Deep learning–based super-resolution ultrashort echo time MRI for pulmonary nodule detection and longitudinal assessment in emphysema

Abstract

Purpose

This study aimed to evaluate the performance of deep learning–based super-resolution ultrashort echo time magnetic resonance imaging (SR-UTE MRI) for the detection of pulmonary nodules and the reliability of longitudinal nodule size assessment compared with conventional ultrashort echo time (UTE) in patients with emphysema.

Materials and methods

This prospective study included 127 patients with emphysema and pulmonary nodules who underwent chest low-dose computed tomography (LDCT) and UTE MRI. SR-UTE images were reconstructed using a deep learning-based algorithm. Pulmonary nodule detection performance, size measurement accuracy, and longitudinal reproducibility were assessed using LDCT as the reference standard. Patients were stratified according to CT-derived emphysema severity quantified by low-attenuation area percentage (LAA%). Objective image quality metrics, including signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and Laplacian variance, were also evaluated.

Results

SR-UTE demonstrated superior image quality compared with conventional UTE, with higher SNR and CNR in patients with severe emphysema (both p < 0.05). For nodules < 6 mm, SR-UTE achieved a higher detection rate than UTE (81.8% vs. 54.5%, p < 0.05), with greater improvement for ground-glass nodules in severe emphysema (66.7% vs. 33.3%). Both UTE and SR-UTE achieved 100% sensitivity for nodules > 8 mm. SR-UTE showed better agreement with CT for nodule size measurement, with higher intraclass correlation coefficients across emphysema subgroups (0.89–0.93 vs. 0.84–0.88 for UTE) and reduced measurement bias. In longitudinal analysis, SR-UTE more accurately reflected CT-derived nodule size changes and demonstrated higher reproducibility than UTE (ICC 0.90 vs. 0.82).

Conclusion

SR-UTE improved the detection of small and subsolid pulmonary nodules and provided more reliable longitudinal size assessment than UTE in patients with emphysema. These findings support the potential role of SR-UTE MRI in MRI-based surveillance of pulmonary nodules.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12890-026-04468-9.

Introduction

Lung cancer remains one of the leading causes of cancer-related mortality worldwide [1]. Early detection, accurate classification, and appropriate management of pulmonary nodules based on their size and type are critical components in the evaluation of patients with suspected lung cancer. Computed tomography (CT) is currently regarded as the reference standard for pulmonary nodule detection owing to its high spatial resolution and rapid acquisition time [2]. However, repeated CT examinations inherently involve cumulative exposure to ionizing radiation, which increases the long-term risk of radiation-induced cancer [3], particularly in patients requiring prolonged imaging surveillance.

Magnetic resonance imaging (MRI) provides a radiation-free alternative for thoracic imaging. Nevertheless, its application in lung imaging has been limited by several technical challenges, including the intrinsically low proton density of lung parenchyma, rapid signal decay associated with extremely short T2* relaxation times, and substantial respiratory and cardiac motion [4]. Ultrashort echo time (UTE) MRI sequences, enabled by non-Cartesian radial sampling, permit signal acquisition from tissues with very short T2* relaxation times by minimizing susceptibility-related signal loss at air–tissue interfaces [5]. These techniques help preserve the signal-to-noise ratio (SNR) of lung parenchyma by minimizing T2* decay, thereby avoiding the signal loss typically seen with conventional MRI sequences.

Clinical studies have demonstrated that UTE MRI enables reliable pulmonary nodule detection with high concordance with low-dose CT (LDCT), including comparable Lung-RADS classification and high sensitivity for nodules larger than 5 mm [5, 6].Therefore, the Fleischner Society and other international organizations have recognized UTE MRI as a promising technique for thoracic MRI in clinical practice [6–8].

Despite these advances, the clinical application of UTE MRI for pulmonary nodule evaluation remains limited due to insufficient spatial resolution and reduced nodule conspicuity, particularly for small lesions. UTE images may demonstrate blurred nodule margins and limited contrast relative to surrounding lung parenchyma. These challenges may be further exacerbated in the presence of underlying pulmonary parenchymal disease, potentially compromising diagnostic reliability.

Emphysema is a well-established independent risk factor for lung cancer [9, 10]. Pathologically, it is characterized by destruction of alveolar walls, enlargement of distal airspaces, and reduced pulmonary tissue density, which collectively result in diminished parenchymal signal on MRI. Data from lung cancer screening studies indicate that approximately 48–58% of patients with emphysema have pulmonary nodules [11]. This high prevalence increases the risk of lung cancer-related mortality and necessitates repeated imaging for longitudinal nodule surveillance. In this context, the inherently low SNR and limited contrast of lung MRI are further exacerbated by emphysematous changes, making accurate visualization and characterization of pulmonary nodules particularly challenging.

Deep learning-based super-resolution (SR) techniques have increasingly been investigated to address intrinsic spatial resolution limitations in medical imaging that arise from hardware constraints, acquisition time restrictions, and radiation dose considerations [12, 13]. In contrast to conventional interpolation-based upsampling methods, which often produce spatial blurring and loss of high-frequency information, SR approaches reconstruct high-resolution images by learning complex mappings between low- and high-resolution image domains. In particular, three-dimensional SR reconstruction based on generative adversarial networks (GANs) enables recovery of fine-scale anatomical structures while maintaining global image consistency [14, 15].

By incorporating multi-component loss functions that combine pixel-wise accuracy, edge preservation, and perceptual similarity, SR approaches have demonstrated superior structural fidelity and enhanced spatial detail across multiple imaging modalities. SR reconstruction has demonstrated potential value in several MRI applications, including neuroimaging [16], rectal imaging [17], and musculoskeletal imaging [18]. However, evidence remains limited regarding whether super-resolution reconstructed UTE (SR-UTE) images provide measurable advantages over conventional UTE images for pulmonary nodule visualization, and whether these advantages are influenced by the severity of emphysema.

To our knowledge, no previous study has systematically investigated deep learning-based super-resolution reconstruction for UTE lung MRI in pulmonary nodule assessment while also evaluating, in the same framework, the impact of emphysema severity and the longitudinal reproducibility of nodule measurements. Therefore, the purpose of this study was to evaluate whether SR-UTE MRI improves pulmonary nodule visualization compared with conventional UTE MRI and to assess the influence of emphysema severity, quantified using CT-derived percentage of low-attenuation areas (LAA%), on this potential diagnostic improvement. It is hypothesized that SR-UTE MRI provides improved nodule boundary delineation, greater conspicuity of internal architectural features, and higher diagnostic confidence, with these benefits maintained across varying degrees of emphysema severity. A secondary objective was to determine whether SR-UTE MRI improves the longitudinal reproducibility of pulmonary nodule evaluation in patients with emphysema during follow-up imaging.

Materials and methods

This study was approved by the Ethics Committee of the Fourth Hospital of Hebei Medical University (No. 2020KY082) and was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants.

Study participants

Between January 2020 and December 2021, a total of 127 consecutive patients with radiologically confirmed emphysema who were participating in a lung cancer screening program were prospectively enrolled. Enrollment followed the recommendations of the Fleischner Society recommendations for patients at increased risk of lung cancer requiring imaging surveillance [19]. All patients underwent chest CT and were additionally asked to undergo UTE-MRI within 24 h.

The inclusion criteria were as follows: (1) presence of indeterminate pulmonary nodules measuring < 3 cm in maximum diameter detected on LDCT; (2) radiologically confirmed emphysema based on quantitative analysis; (3) availability of both chest CT and UTE-MRI examinations with diagnostic image quality adequate for analysis; and (4) age greater than 40 years.

The exclusion criteria were as follows: (1) contraindications to MRI, including pacemakers or ferromagnetic implants; (2) severe claustrophobia precluding MRI acquisition; (3) pregnancy or breastfeeding; and (4) completely calcified pulmonary nodules, which are not suitable for MRI-based assessment. The patient selection process and exclusion criteria are illustrated in Fig. 1.

Fig. 1.

Flowchart illustrating the participant selection process

Pulmonary nodule assessment

Two thoracic radiologists independently reviewed all CT and MRI examinations to identify pulmonary nodules with a maximum diameter greater than 3 mm, in accordance with previously published studies and established clinical guidelines [20, 21]. The reviewers were blinded to all clinical data.

CT images were evaluated using a standardized lung window setting with a window width of 1,500 HU and a window level of − 600 HU. MRI images were evaluated using predefined windowing parameters optimized for pulmonary nodule visualization.

For each detected nodule, the maximum axial (long axis) diameter and the perpendicular short-axis diameter were measured. The mean diameter, defined as the average of the long- and short-axis measurements, was used for nodule characterization and for longitudinal analysis.

Nodule attenuation was classified as solid, part-solid nodule (PSN), or pure ground-glass nodule (GGN). Additional morphological characteristics, including calcification, fat content, spiculation, lobulation, cavitation, and necrosis, were systematically recorded [19, 20, 22]. Nodule location was categorized as peripheral when located within 5 mm of the visceral pleura and non-peripheral when located more than 5 mm from the pleural surface. Proximity to pulmonary fissures was classified as intrafissural when the nodule was in direct contact with a fissure and perifissural when located within 5 mm of a fissure.

According to the Fleischner Society guidelines [19, 23], nodules were categorized for subgroup analysis based on size as follows: Solid nodules: < 6 mm, 6–8 mm, and > 8 mm; GGNs and PSNs: ≤ 6 mm and > 6 mm. Discrepancies between readers were resolved by consensus, with adjudication by a third radiologist when necessary.

CT imaging protocol

All CT examinations were performed with patients in the supine position using a 128–row dual-source multidetector CT system (Somatom Definition Flash, Siemens Healthineers, Forchheim, Germany). Scanning parameters were as follows: tube voltage, 120 kVp; effective tube current, 50 mAs; pitch, 1.2; field of view, 300 mm; slice thickness, 1 mm; and reconstruction kernel, B70f. Patients were instructed to raise their arms and perform a single breath-hold during image acquisition. The scan range extended from the thoracic inlet to the diaphragm to ensure complete lung coverage. The acquisition time was 3.98 s, with an effective radiation dose of approximately 1.2–1.5 mSv.

CT-derived emphysema severity as baseline stratification

CT-derived emphysema severity was used as a baseline stratification variable in this study. Quantitative assessment of emphysema was performed using the open-source software 3D Slicer (version 5.10.0; www.slicer.org). Whole-lung segmentation was initially obtained using a semi-automatic threshold-based algorithm, followed by manual refinement when necessary to exclude central airways and extrapulmonary structures. Low-attenuation areas (LAA) were defined as lung voxels with attenuation values below − 950 HU on inspiratory CT images. The emphysema index was calculated as the percentage of LAA relative to the total lung volume (LAA%). All segmentations and measurements were reviewed by an experienced thoracic radiologist to ensure accuracy and consistency. LAA% was used solely to classify patients into different emphysema severity groups for subsequent subgroup analyses and was not considered a study outcome. Based on previously published thresholds [24–27], patients were categorized as having mild emphysema (LAA% < 10%), moderate emphysema (LAA% 10–25%), or severe emphysema (LAA% > 25%).

MR image acquisition

MRI examinations were performed using a 3.0-T system (Magnetom Skyra, Siemens Healthineers, Erlangen, Germany) equipped with a multi-channel phased-array body coil. Patients were positioned supine with their arms raised and instructed to breathe freely during image acquisition. A free-breathing PETRA sequence was acquired with the following parameters: repetition time, 3 ms; echo time, 0.07 ms; matrix size, 256 × 256; field of view, 400 × 400 mm; slice thickness, 1.6 mm; and axial acquisition plane. Respiratory gating was applied to reduce respiratory and cardiac motion artifacts. The total acquisition time was approximately 3 min.

Deep learning-based super-resolution reconstruction

SR-UTE images were reconstructed from conventional UTE images using the Onekey AI medical imaging platform (OnekeyAI, China), which incorporates a three-dimensional deep learning–based super-resolution reconstruction framework. A generative adversarial network (GAN) architecture consisting of generator and discriminator networks was employed to learn the mapping between low-resolution and high-resolution image volumes using paired training data. The model was trained on millions of medical images from multiple imaging modalities, including CT, MRI, and ultrasound. Low-resolution and high-resolution image pairs were generated through a downsampling strategy and used for supervised training.

The model was optimized using a composite loss function that incorporated L1 (mean absolute error [MAE]) loss, gradient loss, and perceptual loss to improve spatial detail preservation, structural fidelity, and image sharpness. This reconstruction process increased spatial resolution in the slice-encoding direction by a factor of four while preserving the original image dimensions, generating the SR-UTE image datasets used for subsequent analysis.

Image quality assessment

Two thoracic radiologists (Reader 1 with 9 years of experience and Reader 2 with 15 years of experience) independently reviewed all CT and MRI examinations. Image interpretation was performed in randomized order, with a 4-week interval between evaluations of different imaging modalities to minimize recall bias. Both readers were blinded to all clinical information and to the results of the other imaging modality.

Subjective assessment

For subjective assessment, two radiologists independently evaluated pulmonary nodules detected on both conventional UTE and SR-UTE images. Image quality was graded using a 5-point Likert scale across four domains: nodule conspicuity, visualization of anatomical structures (including peripheral vasculature, trachea, and bronchi), overall image quality, and diagnostic acceptability. The scoring criteria were defined as follows: 5, excellent; 4, good; 3, fair; 2, poor; and 1, unacceptable [28]. In cases of disagreement between readers, a final score was determined through consensus review.

Objective assessment

For quantitative assessment, the SNR and contrast-to-noise ratio (CNR) were calculated using selected regions of interest (ROIs) to compare image quality between UTE and SR-UTE images. ROIs used to measure signal intensity (SI) were placed within the lung parenchyma while avoiding major pulmonary vessels and paraspinal muscles. The following parameters were determined for both UTE and SR-UTE images: background noise SI, lung parenchymal SI, muscle SI, and the standard deviation (SD) of background noise. SNR and CNR were calculated using the following equations:

Quantitative assessment of image sharpness

To objectively evaluate the effect of super-resolution reconstruction on image sharpness, Laplacian Variance (LV) was calculated for both UTE and SR-UTE images. LV was measured in corresponding pulmonary nodule regions, with higher values indicating improved edge definition and spatial resolution. LV has been widely used as an objective measure of image sharpness in medical imaging studies [29].

Follow-up subgroup

Follow-up CT and MRI examinations were independently evaluated by two radiologists who were blinded to imaging time points, imaging modality, and all clinical information. Image interpretation followed the same predefined evaluation criteria, measurement protocols, and workstation settings used for the baseline assessment to ensure methodological consistency during longitudinal analyses. Changes in pulmonary nodule size were assessed using the mean diameter derived from the long-axis and short-axis measurements obtained on LDCT at baseline and follow-up examinations. Based on the magnitude of size change, nodules were categorized as follows: Progressive: increase in mean diameter > 2 mm; regressive: decrease in mean diameter > 2 mm; resolved: complete disappearance of the nodule; or stable: change in diameter within ± 2 mm) [19, 23].

Statistical analysis

All statistical analyses were performed using SPSS software (version 22.0; IBM Corp., Armonk, NY, USA) and MedCalc (version 23.0; MedCalc Software). Continuous variables are presented as mean ± standard deviation (SD) or median with interquartile range (IQR), as appropriate, while categorical variables are reported as counts and percentages. Comparisons of quantitative image quality metrics, including SNR, CNR, LV between UTE and SR-UTE images were performed using the paired t-test for normally distributed data or the Wilcoxon signed-rank test for non-normally distributed data.

Differences in subjective image quality scores were evaluated using the Wilcoxon signed-rank test. Pulmonary nodule detection rates and morphologic classifications across imaging sequences and emphysema severity groups were compared using the McNemar test or the chi-squared test, as appropriate. Interobserver agreement for quantitative measurements was assessed using the intraclass correlation coefficient (ICC), whereas agreement for categorical assessments was evaluated using weighted kappa (κ) statistics, both reported with 95% confidence intervals (Cis).

For nodules with follow-up imaging, the reproducibility of mean diameter measurements and the agreement in classification of nodule progression were evaluated using Bland–Altman analysis and ICC. Correlations between emphysema severity (LAA%) and improvements in nodule conspicuity on SR-UTE images were analyzed using Spearman’s rank correlation coefficient. To further evaluate pulmonary nodule detection performance, receiver operating characteristic (ROC) analyses were performed using CT as the reference standard. The area under the ROC curve (AUC) was calculated for UTE and SR-UTE MRI in the overall cohort and within each emphysema severity subgroup. A two-sided p value < 0.05 was considered statistically significant.

Results

Patient and nodule characteristics

Between January 2020 and December 2021, a total of 127 patients were included in the analysis (mean age, 63.5 years ± 16.7 [SD]). The cohort comprised 62 men (49%) and 65 women (51%), of whom 45% (57/127) were current or former smokers. Based on emphysema severity, 49 patients (38%) were classified as having mild emphysema, 41 (32%) as having moderate emphysema, and 37 (30%) as having severe emphysema.

In total, 169 pulmonary nodules were evaluated, with a mean diameter of 10.2 mm ± 6.4 (range, 3.3–25.6 mm). According to morphologic classification, 57 nodules (34%) were ground-glass nodules (GGNs; mean diameter, 7.4 mm ± 2.1), 49 (29%) were part-solid nodules (PSNs; mean diameter, 11.5 mm ± 3.5), and 63 (37%) were solid nodules (SNs; mean diameter, 8.9 mm ± 2.8).

Regarding anatomical distribution, most nodules were located in the right upper lobe (66/169, 39%), followed by the left upper lobe (37/169, 22%), right lower lobe (25/169, 15%), left lower lobe (24/169, 14%), and right middle lobe (17/169, 10%). When categorized by size, 42 nodules (25%) measured < 6 mm, 51 (30%) measured 6–8 mm, and 76 (45%) measured > 8 mm. Detailed characteristics are summarized in Table 1.

Table 1.

Participants and nodules characteristics

Characteristic	Value
Participant Characteristics
No. of participants	127
Age(y)*	63.5 ± 16.7
Gender
Male	62(49)
Female	65(51)
Smoking
Smoker	57(45)
Non-smoker	70(55)
Emphysema category†
Mild Emphysema	49(38)
Moderate Emphysema	41(32)
Severe Emphysema	37(30)
Nodule Characteristics
No. of nodules	169
Mean diameter (mm)*	10.2 ± 6.4 (range: 3.3–25.6)
Nodule type
Ground glass nodule (GGN)	57(34)
Mean diameter (mm)*	7.4 ± 2.1 (range: 4.3–19.1)
Part-solid nodule (PSN)	49(29)
Mean diameter (mm)*	11.5 ± 3.5 (range: 5.5–25.6)
Solid nodule (SN)	63(37)
Mean diameter(mm)*	8.9 ± 2.8 (range: 3.3–20.1)
Nodule Distribution
Right upper lobe	66(39)
Right middle lobe	17(10)
Right lower lobe	25(15)
Left upper lobe	37(22)
Left lower lobe	24(14)
Size category
< 6 mm	42(25)
6–8 mm	51(30)
> 8 mm	76(45)

Data are numbers of participants or nodules, with percentages in parentheses

* Data are mean±standard deviation

 †Emphysema severity was categorized based on LAA%

Image quality assessment

Subjective image quality

Subjective image quality scores were consistently higher for SR-UTE than for conventional UTE across all emphysema severity groups (Table 2). For nodule conspicuity, SR-UTE demonstrated significantly higher scores in patients with mild emphysema (3.4 ± 0.5 vs. 3.3 ± 0.4; p = 0.041), moderate emphysema (3.4 ± 0.5 vs. 3.2 ± 0.6; p = 0.018), and severe emphysema (3.2 ± 0.4 vs. 3.0 ± 0.6; p = 0.009).

Table 2.

Subjective and objective image quality assessment of UTE and SR-UTE MRI according to emphysema severity

Parameter	Sequence	Mild Emphysema (n = 49)	P value	Moderate Emphysema (n = 41)	P value	Severe Emphysema (n = 37)	P value
Subjective assessment
Nodule conspicuity	UTE	3.3 ± 0.4	.041	3.2 ± 0.6	.018	3.0 ± 0.6	.009
	SR-UTE	3.4 ± 0.5		3.4 ± 0.5		3.2 ± 0.4
Anatomical structures depiction	UTE	3.3 ± 0.4	.048	3.2 ± 0.6	.021	3.0 ± 0.5	.006
	SR-UTE	3.4 ± 0.4		3.4 ± 0.5		3.3 ± 0.4
Overall image quality	UTE	3.0 ± 0.4	.032	3.0 ± 0.3	.004	2.8 ± 0.6	.011
	SR-UTE	3.2 ± 0.5		3.3 ± 0.4		3.1 ± 0.6
Diagnostic acceptability	UTE	3.3 ± 0.5	.019	3.2 ± 0.6	.007	3.0 ± 0.5	.010
	SR-UTE	3.5 ± 0.5		3.5 ± 0.4		3.3 ± 0.5
Objective assessment
SNR	UTE	12.6 ± 3.1	.112	11.9 ± 3.6	.003	11.3 ± 4.5	.041
	SR-UTE	13.3 ± 3.6		14.5 ± 3.3		13.1 ± 3.7
CNR	UTE	15.4 ± 2.9	.021	13.8 ± 3.5	.009	13.3 ± 3.9	.015
	SR-UTE	16.9 ± 3.9		15.7 ± 3.2		15.5 ± 3.3
LV	UTE	148.6 ± 34.2	< 0.001	137.9 ± 31.8	< 0.001	124.5 ± 29.7	< 0.001
	SR-UTE	196.2 ± 39.8		179.3 ± 37.5		158.1 ± 34.2

Data are mean scores±standard deviation 

SR-UTE super-resolution ultra-short echo time, UTE  ultra-short echo time, SNR signal-to-noise ratio, CNR contrast-to-noise ratio, LV Laplacian Variance

*P values were calculated using the Wilcoxon signed-rank test to compare UTE and SR-UTE images. Significant values are in bold

Similar improvements with SR-UTE were observed for visualization of anatomical structures and overall image quality across all emphysema severity categories (all p < 0.05). Diagnostic acceptability scores were also significantly higher for SR-UTE, with the largest difference observed in the moderate emphysema group (3.5 ± 0.4 vs. 3.2 ± 0.6; p = 0.007).

Objective image quality

Quantitative analysis further confirmed the superior image quality of SR-UTE compared with UTE. For SNR, SR-UTE demonstrated significantly higher values in patients with moderate emphysema (14.5 ± 3.3 vs. 11.9 ± 3.6; p = 0.003) and severe emphysema (13.1 ± 3.7 vs. 11.3 ± 4.5; p = 0.041), whereas no significant difference was observed in the mild emphysema group (p = 0.112).

Similarly, CNR values were consistently higher for SR-UTE across all emphysema severity groups, including mild (16.9 ± 3.9 vs. 15.4 ± 2.9; p = 0.021), moderate (15.7 ± 3.2 vs. 13.8 ± 3.5; p = 0.009), and severe emphysema (15.5 ± 3.3 vs. 13.3 ± 3.9; p = 0.015).

Quantitative assessment of image sharpness

Laplacian Variance analysis demonstrated significantly higher image sharpness in SR-UTE compared with conventional UTE images across all emphysema severity groups (Table 2). Overall, LV increased from 137.2 ± 33.5 in UTE images to 177.0 ± 38.8 in SR-UTE images, representing a 29.0% improvement. Similar trends were observed across mild, moderate, and severe emphysema subgroups, with LV increases ranging from 27.0% to 32.0% (all P < 0.001).

Nodule detection performance by emphysema severity

Overall pulmonary nodule detection sensitivity was significantly higher for SR-UTE than for conventional UTE across emphysema severity groups (Table 3). In patients with mild emphysema, SR-UTE detected 93.2% (55/59) of nodules compared with 83.1% (49/59) for UTE (p = 0.041). Similar advantages were observed in patients with moderate emphysema (86.9% [53/61] vs. 68.9% [42/61]; p = 0.034) and severe emphysema (75.5% [37/49] vs. 63.3% [31/49]; p = 0.028).

Table 3.

Nodule detection and characterization by emphysema severity

Emphysema Severity	Nodule Type	Fleischner size category	CT(Reference)	UTE Detected	SR-UTE Detected	P value
Mild emphysema	Ground glass nodules	< 6 mm	7	4 (57.1)	6 (85.7)	0.083
		≥ 6 mm	12	10 (83.3)	11 (91.7)	0.317
	Part solid nodules	< 6 mm	6	4 (66.7)	5 (83.3)	0.317
		≥ 6 mm	10	9 (90.0)	10 (100)	0.317
	Solid nodules	< 6 mm	6	5 (83.3)	6 (100)	0.317
		6–8 mm	7	6 (85.7)	7 (100)	0.317
		> 8 mm	11	11 (100)	11 (100)	—
	Total		59	49 (83.1)	55 (93.2)	0.041
Moderate emphysema	Ground glass nodules	< 6 mm	8	4 (50.0)	6 (75.0)	0.157
		≥ 6 mm	14	10 (71.4)	12 (85.7)	0.157
	Part solid nodules	< 6 mm	7	4 (57.1)	6 (85.7)	0.083
		≥ 6 mm	11	8 (72.7)	10 (90.9)	0.157
	Solid nodules	< 6 mm	7	4 (57.1)	6 (85.7)	0.083
		6–8 mm	6	4 (66.7)	5 (83.3)	0.317
		> 8 mm	8	8 (100)	8 (100)	—
	Total		61	42 (68.9)	53 (86.9)	0.034
Severe emphysema	Ground glass nodules	< 6 mm	6	2 (33.3)	4 (66.7)	0.157
		≥ 6 mm	10	6 (60.0)	8 (80.0)	0.157
	Part solid nodules	< 6 mm	5	2 (40.0)	4 (80.0)	0.157
		≥ 6 mm	10	6 (60.0)	8 (80.0)	0.157
	Solid nodules	< 6 mm	3	1 (33.3)	2 (66.7)	0.317
		6–8 mm	6	4 (66.7)	5 (83.3)	0.317
		> 8 mm	9	9 (100)	9 (100)	—
	Total		49	31 (63.3)	37 (75.5)	0.028

Data in parentheses are percentages

CT  computed tomography, GGN ground-glass nodule, PSN part-solid nodule, SN solid nodule, SR-UTE  super-resolution ultrashort echo time, UTE  ultrashort echo time

*P values were calculated using McNemar tests (or Fisher exact tests where appropriate) to compare detection sensitivities between UTE and SR-UTE. Significant values are in bold

Detection performance varied according to nodule type and size. For GGNs measuring < 6 mm, SR-UTE consistently demonstrated higher detection sensitivity than UTE across mild (85.7% vs. 57.1%), moderate (75.0% vs. 50.0%), and severe emphysema groups (66.7% vs. 33.3%), although these subgroup differences did not reach statistical significance. Similar trends favoring SR-UTE were observed for PSNs and SNs measuring < 6 mm. In contrast, both imaging sequences achieved 100% detection sensitivity for SNs larger than 8 mm, irrespective of emphysema severity. ROC analysis showed higher AUC values for SR-UTE than for UTE in the overall cohort (0.900 vs. 0.800, P < 0.001), as well as in the mild (0.920 vs. 0.840, P = 0.012), moderate (0.890 vs. 0.790, P = 0.008), and severe emphysema subgroups (0.830 vs. 0.710, P = 0.021) (Fig. 2).

Fig. 2.

Receiver operating characteristic (ROC) curves for pulmonary nodule detection by conventional UTE and super-resolution UTE (SR-UTE) MRI using CT as the reference standard. a overall cohort; b mild emphysema; c moderate emphysema; and d severe emphysema

Nodule size measurement and intermethod agreement

Intermethod agreement for pulmonary nodule size measurements, using CT as the reference standard, is summarized in Table 4. Across all emphysema severity groups, SR-UTE demonstrated higher agreement with CT than conventional UTE, with ICCs ranging from 0.89 to 0.93 for SR-UTE and from 0.84 to 0.88 for UTE. Among patients with mild emphysema, both sequences showed high agreement with CT; however, SR-UTE demonstrated a higher ICC than UTE (0.93 [95% CI: 0.86–0.97] vs. 0.88 [95% CI: 0.81–0.97]). As emphysema severity increased, the agreement between UTE at CT decreased, whereas SR-UTE maintained relatively stable performance. In the severe emphysema group, the ICC for SR-UTE remained 0.89 (95% CI: 0.83–0.94), whereas the ICC for UTE decreased to 0.84 (95% CI: 0.76–0.94).

Table 4.

Comparison of nodule diameter measurements and intermethod agreement between MRI and LDCT across emphysema severities

Emphysema Severity	Nodule Type	Size	CT Mean Diameter (mm)	UTE Mean Diameter (mm)	ICC (CT vs. UTE)[95%CI]	SR-UTE Mean Diameter (mm)	ICC (CT vs. SR-UTE)[95%CI]
Mild emphysema	Ground glass nodules	< 6 mm	5.2 ± 0.6	4.5 ± 0.5	0.88 [0.81–0.93]	4.9 ± 0.6	0.92 [0.87–0.96]
		≥ 6 mm	8.5 ± 1.2	7.6 ± 1.3	0.89 [0.83–0.94]	8.1 ± 1.2	0.93 [0.88–0.97]
	Part solid nodules	< 6 mm	5.8 ± 0.5	5.0 ± 0.5	0.87 [0.81–0.92]	5.4 ± 0.5	0.91 [0.86–0.95]
		≥ 6 mm	15.2 ± 4.8	14.0 ± 4.6	0.88 [0.83–0.92]	14.8 ± 4.7	0.92 [0.88–0.95]
	Solid nodules	< 6 mm	5.3 ± 0.6	4.5 ± 0.5	0.88 [0.82–0.93]	4.9 ± 0.6	0.93 [0.88–0.96]
		6–8 mm	7.0 ± 0.5	6.1 ± 0.6	0.89 [0.84–0.94]	6.5 ± 0.5	0.94 [0.90–0.97]
		> 8 mm	12.2 ± 3.1	11.2 ± 3.0	0.90 [0.85–0.94]	11.7 ± 3.0	0.94 [0.91–0.97]
	Total		8.0 ± 3.1	7.1 ± 2.9	0.88 [0.81–0.97]	7.8 ± 2.9	0.93 [0.86–0.97]
Moderate emphysema	Ground glass nodules	< 6 mm	5.1 ± 0.7	4.4 ± 0.6	0.86 [0.80–0.91]	4.9 ± 0.6	0.91 [0.86–0.95]
		≥ 6 mm	9.1 ± 1.3	8.2 ± 1.4	0.87 [0.82–0.92]	8.7 ± 1.3	0.92 [0.88–0.96]
	Part solid nodules	< 6 mm	5.9 ± 0.5	5.1 ± 0.6	0.85 [0.80–0.91]	5.5 ± 0.5	0.91 [0.86–0.95]
		≥ 6 mm	14.8 ± 4.9	13.6 ± 4.7	0.87 [0.82–0.92]	14.2 ± 4.8	0.92 [0.88–0.95]
	Solid nodules	< 6 mm	5.4 ± 0.5	4.6 ± 0.5	0.86 [0.81–0.91]	5.0 ± 0.5	0.91 [0.86–0.95]
		6–8 mm	7.1 ± 0.6	6.2 ± 0.6	0.87 [0.82–0.92]	6.6 ± 0.6	0.93 [0.88–0.96]
		> 8 mm	12.5 ± 3.2	11.4 ± 3.0	0.88 [0.83–0.93]	11.9 ± 3.1	0.93 [0.89–0.96]
	Total		8.6 ± 3.1	7.7 ± 3.1	0.87 [0.80–0.96]	8.1 ± 3.0	0.92 [0.86–0.96]
Severe emphysema	Ground glass nodules	< 6 mm	5.0 ± 0.6	4.2 ± 0.6	0.83 [0.77–0.90]	4.7 ± 0.6	0.89 [0.84–0.94]
		≥ 6 mm	8.3 ± 1.5	7.3 ± 1.6	0.85 [0.80–0.91]	7.9 ± 1.5	0.90 [0.85–0.94]
	Part solid nodules	< 6 mm	5.7 ± 0.6	4.8 ± 0.5	0.82 [0.76–0.89]	5.3 ± 0.5	0.89 [0.84–0.93]
		≥ 6 mm	14.5 ± 5.0	13.3 ± 4.9	0.84 [0.79–0.90]	14.0 ± 5.0	0.90 [0.85–0.94]
	Solid nodules	< 6 mm	5.3 ± 0.7	4.3 ± 0.6	0.83 [0.77–0.89]	4.8 ± 0.6	0.88 [0.83–0.93]
	Ground glass nodules	6–8 mm	7.2 ± 0.5	6.3 ± 0.6	0.84 [0.79–0.90]	6.7 ± 0.5	0.89 [0.84–0.93]
		> 8 mm	12.0 ± 3.3	11.0 ± 3.2	0.85 [0.80–0.91]	11.5 ± 3.2	0.90 [0.85–0.94]
	Total		8.4 ± 3.2	7.3 ± 3.2	0.84 [0.76–0.94]	7.9 ± 3.1	0.89 [0.83–0.94]

Data are means ± standard deviations

CI confidence interval, ICC  intraclass correlation coefficient, LDCT low-dose computed tomography, SR-UTE super-resolution ultrashort echo time, UTE ultrashort echo time

Both UTE and SR-UTE tended to underestimate pulmonary nodule diameters compared with CT; however the magnitude of measurement bias was smaller for SR-UTE. For example, in patients with severe emphysema, the mean nodule diameter measured on SR-UTE (7.9 mm ± 3.1) was closer to the CT measurement (8.4 mm ± 3.2) than the diameter measured on UTE (7.3 mm ± 3.2). Subgroup analyses stratified by nodule type and size demonstrated similar findings across GGNs, PSNs, and SNs (Fig. 3). Notably, for nodules measuring < 6 mm in patients with severe emphysema, SR-UTE measurements (range, 4.7–5.3 mm) more closely approximated CT measurements (range, 5.0–5.7 mm) than UTE measurements (range, 4.2–4.8 mm).

Fig. 3.

Representative examples of pulmonary nodules detected using low-dose CT, UTE MRI, and SR-UTE MRI. a–c Axial images of a 19.1 mm solid pulmonary nodule (arrowhead) in the right upper lobe of a patient with mild emphysema (LAA%, 8.9%), shown on a low-dose CT, b UTE, and c SR-UTE. d–f Axial images of a 25.6 mm part-solid nodule (arrowhead) in the left upper lobe of a patient with moderate emphysema (LAA%, 15.6%), shown on d low-dose CT, e UTE, and f SR-UTE. g–i Axial images of a 5.8 mm solid nodule (arrowhead) in the right upper lobe of a patient with severe emphysema (LAA%, 26.3%), shown on g low-dose CT, h UTE, and i SR-UTE

Bland–Altman analysis further demonstrated that both UTE and SR-UTE systematically underestimated pulmonary nodule diameters relative to low-dose CT across all emphysema severity groups (Fig. 4). For UTE, the mean measurement bias increased with worsening emphysema severity, ranging from 0.84 mm in mild emphysema to 0.94 mm in severe emphysema. In contrast, SR-UTE demonstrated a lower and more consistent measurement across severity subgroups (range, 0.47–0.51 mm). In addition, SR-UTE yielded narrower limits of agreement than UTE, with the greatest difference observed in the severe emphysema subgroup (0.18–0.85 mm for SR-UTE vs. 0.47–1.41 mm for UTE).

Fig. 4.

Bland-Altman plots of pulmonary nodule diameter measurements across emphysema severity groups. a–c Comparisons between low-dose CT and UTE MRI in patients with a mild, b moderate, and c severe emphysema. The mean measurement bias ranges from 0.84 to 0.94 mm, with 95% limits of agreement ranging from 0.09 mm to 1.59 mm (bias ± 1.96 SD). d–f Comparisons between low-dose CT and SR-UTE in a patient with d mild, e moderate, and f severe emphysema. The mean measurement bias ranges from 0.47 to 0.51 mm, with 95% limits of agreement ranging from 0.03 mm to 0.92 mm (bias ± 1.96 SD)

Longitudinal assessment of nodule growth

The ability of SR-UTE to assess longitudinal changes in pulmonary nodule size was evaluated using CT-derived diameter change (ΔCT) as the reference standard (Table 5; Fig. 5). SR-UTE demonstrated strong concordance with CT across all growth patterns. For progressive nodules, the mean increase in diameter measured using SR-UTE (2.8 mm ± 0.3) approximated the increase measured on CT (2.9 mm ± 0.3), whereas conventional UTE exhibited a larger negative bias (2.4 mm ± 0.4).

Table 5.

Comparison of nodule diameter changes and longitudinal reproducibility between LDCT and MRI

Nodule Growth Status	ΔCT (mm)	ΔUTE (mm)	ΔSR-UTE (mm)	ICC (UTE1 vs. UTE2) [95% CI]	ICC (SR-UTE1 vs. SR-UTE2) [95% CI]
Progress (> 2 mm)	2.9 ± 0.3	2.4 ± 0.4	2.8 ± 0.3	0.80 (0.69–0.90)	0.89 (0.80–0.95)
Regress (> 2 mm)	−2.3 ± 0.3	−1.7 ± 0.4	−2.1 ± 0.3	0.78 (0.63–0.90)	0.88 (0.77–0.94)
Stable (± 2 mm)	0.2 ± 0.2	0.1 ± 0.2	0.2 ± 0.1	0.84 (0.75–0.91)	0.92 (0.86–0.96)
Total	0.5 ± 0.3	0.4 ± 0.3	0.5 ± 0.2	0.82 (0.75–0.88)	0.90 (0.85–0.95)

Data are means±standard deviations

CI  confidence interval, ICC intraclass correlation coefficient, SR-UTE super-resolution ultrashort echo time, UTE  ultrashort echo time, ICC calculated for the agreement of diameter changes (Δ) relative to LDCT

Δ = change in mean diameter between baseline and follow-up scans

Fig. 5.

Longitudinal assessment of pulmonary nodule growth using low-dose CT, UTE MRI, and SR-UTE MRI

a–c Baseline axial images showing a solid pulmonary nodule (arrowhead) in the right middle lobe measuring 8.3 mm in diameter in a patient with mild emphysema (LAA%, 7.3%), demonstrated on a low-dose CT, b UTE MRI, and c SR-UTE MRI. d–f Corresponding one-year follow-up images of the same nodule demonstrate interval growth to 11.0 mm on d low-dose CT, e UTE MRI, and f SR-UTE MRI. Compared with conventional UTE MRI (b, e), SR-UTE MRI (c, f) provides improved spatial resolution and sharper delineation of nodule boundaries, resulting in reduced partial volume effects and improved visualization of subtle interval growth. The imaging appearance on SR-UTE more closely resembles that of CT, supporting its potential value for longitudinal pulmonary nodule surveillance

Comparable findings were observed for regressive nodules, for which SR-UTE measurements (− 2.1 mm ± 0.3) were closer to CT measurements (− 2.3 mm ± 0.3) than those obtained with UTE. Overall longitudinal reproducibility was higher for SR-UTE than for UTE, with a total ICC of 0.90 (95% CI: 0.85–0.95), compared with 0.82 (95% CI: 0.75–0.88) for UTE. High reproducibility with SR-UTE was also maintained for stable nodules (ICC, 0.92 [95% CI: 0.86–0.96]).

Discussion

In this study, conventional UTE MRI and SR-UTE MRI were compared with CT, the reference standard, for pulmonary nodule detection and characterization in patients with varying degree of emphysema, stratified according to the Fleischner Society criteria [19]. The findings indicate that SR-UTE provided superior image quality, higher nodule detection sensitivity, and more accurate pulmonary nodule size measurements compared with conventional UTE MRI in both cross-sectional and longitudinal evaluations.

MRI of the lung is inherently challenging because of rapid transverse signal decay, which is primarily caused by magnetic susceptibility differences at air-tissue interfaces. These susceptibility effects increase with static magnetic field strength and substantially shorten T2* relaxation times, with reported reductions from approximately 816 ms at 0.2 T to merely 1–1.5 ms at 1.5 T [30].

UTE MRI was therefore developed to overcome these limitations by enabling signal acquisition from tissues with extremely short T2* relaxation times [8]. Although UTE sequences are susceptible to gradient imperfections and eddy current effects, these technical limitations can be mitigated through radial acquisition strategies that reduce pointwise encoding time. The combination of radial and Cartesian encoding trajectories allows more continuous k-space sampling, thereby improving SNR and increasing robustness to B₀ inhomogeneity [31]. Completely calcified pulmonary nodules were excluded because they are typically benign and not clinically relevant for surveillance. Although UTE MRI may show some calcified lesions due to ultra-short T2* signal sensitivity, CT remains far superior for detecting and characterizing calcification. Therefore, these nodules were excluded to maintain a clinically appropriate study population focused on lesions requiring follow-up.

In patients with emphysema, the inherent technical challenges of lung MRI are further exacerbated by reduced proton density, increased susceptibility-related artifacts, and heterogeneous ventilation patterns. These factors collectively reduce signal intensity and degrade lesion conspicuity, creating substantial challenges for the detection and characterization of small or low-density pulmonary nodules. Deep learning-based image reconstruction has demonstrated significant potential for improving MRI performance by reducing image noise, suppressing artifacts, and accelerating image acquisition while simultaneously enhancing spatial resolution [32, 33]. Noise reduction is particularly important in lung parenchymal imaging, where accurate visualization requires detailed representation of fine anatomical structures.

In the present study, SR-UTE consistently demonstrated higher subjective image quality scores and improved objective metrics, including SNR and CNR, compared with conventional UTE across all emphysema severity groups. The improvement was particularly pronounced in patients with moderate and severe emphysema. These findings suggest that super-resolution reconstruction can partially compensate for the signal loss and spatial blurring associated with lung MRI in emphysematous lungs, thereby improving overall image quality. This observation was further supported by quantitative Laplacian Variance analysis. Compared with conventional UTE, SR-UTE demonstrated a 27–32% increase in LV across emphysema severity groups, indicating improved edge definition and spatial resolution. Because accurate delineation of lesion margins is essential for pulmonary nodule measurement and longitudinal assessment, this improvement may contribute to the enhanced measurement consistency observed in the present study. LV has been used as an objective metric for image sharpness assessment [29]. In addition, previous studies evaluating deep learning–based image reconstruction in thoracic imaging have reported improvements in image sharpness and lesion conspicuity [34]. Our findings extend these observations to SR-UTE MRI and provide quantitative evidence that super-resolution reconstruction can improve spatial resolution in pulmonary nodule imaging.

Three-dimensional deep learning-based super-resolution reconstruction techniques have previously been applied primarily in abdominal and pelvic MRI application [17–36]. To the best of our knowledge, this study represents the first investigation of their application in UTE MRI for improving lung image quality.

For pulmonary nodules larger than 8 mm, both UTE and SR-UTE achieved a detection sensitivity of 100%, consistent with previous studies [37]. However, the detection sensitivity for smaller nodules appears to depend on factors such as pulse sequence design, nodule composition, and respiratory motion conditions [38]. Burris et al. and Ohno et al. reported detection rates of 71% and 74.1%, respectively, for UTE MRI in the assessment of pulmonary nodules measuring 4–6 mm [6, 39].

In the present study, conventional UTE detected 30 of 55 pulmonary nodules smaller than 6 mm (54.5%). Missed nodules were predominantly GGNs or PSNs, which are characterized by low proton density and poorly defined margins on MRI. These limitations are further accentuated in moderate to severe emphysema because of increased partial volume effects and respiratory motion, resulting in reduced lesion conspicuity on conventional UTE. However, SR-UTE substantially improved the overall detection rate for nodules smaller than 6 mm to 81.8%, with the most pronounced improvement observed for sub-solid nodules in patients with severe emphysema. This observation was further supported by ROC analysis, which demonstrated higher AUC values for SR-UTE across all emphysema severity groups. These findings suggest that the benefit of SR reconstruction extends beyond improved sensitivity, contributing to enhanced overall pulmonary nodule detection. This improvement may be attributed to reduced high-frequency information loss and edge blurring, thereby improving spatial resolution and lesion conspicuity in lung MRI [17, 40, 41].

Accurate measurement of nodule size is essential for risk stratification and longitudinal management according to the Fleischner Society guidelines [20]. In the present study, both UTE and SR-UTE systematically underestimated pulmonary nodule diameters across all emphysema severity groups, which is consistent with previous reports demonstrating intermethod or inter-reader measurement bias of less than 1 mm [39, 42, 43]. However, the measurement accuracy of conventional UTE declined with increasing emphysema severity. The mean measurement bias increased from 0.84 mm in the mild emphysema to 0.94 mm in the severe emphysema group, suggesting that advancing parenchymal destruction and the associated reduction in SNR exacerbate boundary blurring during radial acquisition.

In contrast, SR-UTE consistently demonstrated superior measurement accuracy compared with conventional UTE, maintaining a stable and lower mean bias (0.47–0.51 mm) across all groups. Significantly, this advantage persisted in patients with severe emphysema, in whom SR-UTE nearly halved the mean measurement bias compared with conventional UTE (0.51 mm vs. 0.94 mm). This improved stability was further evidenced by markedly narrower 95% limits of agreement, reflecting substantially greater measurement reproducibility. Furthermore, in longitudinal follow-up analyses, changes in pulmonary nodule size measured during SR-UTE more closely approximated CT-based measurements, supporting its potential utility for reliable assessment of nodule growth dynamics over time.

From a technical perspective, SR-UTE improves measurement consistency by restoring sharp contrast transitions at the nodule-lung interface and reducing the radial blurring that is inherent to standard acquisitions [44]. The resulting improvement in structural fidelity minimizes voxel-level partial volume averaging and facilitates clearer delineation of nodule boundaries, enabling more consistent linear measurements even in the presence of extensive parenchymal destruction [18]. Consequently, SR-UTE may provide more reliable pulmonary nodule size estimations.

The super-resolution reconstruction used in this study was implemented using the Onekey AI platform, which was developed using large-scale multimodal medical imaging datasets, including CT, MRI, and ultrasound. As a vendor-independent image processing framework, its reconstruction strategy is not inherently restricted to a specific MRI manufacturer or pulmonary imaging protocol. The favorable performance of SR-UTE MRI observed across different emphysema severity groups in the present study suggests that the proposed approach may be applicable under a range of pulmonary imaging conditions. Further evaluation across different imaging environments may provide additional insights into its broader clinical applicability. From a practical perspective, SR reconstruction is performed automatically after image acquisition without extending examination time. This workflow allows SR-UTE MRI to be readily incorporated into routine clinical practice and may support radiation-free longitudinal surveillance of pulmonary nodules.

We acknowledge the difference in Z‑axis resolution between MDCT and MRI. CT provided near‑isotropic resolution (0.625 mm slice thickness), whereas native UTE MRI had a slice thickness of 1.5 mm. The fourfold super‑resolution reconstruction used in this study was designed to enhance resolution in the slice‑encoding direction. Although SR‑UTE cannot fully achieve isotropic resolution comparable to CT, the improved agreement with CT measurements and increased Laplacian Variance suggest that SR reconstruction partially compensates for through‑plane partial volume effects.

This study has several limitations. First, it was conducted at a single center, and the sample size of patients with severe emphysema was relatively small. Further multicenter studies with larger populations are required to validate these findings. Second, CT rather than histopathology was used as the reference standard; therefore, the histologic nature of all pulmonary nodules could not be definitively established. Most nodules were identified in a screening and surveillance setting and were managed according to guideline-based imaging follow-up rather than invasive procedures. Future studies incorporating histopathological confirmation are warranted to further evaluate the role of SR-UTE MRI in differentiating benign and malignant pulmonary nodules. Third, motion artifacts associated with severe dyspnea remain a potential limitation for lung MRI, although SR-UTE demonstrated greater robustness compared with conventional UTE imaging. Another limitation is that SR reconstruction was performed using a commercial software platform. Although the reconstruction workflow is publicly available, detailed information regarding the underlying network architecture and training process remains proprietary, which may limit complete reproducibility. However, the use of a standardized commercial platform may facilitate future external validation.

Conclusion

The findings of this study demonstrate that deep learning-based SR-UTE MRI has potential value for both cross-sectional and longitudinal evaluation of pulmonary nodules in patients with emphysema. SR-UTE provided superior image quality and improved accuracy for pulmonary nodule detection and measurement compared with UTE MRI, with performance approaching that of LDCT. These results indicate that SR-UTE MRI may serve as a promising radiation-free alternative to LDCT for lung cancer screening and long-term pulmonary nodule surveillance in patients at high risk for lung cancer with coexisting emphysema.

Abbreviations

UTE

Ultrashort echo time

SR

Super-resolution

SR-UTE

Super-resolution UTE MRI

LDCT

Low-dose computed tomography

LAA%

Low-attenuation area percentage

GGN

Ground-glass nodule

PSN

Part-solid nodule

SN

Solid nodule

ICC

Intraclass correlation coefficient

Authors' contributions

Conception and design of the research: Hui Feng, Gao-Feng Shi Acquisition of data: Meng-Ting Duan, Hui Liu Analysis and interpretation of the data: Ning Zhang, Meng Li, Statistical analysis: Hui Feng, Meng Li, Ning Zhang Obtaining financing: Hui Feng Writing of the manuscript: Hui Feng, Meng-Ting Duan Critical revision of the manuscript for intellectual content: Gao-Feng Shi, Hui Liu All authors read and approved the final draft.

Funding

1. Medical Science Research Project of Hebei (No. 20201140).

2. Hebei Provincial Department of Finance (No. ZF2025229).

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted with approval from the Ethics Committee of the Fourth Hospital of Hebei Medical University (No. 2020KY082). This study was conducted in accordance with the declaration of Helsinki. Written informed consent was obtained from all participants.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.