is acceptable.

According to the diagnostic reference level (DRL) guidelines published by the Korea Centers for Disease Control and Prevention in the Ministry of Health and Welfare in 2019, the DRLs for chest radiograph posteroanterior and lateral views were 0.40 mGy and 1.26 mGy; considering the tissue-weighting factor of the lung (0.12), the effective doses are 0.05 mSv and 0.15 mSv, respectively [21]. The effective dose of our ULDCT scans was 0.07±0.01mSv, slightly higher than that of posteroanterior chest radiographs and slightly lower than that of chest lateral views. The radiation dose of ULDCT was one-eighth of the LDCT dose. Therefore, ULDCT with DLIR can be considered in situations requiring frequent follow-up of lung lesions.

There are previous studies of a vendor-specific and vendor-agnostic DLIR to create high quality images of ULDCT [25,26]. In a study by Goto et al., ULDCT with vendor-specific DLIR (AiCE, Canon Medical Systems) showed better nodule CNR and comparable subjective image quality assessment (noise in air, noise in soft-tissue, streak artifact, texture fineness, and overall quality) compared to LDCT with iterative reconstruction in the phantom study [25]. In a subjective evaluation, ULDCT with DLIR of 14 nonobese patients achieved a significantly higher preference by radiologists compared to ULDCT with iterative reconstruction [25]. Hata et al. proved a vendor-agnostic DLIR for noise reduction of ULDCT showed better subjective image quality compared to the ULDCT with iterative reconstruction [26]. Also, nodule detection rate of radiologists improved in ULDCT with DLIR compared to ULDCT with iterative reconstruction [26]. These previous studies and ours both align with the same result that ULDCT with DLIR increases objective and subjective image quality compared to iterative reconstruction algorithms and provides image quality comparable to that of LDCT.

There have been concerns regarding the increased radiation exposure from medical imaging of pediatric populations [27]. Efforts to reduce the radiation dose in pediatric CT imaging continued steadily and dose reduction strategies in CT imaging is expected to dramatically reduce the adverse effects of radiation exposure such as radiation-induced cancers [27,28]. One of the proven dose reduction strategies is DLIR which can minimize image noise while preserving the adequate resolution of the CT scans [28,29]. In a study by Yoon et al., in pediatric non-contrast and contrast-enhanced chest CT (CTDIvol, 1.3 ± 0.5 mGy; DLP 49.0 ± 26.3 mGy × cm; effective dose, 2.2 ± 3.2 mSv), DLIR showed significant noise reduction with increased SNR and CNR and showed better subjective overall image quality and noise compared to iterative reconstruction [29]. This result is comparable to our results and application of DLIR is expected to substantially reduce the radiation dose while maintaining the appropriate image quality for diagnosis in pediatric chest CT scans.

Our study has a few limitations. First, we prospectively included a limited number of patients with respiratory disease in this study during the early pandemic. Second, for automated nodule detection and measurement evaluation, we inevitably excluded patients with more than 20 lung nodules because setting the reference was not practically possible. Although the automated nodule detection tool shows a high nodule detection rate, in CT scans with diffuse nodular disease, a reduced specificity would be inevitable [30]. Therefore, the results of our study may have limited applicability to patients with diffuse nodular diseases. Regarding the volume measurements, we did not allow readers to modify the VCAR’s nodule segmentation. If human readers could further edit the VCAR’s measurements, it would help maintain more consistent volumetric measurements between CT scans. Third, in the reliability evaluation of Lung VCAR, ground-glass nodule detection sensitivity of the VCAR was inferior to LuCAS. However the two software had no significant statistical difference in total lung nodule detection sensitivity. The nodule volume measurement difference between the two software was acceptable (mean ± standard deviation, 8.0 ± 78.9 mm³). Therefore, we determined that using Lung VCAR for CT analysis is reasonable. Lastly, we included non-obese patients to consistently maintain the CT parameters and the CTDIvol of 0.1 mGy for a ULDCT scan as much as possible. Obese patients occupied 28% of the NLST screening population [31] and required a three to five times higher CT exposure than our protocol for a ULDCT scan [32,33]. Up-to-date studies evaluating a ULDCT with DLIR included non-obese patients [13,14] and future study is warranted for the applicability of ULDCT with DLIR for obese patients.

In conclusion, our study demonstrated that DLIR showed better subjective and objective image quality than iterative reconstruction. In terms of the nodule detection rate and Lung-RADS 1.1 categories of the nodules, DLIR-ULDCT was comparable to DLIR-LDCT in both the reader study and VCAR analysis, while the radiation dose was similar to that of chest radiographs. ULDCT with DLIR may be a reasonable option for monitoring pulmonary nodules using Lung-RADS or the VCAR system in patients who require repeated CT scans, with an as low as reasonably achievable chest X-ray dose.

Supporting information

S1 Fig. Bland-Altman plots of the X-axis diameter difference of lung nodules.

Compared to low-dose chest CT with adaptive statistical iterative reconstruction-V 50% (ASIR-V₅₀), low-dose chest CT with deep-learning image reconstruction (DLIR) and ultralow-dose chest CT with DLIR showed mean x-axis diameter differences of 0.4 mm (-4.0, 4.8) (a) and -0.1 mm (-2.2, 2.0), respectively (b). In terms of deep learning image reconstruction, low-dose chest CT and ultralow-dose chest CT showed a mean x-axis diameter difference of -0.5 mm (-5.0, 4.0) (c). LDCT = low-dose chest computed tomography, ULDCT = ultralow-dose chest computed tomography, ASIR-V₅₀ = adaptive statistical iterative reconstruction-V 50%, DLIR = deep-learning image reconstruction, VCAR = volume computerized assisted reporting.

(TIF)

pone.0297390.s001.tif^{(788.2KB, tif)}

S2 Fig. Distribution of measurement differences of the X-axis diameter of lung nodules.

Distribution of measurement differences between low-dose chest CT with adaptive statistical iterative reconstruction-V 50% (ASIR-V₅₀) and low-dose chest CT with deep-learning image reconstruction (DLIR) (a,b), low-dose chest CT with ASIR-V₅₀ and ultralow-dose chest CT with DLIR (c, d), and low-dose chest CT and ultralow-dose chest CT with DLIR-H (e, f). In total, 76.9–80.7% of the nodules showed a measurement difference of the diameter equal to or less than 1.5 mm. LDCT = low-dose chest computed tomography, ULDCT = ultralow-dose chest computed tomography, ASIR-V₅₀ = adaptive statistical iterative reconstruction-V 50%, DLIR = deep-learning image reconstruction.

(TIF)

pone.0297390.s002.tif^{(2.1MB, tif)}

S3 Fig. Bland-Altman plots of the volume difference of lung nodules in ASIR-V₅₀-LDCT between lung VCAR and LuCAS.

ASIR-V₅₀ = adaptive statistical iterative reconstruction-V 50%, LDCT = low-dose chest computed tomography, VCAR = volume computerized assisted reporting.

(TIF)

pone.0297390.s003.tif^{(1.5MB, tif)}

S1 Table. Scoring system of subjective image quality.

(DOCX)

pone.0297390.s004.docx^{(31.4KB, docx)}

S2 Table. Results of the observer performance study for the detection of nodules with Lung-RADS 1.1 Categories 3 and 4.

(DOCX)

pone.0297390.s005.docx^{(34.2KB, docx)}

S3 Table. Nodule detection sensitivity of Lung VCAR and LuCAS applied on low-dose chest ct reconstructed with ASIR-V₅₀.

(DOCX)

pone.0297390.s006.docx^{(32.4KB, docx)}

Data Availability

All relevant data except for chest CT images are within the paper. Chest CT image data cannot be shared publicly because of the private health information policies of participating institutions. Chest CT image data can be shared to researchers who meet the criteria for access to confidential data upon request (contact to: yshoka@snu.ac.kr). The non-author contact information (phone/email/hyperlink) is as follows: phone, 82-2-2072-2266; email, irb@snu.ac.rk; hyperlink, http://hrpp.snuh.org/irb/introirb/_/singlecont/view.do.

Funding Statement

This work was supported by GE Healthcare (Grant number: 06-2020-0300). No author received funding in the form of salary from GE Healthcare. URLs to sponsors’ websites is: https://www.gehealthcare.co.kr/. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

1.National Lung Screening Trial Research T, Aberle, Adams AM, Berg CD, Black WC, Clapp, et al. Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med. 2011;365(5):395–409. Epub 2011/07/01. doi: 10.1056/NEJMoa1102873 ; PubMed Central PMCID: PMC4356534. [DOI] [PMC free article] [PubMed] [Google Scholar]
2.de Koning HJ, van der Aalst CM, de Jong PA, Scholten ET, Nackaerts K, Heuvelmans MA, et al. Reduced Lung-Cancer Mortality with Volume CT Screening in a Randomized Trial. N Engl J Med. 2020;382(6):503–13. Epub 2020/01/30. doi: 10.1056/NEJMoa1911793 . [DOI] [PubMed] [Google Scholar]
3.Kim HY. National lung cancer screening in Korea: introduction and imaging quality control. Journal of the Korean Society of Radiology. 2019;80(5):826–36. doi: 10.3348/jksr.2019.80.5.826 [DOI] [Google Scholar]
4.Pinsky PF. Lung cancer screening with low-dose CT: a world-wide view. Transl Lung Cancer Res. 2018;7(3):234–42. Epub 2018/07/28. doi: 10.21037/tlcr.2018.05.12 ; PubMed Central PMCID: PMC6037972. [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Rusinek H, Naidich DP, McGuinness G, Leitman BS, McCauley DI, Krinsky GA, et al. Pulmonary nodule detection: low-dose versus conventional CT. Radiology. 1998;209(1):243–9. Epub 1998/10/14. doi: 10.1148/radiology.209.1.9769838 . [DOI] [PubMed] [Google Scholar]
6.Sakane H, Ishida M, Shi L, Fukumoto W, Sakai C, Miyata Y, et al. Biological Effects of Low-Dose Chest CT on Chromosomal DNA. Radiology. 2020;295(2):439–45. doi: 10.1148/radiol.2020190389 . [DOI] [PubMed] [Google Scholar]
7.Guo R, Zhang Y, Ma Z, Deng C, Fu F, Hu H, et al. Overuse of follow-up chest computed tomography in patients with incidentally identified nodules suspicious for lung cancer. J Cancer Res Clin Oncol. 2022;148(5):1147–52. Epub 20210708. doi: 10.1007/s00432-021-03692-6 . [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Lee SW, Kim Y, Shim SS, Lee JK, Lee SJ, Ryu YJ, et al. Image quality assessment of ultra low-dose chest CT using sinogram-affirmed iterative reconstruction. Eur Radiol. 2014;24(4):817–26. Epub 2014/01/21. doi: 10.1007/s00330-013-3090-9 . [DOI] [PubMed] [Google Scholar]
9.Paks M, Leong P, Einsiedel P, Irving LB, Steinfort DP, Pascoe DM. Ultralow dose CT for follow-up of solid pulmonary nodules: A pilot single-center study using Bland-Altman analysis. Medicine (Baltimore). 2018;97(34):e12019. Epub 2018/08/26. doi: 10.1097/MD.0000000000012019 ; PubMed Central PMCID: PMC6112944. [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Beregi JP, Greffier J. Low and ultra-low dose radiation in CT: Opportunities and limitations. Diagn Interv Imaging. 2019;100(2):63–4. Epub 2019/02/09. doi: 10.1016/j.diii.2019.01.007 . [DOI] [PubMed] [Google Scholar]
11.Lee SC, Yoon SH, Goo JM, Yim JJ, Kim CK. Submillisievert Computed Tomography of the Chest in Contact Investigation for Drug-Resistant Tuberculosis. J Korean Med Sci. 2017;32(11):1779–83. doi: 10.3346/jkms.2017.32.11.1779 ; PubMed Central PMCID: PMC5639057. [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Hsieh J, Liu E, Nett B, Tang J, Thibault J-B, Sahney S. A new era of image reconstruction: TrueFidelity™. Technical white paper on deep learning image reconstruction GE Healthcare. 2019. [Google Scholar]
13.Jiang B, Li N, Shi X, Zhang S, Li J, de Bock GH, et al. Deep Learning Reconstruction Shows Better Lung Nodule Detection for Ultra-Low-Dose Chest CT. Radiology. 2022;303(1):202–12. doi: 10.1148/radiol.210551 . [DOI] [PubMed] [Google Scholar]
14.Wang H, Li LL, Shang J, Song J, Liu B. Application of deep learning image reconstruction in low-dose chest CT scan. Br J Radiol. 2022;95(1133):20210380. doi: 10.1259/bjr.20210380 . [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Kim JH, Yoon HJ, Lee E, Kim I, Cha YK, Bak SH. Validation of Deep-Learning Image Reconstruction for Low-Dose Chest Computed Tomography Scan: Emphasis on Image Quality and Noise. Korean J Radiol. 2021;22(1):131–8. doi: 10.3348/kjr.2020.0116 ; PubMed Central PMCID: PMC7772377. [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Lee HN, Kim JI, Shin SY. Measurement accuracy of lung nodule volumetry in a phantom study: Effect of axial-volume scan and iterative reconstruction algorithm. Medicine (Baltimore). 2020;99(23):e20543. Epub 2020/06/06. doi: 10.1097/MD.0000000000020543 ; PubMed Central PMCID: PMC7306330. [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Rotzinger DC, Racine D, Beigelman-Aubry C, Alfudhili KM, Keller N, Monnin P, et al. Task-Based Model Observer Assessment of A Partial Model-Based Iterative Reconstruction Algorithm in Thoracic Oncologic Multidetector CT. Sci Rep. 2018;8(1):17734. Epub 2018/12/12. doi: 10.1038/s41598-018-36045-4 ; PubMed Central PMCID: PMC6286352. [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Cicchetti DV. Guidelines, criteria, and rules of thumb for evaluating normed and standardized assessment instruments in psychology. Psychological Assessment. 1994;6(4):284–90. doi: 10.1037/1040-3590.6.4.284 [DOI] [Google Scholar]
19.Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977;33(1):159–74. Epub 1977/03/01. [PubMed] [Google Scholar]
20.Viera AJ, Garrett JM. Understanding interobserver agreement: the kappa statistic. Fam med. 2005;37(5):360–3. [PubMed] [Google Scholar]
21.Deak PD, Smal Y, Kalender WA. Multisection CT protocols: sex- and age-specific conversion factors used to determine effective dose from dose-length product. Radiology. 2010;257(1):158–66. Epub 2010/09/21. doi: 10.1148/radiol.10100047 . [DOI] [PubMed] [Google Scholar]
22.Liang M, Yip R, Tang W, Xu D, Reeves A, Henschke CI, et al. Variation in Screening CT-Detected Nodule Volumetry as a Function of Size. AJR Am J Roentgenol. 2017;209(2):304–8. Epub 2017/06/02. doi: 10.2214/AJR.16.17159 . [DOI] [PubMed] [Google Scholar]
23.Peters AA, Christe A, von Stackelberg O, Pohl M, Kauczor HU, Heußel CP, et al. "Will I change nodule management recommendations if I change my CAD system?"-impact of volumetric deviation between different CAD systems on lesion management. Eur Radiol. 2023;33(8):5568–77. Epub 2023/03/10. doi: 10.1007/s00330-023-09525-z ; PubMed Central PMCID: PMC10326095. [DOI] [PMC free article] [PubMed] [Google Scholar]
24.van Riel SJ, Jacobs C, Scholten ET, Wittenberg R, Winkler Wille MM, de Hoop B, et al. Observer variability for Lung-RADS categorisation of lung cancer screening CTs: impact on patient management. Eur Radiol. 2019;29(2):924–31. Epub 2018/08/02. doi: 10.1007/s00330-018-5599-4 ; PubMed Central PMCID: PMC6302878. [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Goto M, Nagayama Y, Sakabe D, Emoto T, Kidoh M, Oda S, et al. Lung-Optimized Deep-Learning-Based Reconstruction for Ultralow-Dose CT. Acad Radiol. 2023;30(3):431–40. Epub 2022/06/24. doi: 10.1016/j.acra.2022.04.025 . [DOI] [PubMed] [Google Scholar]
26.Hata A, Yanagawa M, Yoshida Y, Miyata T, Tsubamoto M, Honda O, et al. Combination of Deep Learning-Based Denoising and Iterative Reconstruction for Ultra-Low-Dose CT of the Chest: Image Quality and Lung-RADS Evaluation. AJR Am J Roentgenol. 2020;215(6):1321–8. Epub 2020/10/15. doi: 10.2214/AJR.19.22680 . [DOI] [PubMed] [Google Scholar]
27.Miglioretti DL, Johnson E, Williams A, Greenlee RT, Weinmann S, Solberg LI, et al. The use of computed tomography in pediatrics and the associated radiation exposure and estimated cancer risk. JAMA Pediatr. 2013;167(8):700–7. Epub 2013/06/12. doi: 10.1001/jamapediatrics.2013.311 ; PubMed Central PMCID: PMC3936795. [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Nagayama Y, Sakabe D, Goto M, Emoto T, Oda S, Nakaura T, et al. Deep Learning-based Reconstruction for Lower-Dose Pediatric CT: Technical Principles, Image Characteristics, and Clinical Implementations. Radiographics. 2021;41(7):1936–53. Epub 2021/10/02. doi: 10.1148/rg.2021210105 . [DOI] [PubMed] [Google Scholar]
29.Yoon H, Kim J, Lim HJ, Lee MJ. Image quality assessment of pediatric chest and abdomen CT by deep learning reconstruction. BMC Med Imaging. 2021;21(1):146. Epub 2021/10/12. doi: 10.1186/s12880-021-00677-2 ; PubMed Central PMCID: PMC8503996. [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Kim H, Kim HY, Goo JM, Kim Y. Lung Cancer CT Screening and Lung-RADS in a Tuberculosis-endemic Country: The Korean Lung Cancer Screening Project (K-LUCAS). Radiology. 2020;296(1):181–8. doi: 10.1148/radiol.2020192283 . [DOI] [PubMed] [Google Scholar]
31.Zhao J, Barta JA, McIntire R, Shusted C, Zeigler-Johnson C, Juon HS. Racial difference in BMI and lung cancer diagnosis: analysis of the National Lung Screening Trial. BMC Cancer. 2022;22(1):797. doi: 10.1186/s12885-022-09888-4 ; PubMed Central PMCID: PMC9297592. [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Svahn TM, Sjoberg T, Ast JC. Dose estimation of ultra-low-dose chest CT to different sized adult patients. Eur Radiol. 2019;29(8):4315–23. doi: 10.1007/s00330-018-5849-5 . [DOI] [PubMed] [Google Scholar]
33.Guo X, Jia D, He L, Jia X, Zhang D, Dou Y, et al. Evaluation of ultralow-dose computed tomography on detection of pulmonary nodules in overweight or obese adult patients. J Appl Clin Med Phys. 2022;23(4):e13589. doi: 10.1002/acm2.13589 ; PubMed Central PMCID: PMC8992951. [DOI] [PMC free article] [PubMed] [Google Scholar]

PLoS One. doi: 10.1371/journal.pone.0297390.r001

Decision Letter 0

Gayle E Woloschak

7 Jul 2023

PONE-D-23-05687Prospective Evaluation of Deep Learning Image Reconstruction for Lung-RADS and Automatic Nodule Volumetry on Ultralow-Dose Chest CTPLOS ONE

Dear Dr. Yoon:

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Revisions were proposed by both reviewers. Please also note that improvements in the figures were recommended for the next revision.

Please submit your revised manuscript by Aug 21 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Gayle E. Woloschak, PhD

Section Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf.

2. Please note that PLOS ONE has specific guidelines on code sharing for submissions in which author-generated code underpins the findings in the manuscript. In these cases, all author-generated code must be made available without restrictions upon publication of the work. Please review our guidelines at https://journals.plos.org/plosone/s/materials-and-software-sharing#loc-sharing-code and ensure that your code is shared in a way that follows best practice and facilitates reproducibility and reuse.

3. Thank you for stating in your Funding Statement:

“This work was supported by GE Healthcare. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.”

Please provide an amended statement that declares *all* the funding or sources of support (whether external or internal to your organization) received during this study, as detailed online in our guide for authors at http://journals.plos.org/plosone/s/submit-now. Please also include the statement “There was no additional external funding received for this study.” in your updated Funding Statement.

Please include your amended Funding Statement within your cover letter. We will change the online submission form on your behalf.

4. Thank you for stating the following in the Competing Interests section:

“Soon Ho Yoon works in the MEDICALIP as a chief medical officer.

The other author do not have any conflicts of interest to declare for this article.”

We note that one or more of the authors are employed by a commercial company: MEDICALIP

A. Please provide an amended Funding Statement declaring this commercial affiliation, as well as a statement regarding the Role of Funders in your study. If the funding organization did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript and only provided financial support in the form of authors' salaries and/or research materials, please review your statements relating to the author contributions, and ensure you have specifically and accurately indicated the role(s) that these authors had in your study. You can update author roles in the Author Contributions section of the online submission form.

Please also include the following statement within your amended Funding Statement.

“The funder provided support in the form of salaries for authors [insert relevant initials], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section.”

If your commercial affiliation did play a role in your study, please state and explain this role within your updated Funding Statement.

B. Please also provide an updated Competing Interests Statement declaring this commercial affiliation along with any other relevant declarations relating to employment, consultancy, patents, products in development, or marketed products, etc.

Within your Competing Interests Statement, please confirm that this commercial affiliation does not alter your adherence to all PLOS ONE policies on sharing data and materials by including the following statement: "This does not alter our adherence to PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors http://journals.plos.org/plosone/s/competing-interests) . If this adherence statement is not accurate and there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared.

Please include both an updated Funding Statement and Competing Interests Statement in your cover letter. We will change the online submission form on your behalf.

5. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: http://journals.plos.org/plosone/s/supporting-information.

Additional Editor Comments (if provided):

One reviewer suggested major revisions, the other suggested minor changes. Please work to address all of these concerns in a revision. Also note that there were concerns about the quality of figures which need to be improved.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This study evaluated whether lung nodule detection, Lung-RADS classification, and volumetric nodule

assessments were feasible with ULD chest CT scans with DLIR. Study included 40 patient cases reviewed by 2 observers with a commercialized assisted tool.

Authors concluded DLIR enabled comparable Lung-RADS and volumetric nodule assessments on

ULDCT images to LDCT images.

1. Study included assessment with a commercialized assisted tool (VCAR) in addition to observers' evaluation. Authors did not state the rationale of including the assessment using a software tool, and the reason why VCAR was introduced in this study is not clear.

If authors intended to use VCAR as a detection and measurement device, and compared nodule assessment performance by VCAR for LDCT and ULDCT scans with IR and DLIR, then the performance reliability of VCAR in terms of nodule detection and volume measurement need to be first assessed before drawing any conclusion based on measurements using that particular device. And yet, I could not find any description on that.

Authors should make it clear the role of VCAR assessment in this study and the way how its reliability was assessed.

2. In Fig. 5 (A) and (B), VCAR measurements already differ btw ASiR-LDCT and DLIR-LDCT quite a lot, which indicate VCAR's performance is not sufficiently reliable to distinguish between LDCT and ULDCT. Authors need to present the reliability range of VCAR's performance in order to justify its use in this study purpose.

3. In Fig. 5 (B) and (C), volume measurement difference of case no. 17 exceeded 100%, which may result in a different patient management and represent a potential risk. And yet, authors did not mention that risk. Study results should be discussed in both positive and negative aspects.

4. M&M section states a total of 43 nodules were assessed with VCAR, but the number is 27 in Fig. 5. Needs to be clarified.

5. Authors stated that there was no significant difference in the nodule detection sensitivity between the DLIR-LDCT and DLIR-ULDCT images. I assume authors used a chi-squire test for calculating the statistical significance, which is not clear to the reviewer. Please explain the way in detail.

6. For nodule detection study with observers, how were the gold standard nodules established ?

Reviewer #2: LDCT reduces lung cancer mortality, but it needs annual/short-term follow-ups which deliver radiation exposures.a ULDCT is 6-10 times less dose than LDCT, and DLIR were shown to provide better image quality. This work attempts to evaluate classification and evaluation of the images, and the authors reached this goal using DLIR for hospital measurements. Given that it is informative on using DLIR for Ultra low dose chest CT, I think readers will benefit from this manuscript for their own practices. Before recommending for publication, I only have a few minor concerns on this manuscript.

Introduction: "better image quality ... than the preexisting iterative reconstruction method (13-15)" suggests one single method but with multiple citations. Do the authors mean multiple methods?

Study population: is the period of Feb-Jul 2020 chosen on purpose, because it overlaps with the onset of COVID-19 pandemic?

Objective and subjective image quality assessment: the authors have four thoracic radiologists, with two for each task. Is the number too small for the assessment, if their assessment results diverge significantly?

Statistical analysis:

(1) For the Bonferroni correction, why is 6 used?

(2) For ICC thresholds are different across different studies, why did the authors choose 0.40 and 0.75? If this is a field-specific choice, can the authors provide citation to the choice?

(3) Does Fleiss' kappa offer a different result, if offered?

Table 1: what are the two values in the brackets of 21 (53) and 19 (47)?

Table 2: * - could the mention of Bonferroni here be more clear for the adopted significance level?

Nodule detectability and Lung-RADS categorization by VCAR: given that there is no significant difference between the two sets of images, would the authors like to mention that DLIR-ULDCT should be preferred given its ultra low dose?

Discussion: nice summary of the findings and the limitations.

Figure 3: can the authors highlight certain features that were not visible with ASIR-V50 but evident with DLIR?

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

**********

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

PLoS One. 2024 Feb 22;19(2):e0297390. doi: 10.1371/journal.pone.0297390.r002

Author response to Decision Letter 0

22 Oct 2023

Reviewer #1: This study evaluated whether lung nodule detection, Lung-RADS classification, and volumetric nodule assessments were feasible with ULD chest CT scans with DLIR. Study included 40 patient cases reviewed by 2 observers with a commercialized assisted tool.

Authors concluded DLIR enabled comparable Lung-RADS and volumetric nodule assessments on ULDCT images to LDCT images.

Authors should make it clear the role of VCAR assessment in this study and the way how its reliability was assessed.

Response: Thank you for the keen comment.

Although there is a variety of software for automated detection and measurement of lung nodules in chest CT scans, the reason why we choose to use Lung VCAR is as follows:

First, among the various software for nodule detection and measurement for Chest CT scans, we thought using FDA-approved VCAR software from the same vendor (GE Healthcare) with our CT scanner could ensure consistency to this study. The purpose of our study is to compare the quality of DLIR compared to ASIR-V50 reconstruction and ULDCT to LDCT rather than evaluate the CAD software’s nodule detection ability and measurement accuracy.

Second, a previous study done by Lee et al. (Medicine 2020;99:23(e20543)) showed acceptable performance of Lung VCAR in nodule volume measurement with absolute percentage volume error around 10% in low-dose and standard-dose chest CT scans reconstructed with FBP and ASIR-V. Liang et al. (AJR 2017; 209:304–308) compared the volume of stable nodules in chest CT scans using two commercial software, Lung VCAR and SyngoVia. Lung VCAR showed a mean volume variation of 0.7% ± 18.6% (range, −39.3% to 83.3%) which was better than SyngoVia. These previous studies supported that Lung VCAR was reliable and became for our purpose.

Third, we initially assessed the performance of nodule detection by Lung VCAR for LDCT and ULDCT scans reconstructed with FBP, ASIR-V50, ASIR-V100 and DLIR and the nodule detection sensitivity showed consistent results, which was sufficient for analysis. The nodule detection sensitivities are listed in the table below.

LDCT ULDCT

Nodules FBP ASIR-V 50 ASIR-V 100 DLIR-H FBP ASIR-V 50 ASIR-V 100 DLIR-H

Total 0.70 (30/43) 0.72 (31/43) 0.74 (32/43) 0.72 (31/43) 0.63 (27/43) 0.65 (28/43) 0.65 (28/43) 0.67 (29/43)

Solid 0.78 (14/18) 0.78 (14/18) 0.78 (14/18) 0.78 (14/18) 0.67 (12/18) 0.72 (13/18) 0.72 (13/18) 0.72 (13/18)

Subsolid 0.00 (0/7) 0.14 (1/7) 0.14 (1/7) 0.14 (1/7) 0.00 (0/7) 0.00 (0/7) 0.00 (0/7) 0.14 (1/7)

Calcified 0.89 (16/18) 0.89 (16/18) 0.94 (17/18) 0.89 (16/18) 0.83 (15/18) 0.83 (15/18) 0.83 (15/18) 0.83 (15/18)

In response to your comment, we further thought about solutions to evaluate Lung VCAR’s reliability. Identifying a gold standard for nodule detection sensitivity and measurement was challenging since every software has limitations. Even in experienced radiologists, inter and intra-observer variability regarding nodule detection and measurement is inevitable.

We have compared VCAR with another commercially available software called LuCAS (Monitor Corporation, Seoul, Korea). LuCAS is a deep learning-based computer-aided diagnosis (DL-CAD) for nodule detection and segmentation in chest CT scans. The nodule detection sensitivity in LDCT with ASIR-V50 reconstruction showed no statistically significant difference between the two software (VCAR vs. LuCAS, 72% vs. 86%, p=.113). However, LuCAS performed better than VCAR in detecting subsolid nodules.

LDCT – ASIR-V50

Nodules VCAR LuCAS

Total 0.72 (31/43) 0.86 (37/43)

Solid 0.78 (14/18) 0.78 (14/18)

Subsolid 0.14 (1/7) 1.00 (7/7)

Calcified 0.89 (16/18) 0.89 (16/18)

The mean ± standard deviation of nodule volume difference between VCAR and LuCAS was 8.0 ± 78.9 mm3. The Bland-Altman plot below shows an acceptable measurement difference between the two software.

Response: Thank you for the comment. The answer to this comment will be presented together in questions 1 and 3.

Response: Thank you for your keen comment.

In the waterfall charts of Figure 5 (B) and (D), one nodule’s volume difference exceeds 100% between ASIR-V50-LDCT and DLIR-LDCT and between ASIR-V50-LDCT and DLIR-ULDCT. The nodule was the smallest among the 26 nodules: a 4.5 mm-sized solid nodule. The Lung VCAR measured nodule volume was 25 mm3 in ASIR-V50-LD CT, 53 mm3 in DLIR-LDCT, and 60 mm3 in DLIR-ULDCT. Since the nodule is a small-sized solid nodule, Lung-RADS 1.1 category was 2 and despite the large percentage volume difference among ASIR-V50-LD CT, DLIR-LDCT, DLIR-ULDCT, the Lung-RADS 1.1 category did not change.

Previous studies also found larger measurement error in small-sized nodules by CAD and manual measurement by radiologists (AJR 2017; 209:304–308. Medicine 2020;99:23(e20543). Eur Radiol 2023;33:5568–5577.). This reaffirms that such inconsistencies aren't solely a limitation of the CAD software but are observed across different measurement techniques, particularly for smaller nodules.

In our study, Lung-RADS 1.1 category of 12 nodules with categories 3 and 4 stayed the same among ASIR-V50-LDCT, DLIR-LDCT, and DLIR-ULDCT. CAD software, including Lung VCAR, has limitations, but if used in the appropriate and specific situations, it would be helpful in clinical practice.

4. M&M section states a total of 43 nodules were assessed with VCAR, but the number is 27 in Fig. 5. Needs to be clarified.

Response: Thank you for pointing out the discrepancy. To clarify, we assessed the nodule detection sensitivity of VCAR for a total of 43 nodules. Out of theses, 26 nodules were consistently detected and their volume were measured by VCAR in all ASIR-V50-LDCT, DLIR-LDCT, and DLIR-ULDCT reconstruction images. Therefore, the comparisons in Lung-RADS 1.1 categorization and nodule volume difference were focused on these 26 nodules. Bland-Altman plots in Figure 4 and waterfall charts in Figure 5 show the 26 nodules’ volume difference. You are right about the number 27 in Figure 5d - it was an oversight on our part and has now been corrected to “26”. We apologize for any confusion this may have caused and appreciate your keen observation.

Response: We used the chi-square test for calculating the nodule detection sensitivity of each radiologist in DLIR-LDCT and DLIR-ULDCT scans. We added the detail in the statistical analysis as follows.

“ The chi-square test was conducted to compare the nodule detection sensitivities of each radiologist in DLIR-LDCT and DLIR-ULDCT.”

6. For nodule detection study with observers, how were the gold standard nodules established ?

Response: The gold standard nodules for nodule detection and Lung-RADS categorization was done by one board-certified radiologist with 7 years of experience in chest CT interpretation. We added the detail in the manuscript as follows.

“ One board-certified thoracic radiologist (***) reviewed 40 DLIR-LDCT scans thoroughly and detected nodules of Lung-RADS 1.1 category 3 or over for gold standard.”

1. Introduction: "better image quality ... than the preexisting iterative reconstruction method (13-15)" suggests one single method but with multiple citations. Do the authors mean multiple methods?

Response: We tried to cite three previous studies showing better image quality of deep-learning image reconstruction compared to the iterative reconstruction method. The citations (13-15) represent different studies, each comparing DLIR with a particular iterative reconstruction method. We apologize for any confusion caused by the wording.

2. Study population: is the period of Feb-Jul 2020 chosen on purpose, because it overlaps with the onset of COVID-19 pandemic?

Response: It was a coincidence. We started to plan this study in December 2019. IRB approval and clinical trial registration were completed in January 2020, before COVID-19 became a pandemic. In our country, the first COVID-19 patient occurred on January 1st, 2020. Due to the unexpected pandemic disease, the patient enrollment period took longer than we expected, but there was no significant impact on our study progress.

3. Objective and subjective image quality assessment: the authors have four thoracic radiologists, with two for each task. Is the number too small for the assessment, if their assessment results diverge significantly?

Response: Thank you for your concern. While we acknowledge that having only two radiologists for each task may seem limited, the reviewers were composed of board-certified thoracic radiologists with sufficient CT interpretation experience. Importantly, ICC showed excellent interobserver agreement for subjective assessment of image noise and diagnostic acceptability (0.82 and 0.86, respectively). Given the strong consistency in their evaluations, we believe their assessments provided meaningful and reliable results for this study.

4. Statistical analysis:

(1) For the Bonferroni correction, why is 6 used?

Response: Bonferroni correction compensates for the multiple comparisons we made in our analysis. For our study, we carried out multiple comparisons across 4 types of CT scans (ASIR-V50-LDCT, ASIR-V50-ULDCT, DLIR-LDCT, DLIR-ULDCT scans). The number of unique pairwise comparisons among these four scans is 6 (calculated as 4C2=6) Hence, we adjusted our significance level as follows:

Bonferroni-corrected p-value = original p-value / number of tests performed

0.008 = 0.05 / 6

(2) For ICC thresholds are different across different studies, why did the authors choose 0.40 and 0.75? If this is a field-specific choice, can the authors provide citation to the choice?

Response: Thank you for the comment. We do agree with the reviewer’s opinion that ICC thresholds differ across studies, therefore reference is necessary. We cited the article written by Domenic V. Cicchetti written in 1994 and added the reference no. 18 for the ICC thresholds in the manuscript.

(3) Does Fleiss' kappa offer a different result, if offered?

Response: Thank you for your keen question. Prior to writing this manuscript, we had statistical consultation. ICC was recommended for the interobserver agreement in subjective image quality assessment of the CT scans, and kappa was not recommended. Since the subjective image quality assessment was a scoring task (ordinal variable), kappa, which is for the nominal variable, was not applicable. Cohen’s Kappa was recommended for the Lung-RADS 1.1 categorization agreement between two radiologists, and Fleiss’ Kappa, which is for interobserver agreement for 3 or more raters, was not applicable.

5. Table 1: what are the two values in the brackets of 21 (53) and 19 (47)?

Response: The data are the number of patients with percentages in parentheses. We indicated in the footnotes as follows.

“ † Data are number of patients with percentage in parentheses.”

6. Table 2: * - could the mention of Bonferroni here be more clear for the adopted significance level?

Response: Thank you for the comment. We amended our footnotes and mentioned Bonferroni as follows.

“ * Bonferroni corrected p value < .008 indicated a statistically significant difference.”

7. Nodule detectability and Lung-RADS categorization by VCAR: given that there is no significant difference between the two sets of images, would the authors like to mention that DLIR-ULDCT should be preferred given its ultra low dose?

Response: We appreciate your observation. There was no significant difference in the nodule detection sensitivity of VCAR between the DLIR-LDCT and DLIR-ULDCT images (72.1% and 67.4%, P=.64). Lung-RADS 1.1 categorization of 12 nodules of category 3, 4 nodules by VCAR were exactly the same. Therefore, DLIR-ULDCT is a scan comparable to DLIR-LDCT. However, in previous studies, it is known that the measurement accuracy of VCAR is poor when the size of the nodule is small, so the accuracy may be low for category 2 nodules. In actual clinical practice, nodule detection CAD for chest CT scans is not popularized, and nodule detection and risk assessment are usually performed by a radiologist. In our observer study, radiologists’ nodule detection sensitivities did not significantly differ between DLIR-LDCT and DLIR-ULDCT (70.6%-88.2% vs. 64.7%-82.4%, P=.57). In terms of Lung-RADS 1.1 categories, the two radiologists did not detect or down-categorized three and two nodules each in DLIR-ULDCT compared to in DLIR-LDCT. Therefore, we advocate DLIR-ULDCT primarily for follow-up scans for monitoring pulmonary nodules rather than initial assessments. We have incorporated this in the conclusion.

8. Figure 3: can the authors highlight certain features that were not visible with ASIR-V50 but evident with DLIR?

Response: DLIR showed superior lesion conspicuity and small vessel visibility in the lung window setting compared to the ASIR-V50 reconstruction algorithm in both low-dose and ultralow-dose chest CT scans. Additionally, in the mediastinal window setting, mediastinal lesions and anatomical structures were more clearly depicted in DLIR compared to ASIR-V50 reconstruction algorithm in both low-dose and ultralow-dose chest CT scans. We have added some figures in Figure 3 to exemplify these points.

Attachment

Submitted filename: Response to Reveiwers.docx

pone.0297390.s007.docx^{(433.8KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0297390.r003

Decision Letter 1

Gayle E Woloschak

8 Nov 2023

PONE-D-23-05687R1Prospective Evaluation of Deep Learning Image Reconstruction for Lung-RADS and Automatic Nodule Volumetry on Ultralow-Dose Chest CTPLOS ONE

Dear Dr. Yoon,

Minor recommendations were submitted by one reviewer. Please address those in a revision.

Please submit your revised manuscript by Dec 23 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.
A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.
An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

We look forward to receiving your revised manuscript.

Kind regards,

Gayle E. Woloschak, PhD

Section Editor

PLOS ONE

Journal Requirements:

Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice.

Additional Editor Comments:

One reviewer had agreed to accept the work, one had suggested minor revisions. Please address the minor revisions in a resubmission.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #2: All comments have been addressed

Reviewer #3: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #2: Yes

Reviewer #3: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: Yes

Reviewer #3: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

Reviewer #2: Yes

Reviewer #3: No

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #2: Yes

Reviewer #3: (No Response)

**********

6. Review Comments to the Author

Reviewer #2: (No Response)

Reviewer #3: This article is well-written and interesting, and it holds clinical importance, given that the authors have revealed that DLIR is suitable for low-dose lung CT.

I have several suggestions that the authors need to address.

Introduction

1. There are significant controversies surrounding the actual risk of low-dose radiation exposure for adults. For instance, a recent study has indicated that LDCT does not induce DNA damage (source: https://doi.org/10.1148/radiol.2020190389). Whether LDCT entails "non-negligible radiation exposure" remains unclear. The description requires modification.

Methods

2. It is unclear how to determine the sample size (n=40) for this prospective study. Please provide clarification.

3. Please specify the number of patients excluded based on each exclusion criterion.

CT acquisition parameters

4. Please include information about the noise index for both LDCT and ULDCT protocols if the authors utilized automated tube current modulation. This crucial information should be presented in the main text rather than the supporting text.

5. Why did the authors need to treat S2, S3, and S4 texts as supporting text? This important information should be included in the main text, unless there are reasonable constraints such as word count limitations.

6. The McNemar test is recommended for intraindividual comparisons of nodule detection sensitivity, rather than the Chi-square test.

Discussion

7. One of the major limitations in this study is that the authors used a vendor-specific DLIR algorithm, and as a result, the applicability to other DLIR algorithms provided by different vendors is not discussed. Recent investigations employed a different vendor-specific DLR for ULDCT of the lung to improve image quality (doi: 10.1016/j.acra.2022.04.025 and doi: 10.2214/AJR.19.22680.). Please provide a concise discussion of whether your results align with these prior investigations to offer readers the wide applicability of study findings.

8. The study findings may be applicable for pediatric CT, where radiation exposure is of concern　in comparison to adults who require lung cancer screening. Discussing the applicability to low-dose pediatric CT by citing the relevant articles (such as doi: 10.1148/rg.2021210105. and doi: 10.1186/s12880-021-00677-2.) would further emphasize the value of the authors' findings.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #2: No

Reviewer #3: No

**********

PLoS One. 2024 Feb 22;19(2):e0297390. doi: 10.1371/journal.pone.0297390.r004

Author response to Decision Letter 1

29 Dec 2023

Response: Thank you for your comment. The study by Sakane et al. proved that a single low-dose CT with a median effective dose of 1.5mSv does not damage human DNA in peripheral blood lymphocytes. While these findings are innovative and potentially impactful, they do not encompass patient groups who undergo annual or more frequent low-dose CT scans.

In clinical practice, repeated low-dose CT scans are common due to various conditions, making difficult to ignore the effects of cumulative radiation exposure in these cases. Therefore, clinicians and radiologists continue to make efforts to minimize radiation exposure in patients.

We amended the description as below:

A recent study by Sakane et al. reported no damage of human DNA from a single low-dose CT scan [6]. However, lung cancer CT screening and pulmonary disease evaluations may often involve the frequent use of follow-up CT scans [7], and the safety of the cumulative radiation exposure remains underexplored.

2. It is unclear how to determine the sample size (n=40) for this prospective study. Please provide clarification.

Response: We assumed that deep-learning image reconstruction in ultra low-dose CT scan group will obtain 90% of satisfactory subjective overall quality and iterative reconstruction in ultra low-dose CT scan group will obtain 65% of satisfactory subjective overall quality. Applying the alpha error =0.05, and beta error =0.20 (power=0.80), the calculated total sample size was 40.

3. Please specify the number of patients excluded based on each exclusion criterion.

Response: Since this study was a prospective cross-sectional study, the patient enrollment was done by asking patients who met the inclusion criteria and did not meet the exclusion criteria their willingness to participate in this study and obtaining their consent at a pulmonology outpatient clinic center about. Therefore, the total number of the all potential participants who met the inclusion criteria or met the exclusion criteria was not separately recorded.

Response: Thank you for the comment. The noise index for LDCT was 28 and for ULDCT was 33. We relocated the S1 text in the main manuscript and added the noise index as follows:

Response: Thank you for the comment. As you suggested we relocated the S2, S3, S4 text within the main manuscript.

6. The McNemar test is recommended for intraindividual comparisons of nodule detection sensitivity, rather than the Chi-square test.

Response: Thank you for the keen comment. As your advice, we reconducted McNemar test to compare the nodule detection sensitivities of each radiologist and VCAR between DLIR-LDCT and DLIR-ULDCT and to compare the nodule detection sensitivity of VCAR and LuCAS. Each p value number itself has changed, but the overall statistical significance has not changed. The changes were as follows:

Radiologists’ nodule detection sensitivities between DLIR-LDCT and DLIR-ULDCT, P=.57 → P=1

VCAR’s nodule detection sensitivity between DLIR-ULDCT and ASIR-V50-ULDCT, P=.99 → P=1

VCAR’s nodule detection sensitivity between DLIR-LDCT and DLIR-ULDCT, P=.64 → P=.5

Nodule detection sensitivity between VCAR and LuCAS, P=.113 → P=.18

Subsolid nodule detection sensitivity between VCAR and LuCAS, P=.002 → P=.03

Response: We appreciate the comments. As you suggested, we added previous studies of DLIR in discussion section as follows:

There are previous studies of a vendor-specific and vendor-agnostic DLIR to create high quality images of ULDCT [26, 27]. In a study by Goto et al., ULDCT with vendor-specific DLIR (AiCE, Canon Medical Systems) showed better nodule CNR and comparable subjective image quality assessment (noise in air, noise in soft-tissue, streak artifact, texture fineness, and overall quality) compared to LDCT with iterative reconstruction in the phantom study [26]. In a subjective evaluation, ULDCT with DLIR of 14 nonobese patients achieved a significantly higher preference by radiologists compared to ULDCT with iterative reconstruction [26]. Hata et al. proved a vendor-agnostic DLIR for noise reduction of ULDCT showed better subjective image quality compared to the ULDCT with iterative reconstruction [27]. Also, nodule detection rate of radiologists improved in ULDCT with DLIR compared to ULDCT with iterative reconstruction [27]. These previous studies and ours both align with the same result that ULDCT with DLIR increases objective and subjective image quality compared to iterative reconstruction algorithms and provides image quality comparable to that of LDCT.

8. The study findings may be applicable for pediatric CT, where radiation exposure is of concern in comparison to adults who require lung cancer screening. Discussing the applicability to low-dose pediatric CT by citing the relevant articles (such as doi: 10.1148/rg.2021210105. and doi: 10.1186/s12880-021-00677-2.) would further emphasize the value of the authors' findings.

Response: Dose reduction in pediatric CT using DLIR and ULDCT is a very novel and influential idea and thank you for suggesting a perspective that we hadn’t thought of before. We added this to discussion section as follows:

There have been concerns regarding the increased radiation exposure from medical imaging of pediatric populations [28]. Efforts to reduce the radiation dose in pediatric CT imaging continued steadily and dose reduction strategies in CT imaging is expected to dramatically reduce the adverse effects of radiation exposure such as radiation-induced cancers [28, 29]. One of the proven dose reduction strategies is DLIR which can minimize image noise while preserving the adequate resolution of the CT scans [29, 30]. In a study by Yoon et al., in pediatric non-contrast and contrast-enhanced chest CT (CTDIvol, 1.3 ± 0.5 mGy; DLP 49.0 ± 26.3 mGy × cm; effective dose, 2.2 ± 3.2 mSv), DLIR showed significant noise reduction with increased SNR and CNR and showed better subjective overall image quality and noise compared to iterative reconstruction [30]. This result is comparable to our results and application of DLIR is expected to substantially reduce the radiation dose while maintaining the appropriate image quality for diagnosis in pediatric chest CT scans.

Attachment

Submitted filename: Response to reveiwer_231219.docx

pone.0297390.s008.docx^{(23.6KB, docx)}

PLoS One. doi: 10.1371/journal.pone.0297390.r005

Decision Letter 2

Gayle E Woloschak

4 Jan 2024

Prospective Evaluation of Deep Learning Image Reconstruction for Lung-RADS and Automatic Nodule Volumetry on Ultralow-Dose Chest CT

PONE-D-23-05687R2

Dear Dr. Yoon:

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Gayle E. Woloschak, PhD

Section Editor

PLOS ONE

Additional Editor Comments (optional):

Thank you for addressing the concerns raised by the reviewers.

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

Reviewer #3: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

Reviewer #3: (No Response)

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #3: (No Response)

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

Reviewer #3: (No Response)

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

Reviewer #3: (No Response)

**********

6. Review Comments to the Author

Reviewer #3: (No Response)

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #3: No

**********

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

S1 Fig. Bland-Altman plots of the X-axis diameter difference of lung nodules.

(TIF)

pone.0297390.s001.tif^{(788.2KB, tif)}

S2 Fig. Distribution of measurement differences of the X-axis diameter of lung nodules.

(TIF)

pone.0297390.s002.tif^{(2.1MB, tif)}

S3 Fig. Bland-Altman plots of the volume difference of lung nodules in ASIR-V₅₀-LDCT between lung VCAR and LuCAS.

ASIR-V₅₀ = adaptive statistical iterative reconstruction-V 50%, LDCT = low-dose chest computed tomography, VCAR = volume computerized assisted reporting.

(TIF)

pone.0297390.s003.tif^{(1.5MB, tif)}

S1 Table. Scoring system of subjective image quality.

(DOCX)

pone.0297390.s004.docx^{(31.4KB, docx)}

S2 Table. Results of the observer performance study for the detection of nodules with Lung-RADS 1.1 Categories 3 and 4.

(DOCX)

pone.0297390.s005.docx^{(34.2KB, docx)}

S3 Table. Nodule detection sensitivity of Lung VCAR and LuCAS applied on low-dose chest ct reconstructed with ASIR-V₅₀.

(DOCX)

pone.0297390.s006.docx^{(32.4KB, docx)}

Attachment

Submitted filename: Response to Reveiwers.docx

pone.0297390.s007.docx^{(433.8KB, docx)}

Attachment

Submitted filename: Response to reveiwer_231219.docx

pone.0297390.s008.docx^{(23.6KB, docx)}

Data Availability Statement

[pone.0297390.ref001] 1.National Lung Screening Trial Research T, Aberle, Adams AM, Berg CD, Black WC, Clapp, et al. Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med. 2011;365(5):395–409. Epub 2011/07/01. doi: 10.1056/NEJMoa1102873 ; PubMed Central PMCID: PMC4356534. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref002] 2.de Koning HJ, van der Aalst CM, de Jong PA, Scholten ET, Nackaerts K, Heuvelmans MA, et al. Reduced Lung-Cancer Mortality with Volume CT Screening in a Randomized Trial. N Engl J Med. 2020;382(6):503–13. Epub 2020/01/30. doi: 10.1056/NEJMoa1911793 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref003] 3.Kim HY. National lung cancer screening in Korea: introduction and imaging quality control. Journal of the Korean Society of Radiology. 2019;80(5):826–36. doi: 10.3348/jksr.2019.80.5.826 [DOI] [Google Scholar]

[pone.0297390.ref004] 4.Pinsky PF. Lung cancer screening with low-dose CT: a world-wide view. Transl Lung Cancer Res. 2018;7(3):234–42. Epub 2018/07/28. doi: 10.21037/tlcr.2018.05.12 ; PubMed Central PMCID: PMC6037972. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref005] 5.Rusinek H, Naidich DP, McGuinness G, Leitman BS, McCauley DI, Krinsky GA, et al. Pulmonary nodule detection: low-dose versus conventional CT. Radiology. 1998;209(1):243–9. Epub 1998/10/14. doi: 10.1148/radiology.209.1.9769838 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref006] 6.Sakane H, Ishida M, Shi L, Fukumoto W, Sakai C, Miyata Y, et al. Biological Effects of Low-Dose Chest CT on Chromosomal DNA. Radiology. 2020;295(2):439–45. doi: 10.1148/radiol.2020190389 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref007] 7.Guo R, Zhang Y, Ma Z, Deng C, Fu F, Hu H, et al. Overuse of follow-up chest computed tomography in patients with incidentally identified nodules suspicious for lung cancer. J Cancer Res Clin Oncol. 2022;148(5):1147–52. Epub 20210708. doi: 10.1007/s00432-021-03692-6 . [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref008] 8.Lee SW, Kim Y, Shim SS, Lee JK, Lee SJ, Ryu YJ, et al. Image quality assessment of ultra low-dose chest CT using sinogram-affirmed iterative reconstruction. Eur Radiol. 2014;24(4):817–26. Epub 2014/01/21. doi: 10.1007/s00330-013-3090-9 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref009] 9.Paks M, Leong P, Einsiedel P, Irving LB, Steinfort DP, Pascoe DM. Ultralow dose CT for follow-up of solid pulmonary nodules: A pilot single-center study using Bland-Altman analysis. Medicine (Baltimore). 2018;97(34):e12019. Epub 2018/08/26. doi: 10.1097/MD.0000000000012019 ; PubMed Central PMCID: PMC6112944. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref010] 10.Beregi JP, Greffier J. Low and ultra-low dose radiation in CT: Opportunities and limitations. Diagn Interv Imaging. 2019;100(2):63–4. Epub 2019/02/09. doi: 10.1016/j.diii.2019.01.007 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref011] 11.Lee SC, Yoon SH, Goo JM, Yim JJ, Kim CK. Submillisievert Computed Tomography of the Chest in Contact Investigation for Drug-Resistant Tuberculosis. J Korean Med Sci. 2017;32(11):1779–83. doi: 10.3346/jkms.2017.32.11.1779 ; PubMed Central PMCID: PMC5639057. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref012] 12.Hsieh J, Liu E, Nett B, Tang J, Thibault J-B, Sahney S. A new era of image reconstruction: TrueFidelity™. Technical white paper on deep learning image reconstruction GE Healthcare. 2019. [Google Scholar]

[pone.0297390.ref013] 13.Jiang B, Li N, Shi X, Zhang S, Li J, de Bock GH, et al. Deep Learning Reconstruction Shows Better Lung Nodule Detection for Ultra-Low-Dose Chest CT. Radiology. 2022;303(1):202–12. doi: 10.1148/radiol.210551 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref014] 14.Wang H, Li LL, Shang J, Song J, Liu B. Application of deep learning image reconstruction in low-dose chest CT scan. Br J Radiol. 2022;95(1133):20210380. doi: 10.1259/bjr.20210380 . [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref015] 15.Kim JH, Yoon HJ, Lee E, Kim I, Cha YK, Bak SH. Validation of Deep-Learning Image Reconstruction for Low-Dose Chest Computed Tomography Scan: Emphasis on Image Quality and Noise. Korean J Radiol. 2021;22(1):131–8. doi: 10.3348/kjr.2020.0116 ; PubMed Central PMCID: PMC7772377. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref016] 16.Lee HN, Kim JI, Shin SY. Measurement accuracy of lung nodule volumetry in a phantom study: Effect of axial-volume scan and iterative reconstruction algorithm. Medicine (Baltimore). 2020;99(23):e20543. Epub 2020/06/06. doi: 10.1097/MD.0000000000020543 ; PubMed Central PMCID: PMC7306330. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref017] 17.Rotzinger DC, Racine D, Beigelman-Aubry C, Alfudhili KM, Keller N, Monnin P, et al. Task-Based Model Observer Assessment of A Partial Model-Based Iterative Reconstruction Algorithm in Thoracic Oncologic Multidetector CT. Sci Rep. 2018;8(1):17734. Epub 2018/12/12. doi: 10.1038/s41598-018-36045-4 ; PubMed Central PMCID: PMC6286352. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref018] 18.Cicchetti DV. Guidelines, criteria, and rules of thumb for evaluating normed and standardized assessment instruments in psychology. Psychological Assessment. 1994;6(4):284–90. doi: 10.1037/1040-3590.6.4.284 [DOI] [Google Scholar]

[pone.0297390.ref019] 19.Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977;33(1):159–74. Epub 1977/03/01. [PubMed] [Google Scholar]

[pone.0297390.ref020] 20.Viera AJ, Garrett JM. Understanding interobserver agreement: the kappa statistic. Fam med. 2005;37(5):360–3. [PubMed] [Google Scholar]

[pone.0297390.ref021] 21.Deak PD, Smal Y, Kalender WA. Multisection CT protocols: sex- and age-specific conversion factors used to determine effective dose from dose-length product. Radiology. 2010;257(1):158–66. Epub 2010/09/21. doi: 10.1148/radiol.10100047 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref022] 22.Liang M, Yip R, Tang W, Xu D, Reeves A, Henschke CI, et al. Variation in Screening CT-Detected Nodule Volumetry as a Function of Size. AJR Am J Roentgenol. 2017;209(2):304–8. Epub 2017/06/02. doi: 10.2214/AJR.16.17159 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref023] 23.Peters AA, Christe A, von Stackelberg O, Pohl M, Kauczor HU, Heußel CP, et al. "Will I change nodule management recommendations if I change my CAD system?"-impact of volumetric deviation between different CAD systems on lesion management. Eur Radiol. 2023;33(8):5568–77. Epub 2023/03/10. doi: 10.1007/s00330-023-09525-z ; PubMed Central PMCID: PMC10326095. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref024] 24.van Riel SJ, Jacobs C, Scholten ET, Wittenberg R, Winkler Wille MM, de Hoop B, et al. Observer variability for Lung-RADS categorisation of lung cancer screening CTs: impact on patient management. Eur Radiol. 2019;29(2):924–31. Epub 2018/08/02. doi: 10.1007/s00330-018-5599-4 ; PubMed Central PMCID: PMC6302878. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref025] 25.Goto M, Nagayama Y, Sakabe D, Emoto T, Kidoh M, Oda S, et al. Lung-Optimized Deep-Learning-Based Reconstruction for Ultralow-Dose CT. Acad Radiol. 2023;30(3):431–40. Epub 2022/06/24. doi: 10.1016/j.acra.2022.04.025 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref026] 26.Hata A, Yanagawa M, Yoshida Y, Miyata T, Tsubamoto M, Honda O, et al. Combination of Deep Learning-Based Denoising and Iterative Reconstruction for Ultra-Low-Dose CT of the Chest: Image Quality and Lung-RADS Evaluation. AJR Am J Roentgenol. 2020;215(6):1321–8. Epub 2020/10/15. doi: 10.2214/AJR.19.22680 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref027] 27.Miglioretti DL, Johnson E, Williams A, Greenlee RT, Weinmann S, Solberg LI, et al. The use of computed tomography in pediatrics and the associated radiation exposure and estimated cancer risk. JAMA Pediatr. 2013;167(8):700–7. Epub 2013/06/12. doi: 10.1001/jamapediatrics.2013.311 ; PubMed Central PMCID: PMC3936795. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref028] 28.Nagayama Y, Sakabe D, Goto M, Emoto T, Oda S, Nakaura T, et al. Deep Learning-based Reconstruction for Lower-Dose Pediatric CT: Technical Principles, Image Characteristics, and Clinical Implementations. Radiographics. 2021;41(7):1936–53. Epub 2021/10/02. doi: 10.1148/rg.2021210105 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref029] 29.Yoon H, Kim J, Lim HJ, Lee MJ. Image quality assessment of pediatric chest and abdomen CT by deep learning reconstruction. BMC Med Imaging. 2021;21(1):146. Epub 2021/10/12. doi: 10.1186/s12880-021-00677-2 ; PubMed Central PMCID: PMC8503996. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref030] 30.Kim H, Kim HY, Goo JM, Kim Y. Lung Cancer CT Screening and Lung-RADS in a Tuberculosis-endemic Country: The Korean Lung Cancer Screening Project (K-LUCAS). Radiology. 2020;296(1):181–8. doi: 10.1148/radiol.2020192283 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref031] 31.Zhao J, Barta JA, McIntire R, Shusted C, Zeigler-Johnson C, Juon HS. Racial difference in BMI and lung cancer diagnosis: analysis of the National Lung Screening Trial. BMC Cancer. 2022;22(1):797. doi: 10.1186/s12885-022-09888-4 ; PubMed Central PMCID: PMC9297592. [DOI] [PMC free article] [PubMed] [Google Scholar]

[pone.0297390.ref032] 32.Svahn TM, Sjoberg T, Ast JC. Dose estimation of ultra-low-dose chest CT to different sized adult patients. Eur Radiol. 2019;29(8):4315–23. doi: 10.1007/s00330-018-5849-5 . [DOI] [PubMed] [Google Scholar]

[pone.0297390.ref033] 33.Guo X, Jia D, He L, Jia X, Zhang D, Dou Y, et al. Evaluation of ultralow-dose computed tomography on detection of pulmonary nodules in overweight or obese adult patients. J Appl Clin Med Phys. 2022;23(4):e13589. doi: 10.1002/acm2.13589 ; PubMed Central PMCID: PMC8992951. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Prospective evaluation of deep learning image reconstruction for Lung-RADS and automatic nodule volumetry on ultralow-dose chest CT

Seung-Jin Yoo

Young Sik Park

Hyewon Choi

Da Som Kim

Jin Mo Goo

Soon Ho Yoon

Roles

Abstract

Purpose

Methods

Results

Conclusion

Introduction

Materials and methods

Study population

CT acquisition

Objective and subjective image quality assessment

Nodule detection and Lung-RADS evaluation by readers

Volumetric nodule evaluation

Fig 1. Flowchart of participants and lung nodules included for the nodule detectability and Lung-RADS category evaluation.

Fig 2. Detection and volumetric measurement of a lung nodule by VCAR.

Statistical analysis

Results

Demographics

Table 1. Characteristics of study participants.

Objective and subjective image quality assessment

Table 2. Results of objective and subjective image quality assessment.

Fig 3. Representative images of a subsolid nodule in Low-dose and Ultralow-dose CT scans with iterative and deep-learning reconstruction algorithms.

Fig 4. Representative images of mediastinal structures in Low-dose and Ultralow-dose CT scans with iterative and deep-learning reconstruction algorithm.

Nodule detectability and Lung-RADS categorization by readers

Table 3. Nodule detection sensitivity of radiologists.

Nodule detectability and Lung-RADS categorization by VCAR

Table 4. Nodule detection sensitivity of Lung VCAR.

Nodule volume and diameter difference by VCAR

Fig 5. Bland-Altman plots of nodule volume measurement differences.

Fig 6. Distribution of measurement differences of the nodule volume.

Discussion

Supporting information

Data Availability

Funding Statement

References

Decision Letter 0

Gayle E Woloschak

Roles

Author response to Decision Letter 0

Decision Letter 1

Gayle E Woloschak

Roles

Author response to Decision Letter 1

Decision Letter 2

Gayle E Woloschak

Roles

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases