Ultra-High Resolution Photon Counting Detector Computed Tomography Imaging for Quantitative Lung Assessment -An Anthropomorphic Phantom Study

Abstract

Background:

Quantitative lung imaging is utilized to understand, characterize, and monitor lung disease and response to interventions. X-ray computed tomography has remained the modality of choice for clinical lung assessment and photon counting detector-computed tomography (PCD-CT) is the latest advancement. PCD-CT provides increased spatial and contrast resolution, decreased image noise and artifacts (such as beam hardening) and, thus, a potential for enhanced image quality for equivalent or reduced radiation dose levels. However, evaluation of the ultra-high resolution capabilities of PCD-CT for quantitative lung imaging have not yet been systematically investigated.

Purpose:

This study aims to evaluate two ultra-high resolution acquisition modes and four reconstruction kernels for optimal quantitative chest imaging at high radiation dose (9mGy). We assess the stability of measurements across different scan modes and reconstruction kernels when the radiation dose level is reduced.

Methods:

A customized anthropomorphic chest phantom, containing standardized insert materials including air, water, various density foam inserts, and a modulation transfer function (MTF) cube was repeatedly scanned with PCD-CT (NAEOTOM Alpha; Siemens Healthineers). Two ultra-high resolution acquisition modes, quantum plus (UHRQ+) and quantum with tin filtering (UHRQSn), and four reconstruction kernels (Br64, Bl60, Qr60 and Qr40, all with iterative reconstruction level 3) were examined with acquisitions at three radiation dose levels (9.1mGy, 6.8mGy, 3.2mGy). Quantitative density measures, airway measurements, contrast-to-noise ratio (CNR), signal-to-noise ratio (SNR), and MTF values were compared, along with the percentage change in measurement values from high to low radiation dose levels.

Results:

At the highest radiation dose levels, UHRQ+ acquisition resulted in lower density values with higher standard deviation compared to UHRQSn. UHRQ+ mode demonstrated higher CNR, SNR, and MTF values. Only UHRQ+ with Qr40 reconstruction provided accurate air measurements, both inside and outside the phantom, across all radiation dose levels. Quantitative density measurements remained highly stable (<2% change) as the radiation dose was reduced from 9.1 mGy to 3.2 mGy. Airway wall thickness, diameter and lumen area measurements were all larger with UHRQ+ acquisition compared to UHRQSn for the high radiation dose level. At low radiation dose levels, the UHRQ+ acquisition with Br64 reconstruction maintained the highest consistency in airway metrics compared to the values from the high dose acquisition, with less than 5% measurement percentage change.

Conclusion:

The UHRQ+ mode is recommended for quantitative lung assessment, leveraging the PCD-CT voxel size potential (1024×1024 in plane matrix with 0.2mm slice thickness). The choice of reconstruction kernel at ultra-high resolution should be task-specific, with Qr40 being optimal for density assessment due to its accuracy in air measurement across regions and Br64 for airway assessment. The high consistency of measurements across the radiation dose levels for these kernels (<5% measurement change from 9mGy measurements) suggests that acquisition at 3mGy is sufficient for quantitative analysis.

INTRODUCTION

Over the past 40 years, quantitative, volumetric x-ray CT has evolved, offering an increased understandings of lung disease etiology and progression (1). Measurements are affected by the lung inflation state, altered voxel sizes influenced by reconstructed field of view, reconstruction kernel, artifacts such as beam hardening and image noise contributed to by the need for digital conversion required to capture the signal residing on energy integrating detectors. The recent advent of ultra-high-resolution photon-counting detector (PCD) technology has the potential to significantly improve quantitative computed tomography (CT) assessment of lung anatomy and regional indices of lung function. PCD-CT offers enhanced contrast-to-noise ratios, superior spatial resolution, reduced scanning artifacts, and better structural visualization, at equivalent or reduced radiation exposure compared to energy integrating detector (EID) technology (2).

The key innovation of PCD technology is the direct transformation of x-ray photons into electrical signals in the semiconductor, bypassing the two-step process of EIDs (3, 4). In EID systems, x-rays are converted to visible light within the scintillation crystal and then to electrical signal by a photodiode. PCDs eliminate the need for septa between detector pixels to minimize cross talk, thereby enabling smaller pixel sizes and greater dose efficiency. Furthermore, PCD-CT’s capability to measure the energy level of detected x-ray photons, coupled with advanced iterative reconstruction methods, facilitates improved contrast at lower X-ray energies and ultra-high resolution imaging (0.2 mm detector at 1024×1024 pixel matrix, compared to traditional 0.6 mm detector at 512×512). Ultra-high resolution acquisitions therefore allow for unprecedented in vivo anatomical detail in the lung.

Previous studies have compared the ultra-high resolution acquisition mode of PCD-CT (1024×1024) to EID-CT in human subjects for lung disease evaluation, using equivalent or reduced radiation doses and expert reader qualitative assessments (5–7). These studies have demonstrated that PCD-CT provides enhanced visualization of distal bronchial divisions (5, 6), sharpness of bronchial wall (5, 6), sharpness of fissure (5, 6), linear opacities (6), bronchiectasis (6), honeycombing (6), and reticulation (7). However, findings for ground-glass opacities were mixed, with one study indicating improvement with PCD-CT (7) and another preferring EID-CT (5). Additionally, one study demonstrated that EID-CT was favored for sharpness of pulmonary vessels (5), and interstitial abnormalities (5). Expert reviews further discuss the potential of PDC-CT for lung assessment (8, 9).

For EID-CT, spectral shaping via tin (Sn) filtering has been a valuable option for retaining image quality while reducing radiation dose levels in chest imaging (10–12). Tin filtering is a scan acquisition option aimed at optimizing the spectrum for dose efficiency by blocking low energy x-rays produced by the source from reaching the patient which optimizes image quality at the interface between soft tissue and air and improves beam hardening. We incorporate tin filtering as an acquisition mode for this study to observe if spectral shaping for PCD-CT increases the stability of quantitative metrics as radiation dose level is reduced.

Given the advancements and potential clinical advantages of PCD-CT, it is crucial to examine and optimize acquisition and reconstruction settings for quantitative assessments of lung density and structure. To date, it has been well recognized that image noise increases with reduced voxel size if radiation dose is held constant. This study aims to evaluate the performance of ultra-high resolution PCD-CT in the context of lung imaging by utilizing an anthropomorphic phantom. We compare two ultra-high resolution acquisition modes (with and without tin filtered spectral shaping) at a radiation dose level comparable to a standard clinical chest CT exam (9mGy), to determine optimal performance capacity and examine the impact on quantitative measurements as radiation dose is reduced to levels used for quantitative chest CT multi-center trials (7–3mGy) and low-dose CT for lung cancer screening (3mGy).

METHODS

PCD-CT Acquisition & Reconstruction

Ultra-high resolution scans were acquired on the PCD-CT (NAEOTOM Alpha; Siemens Healthineers, Forchheim, Germany; software version VA50) using 120 × 0.2 mm detector configuration, 140 kVp tube voltage, dose modulation (CARE keV), and rotation time of 0.25 seconds (Table 1). Two scan modes were included, the ultra-high resolution quantum plus (UHRQ+) and the ultra-high resolution quantum with tin (UHRQSn), both had 0.2 mm slice thickness and 1024 × 1024 matrix size. Pitch varied for the tin (Sn) filtered images (0.35 – 1.0) to match CTDI_vol to the non-tin filtered images which had a consistent pitch of 1.0. Three radiation levels were targeted – clinical chest CT (~9mGy), multi-center trial CT (~7mGy), and low dose CT (~3mGy). Using quantum iterative reconstruction at strength three (QIR-3), four reconstruction kernels were obtained - two sharp kernels (Bl60 and Br64) and two quantitative kernels (Qr60 and Qr40).

Table 1:

Acquisition and reconstruction parameters for photon counting detector computed tomography (PCD-CT) chest phantom scans.

Parameters	UHRQ+	UHRQSn
Acquisition
Detector Configuration, # x mm	120 x 0.2 mm	120 x 0.2 mm
Rotation time, s	0.25	0.25
Pitch	1.0	0.35, 0.45, 1.0
Tube potential, kVp	140	Sn140
Care keV	On	On
Care keV IQ	80, 57, 26	103, 75, 36
CTDIVol, mGy	9.1, 6.8, 3.2	9.1, 6.8, 3.2
Reconstruction
Slice thickness x spacing, mm	0.2 x 0.15	0.2 x 0.15
Matrix Size	1024 x 1024	1024 x 1024
Field of view, mm	260	260
Kernels	Bl60, Br64, Qr40, Qr60	Bl60, Br64, Qr40, Qr60
Iterative reconstruction	QIR; strength 3	QIR; strength 3

Phantom

To evaluate the effects of acquisition scan mode, radiation dose, and reconstruction kernel on the PCD-CT images, a PH-1 N1 LUNGMAN chest phantom (Kyoto Kagaku Co., Ltd, Japan) with an added attenuation vest was utilized (Figure 1). The phantom contained tubes with differing types of material inserts for quantitative evaluation. A 42mm diameter tube, placed in the right lung, included two water inserts, a porcine fixed lung tissue sample (Lung Sample A), and a modulation transfer function (MTF) cube which was custom-built at the Medical Instruments Shop at University of Iowa Health Care. A 22mm diameter tube, in the left lung, contained standardized foam inserts of varying densities from the National Institute of Standards and Technology (NIST). Abdominal air was calculated via a region of interest (ROI) within the 42mm tube, tracheal air via an ROI in the phantom trachea, and outside air was calculated via a ROI outside of the phantom. Schematic of the phantom inserts and ROIs can be seen in Figure 1A.

Figure 1:

Anthropomorphic chest phantom with inserts for quantitative measurement evaluation. A) Phantom insert configuration including right lung Large Tube (42mm) with water (H₂O) inserts, fixed Lung Sample A, modulation transfer function (MTF) cube, and Abdominal air region. Air measurements were also assessed in the trachea of the phantom and outside the phantom (not illustrated in the schematic). Left lung small tube includes varying density foam inserts from the National Institute of Standards and Technology (NIST) (acrylic insert not analyzed). B) Compilation of coronal slices showing the automated segmentations (blue outlines) of the inserts, including the parenchymal lung foam region. C) Example of the automated airway measurement made in the axial plane of Lung Sample A. D) Zoomed in example of the three parenchymal holes measuring approximately 1.5mm.

Image Analysis

Image analysis was performed using the in-house developed Pulmonary Analysis Software Suite (PASS) (13), which automatically segmented the regions of interest. Analysis consisted of the average and standard deviation (SD) of Hounsfield units (HU) within the insert samples. The percentage change in quantitative measures for a given scan acquisition mode, from high dose to medium and low dose was also calculated. Contrast-to-noise (CNR) and signal-to-noise (SNR) ratios were calculated for each scan using the following equations:

$$CNR=\frac{(NIST12mean-NIST8mean)}{{SD}_{outsideair}}$$

$$SNR=\frac{\left|meanofparechymallungequivalentfoam\right|}{{SD}_{outsideair}}$$

The 3D MTF cube was made of solid high-density polyethylene. Five of the cube’s sides were used to create the edge spread and line spread functions which were then used to produce the MTF along the x, y, and z-axes (14). The 20% and 50% of peak modulations were used to assess delineation. MTF curves were plotted using R (version 4.3.0).

Airway measurements including inner and outer diameter, wall thickness, and lumen area were also calculated via PASS. These measurements were created by placing a seed point in the center of the airway. The same airway, at approximately the same axial slice, was used for all scans. Figure 1C shows an example of the output after the seed point has been placed.

Qualitative visual comparison between scan acquisition modes, reconstruction kernels and radiation dose levels included axial and coronal views, visualization of details in the fixed Lung Sample A, and three small parenchymal holes within the phantom lung that appeared outside of the insert tubes as seen in Figure 1D.

RESULTS

Visual inspection

At a high radiation dose (9mGy) level, for any given reconstruction kernel, the UHRQSn acquisition featured less streaking (beam hardening) compared to the UHRQ+ acquisition in the simulated fat surrounding the ribs in the axial view (Figure 2) and at the apex of the lungs in the coronal view (Supplemental Figure 1). This reduction in noise around the ribs and apex of the lung is less evident at low dose (3mGy). At 3mGy, the UHRQ+ acquisition provides clearer small structure definition compared with UHRQSn, as made evident by the clarity of the boundaries of the small parenchymal holes shown in Figure 2 (right inset). Visual inspection of the four reconstruction kernels investigated, identified Qr60 as providing a balance between low noise and edge detection (Figure 2).

Figure 2:

Representative images for photon counting protocol comparison across two radiation dose levels (9mGy and 3mGy), two scan acquisition modes (HRUQ+ and UHRQSn), and four reconstruction kernels (Br64, Bl60, Qr60, and Qr40). Axial cross section of phantom at the level of Lung Sample A, NIST8 and Bone 50% (not analyzed). Inset imaging show the Lung Sample A which is contained in a tube of 42 mm diameter (left inset) and three parenchymal holes of approximately 1.5 mm each (right inset).

High radiation dose protocol comparison

Comparing quantitative measurements for the phantom inserts from UHRQ+ (Table 2) and UHRQSn (Table 3) scan modes for the high radiation dose protocol revealed a consistent trend in lower standard deviations in the UHRQSn acquisitions compared to the equivalent kernel reconstruction acquired with UHRQ+ mode. For the outside air measurement, the standard deviation was lower in the UHRQ+ acquired data. The NIST foam insert measurements showed a consistent trend (across all reconstruction kernels) to lower mean HU density with UHRQ+ compared to UHRQSn acquisition mode. For the lowest density NIST foam insert (NIST 4 lb), which mimicked emphysematous lung parenchyma, the mean HU difference between scan modes was the greatest (UHRQ+ - UHRQSn ranged from −12.73HU (Br64) to −11.46HU (Qr60)). These observed trends in standard deviation, and mean HU measurements of the NIST foam persisted at the low radiation dose (Supplemental Tables 1 and 2).

Table 2:

Quantitative measurements for PCCT phantom acquisition with high radiation dose (9mGy) and UHRQ+ acquisition mode (no tin filtering)

UHRQ+	Br64		Bl60		Qr60		Qr40
HU	mean	SD	mean	SD	mean	SD	mean	SD
Abdominal Air	−1023.99	106.57	−1023.96	118.66	−1015.10	65.53	−1004.16	24.85
Tracheal Air	−1006.06	95.75	−1017.23	106.95	−999.89	59.37	−990.93	25.23
Outside Air	−1012.17	56.21	−1012.92	63.62	−1004.34	35.83	−1000.43	12.76
Water Vial 1	−0.74	119.43	0.05	136.33	0.62	73.30	−1.12	28.75
Water Vial 2	−1.28	113.39	−0.30	127.88	−2.62	70.49	−1.79	28.75
Lung Sample A	−630.09	289.65	−648.86	363.19	−614.10	248.77	−601.93	221.94
Foam Inserts
NIST 4 lb	−955.62	98.67	−957.84	111.23	−950.70	60.03	−942.50	23.47
NIST 8 lb	−902.42	92.41	−903.10	103.36	−896.02	57.26	−889.50	21.29
NIST 12 lb	−834.73	88.91	−835.39	99.50	−828.11	55.57	−822.00	20.10
NIST 14 lb	−791.24	90.99	−791.56	101.75	−784.34	55.83	−779.92	20.56
NIST 20 lb	−694.30	110.05	−694.86	123.12	−687.84	67.13	−686.52	25.07
Ratios
CNR	1.20		1.06		1.90		5.29
SNR	11.36		10.6		17.71		49.66
MTF cycles/cm
CF50%, In Plane	8.34		8.56		6.19		3.47
CF20%, In Plane	11.35		12.51		9.47		5.21
CF50%, Z-Plane	7.80		9.93		7.67		7.74
CF20%, Z-Plane	14.73		17.42		14.29		14.52

Table 3:

Quantitative measurements for PCD-CT phantom acquisition with high radiation dose (9mGy) and UHRQSn acquisition mode (includes tin filtering)

UHRQSn	Br64		Bl60		Qr60		Qr40
HU	mean	SD	mean	SD	mean	SD	mean	SD
Abdominal Air	−1010.54	86.55	−1011.72	96.42	−1002.13	53.40	−992.86	20.87
Tracheal Air	−991.20	85.61	−1002.52	95.58	−984.89	52.85	−977.02	23.46
Outside Air	−1009.64	62.79	−1009.53	71.22	−1000.68	38.79	−995.94	14.43
Water Vial 1	−0.49	101.78	−0.35	115.97	−1.56	62.35	−0.89	25.09
Water Vial 2	−3.17	98.42	−2.85	111.28	−1.88	61.06	−3.58	24.74
Lung Sample A	−617.77	278.49	−636.68	348.96	−603.61	242.63	−592.68	219.34
Foam Inserts
NIST 4 lb	−942.90	87.14	−945.40	98.06	−939.24	53.89	−930.64	21.09
NIST 8 lb	−892.88	80.15	−894.36	90.39	−887.11	48.84	−880.72	18.83
NIST 12 lb	−827.70	76.05	−829.41	85.23	−821.20	46.13	−816.65	17.07
NIST 14 lb	−784.17	78.31	−785.88	87.78	−777.68	47.80	−774.76	17.77
NIST 20 lb	−685.63	87.26	−687.65	97.73	−680.73	53.79	−680.09	20.32
Ratios
CNR	1.04		0.91		1.70		4.44
SNR	9.99		8.84		16.12		43.31
MTF cycles/cm
CF50%, In Plane	8.23		8.59		5.74		3.37
CF20%, In Plane	11.09		12.56		9.07		5.11
CF50%, Z-Plane	7.48		10.07		7.53		7.49
CF20%, Z-Plane	13.80		16.80		13.40		13.81

There was a systematically higher contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR) in the UHRQ+ compared to UHRQSn mode acquisitions at the high radiation dose level (9mGy). The highest CNR (5.29) and SNR (49.66) was with the UHRQ+ acquisition mode and Qr40 reconstruction kernel. The lowest CNR (0.91) and SNR (8.84) was with the UHRQSn acquisition mode, and Bl60 kernel.

The spatial frequency cutoffs for the modulation transfer function (MTF) were slightly higher for the UHRQ+ versus UHRQSn acquisition mode for the Br64, Qr60 and Qr40 reconstruction kernels respectively (refer to Supplemental Figure 2 for the MTF curves). The highest spatial frequency cutoffs for the MTF were achieved with the Bl60 reconstruction kernel for which the 50% modulation cutoff was 8.59 cycles/cm in-plane and 10.07 cycles/cm in the z-direction (Table 2). As expected, the Qr40 kernel, being the softest kernel included in the evaluation, had the lowest spatial frequency cutoff for the MTF, with the 50% modulation cutoff being 3.37 cycles/cm in-plane and 7.49 cycles/cm in the z-direction.

Comparison to reference values for water and air measurements

Table 4 illustrates the reference value minus the measurement value for air and water regions of interest within the phantom. For quantitative measurement of water (water 1 and water 2) in the anthropomorphic chest phantom, measurements were within ±4HU of the reference value (0HU) for all radiation dose levels, scan modes and reconstruction kernels.

Table 4:

The difference between reference values for air (-1000HU) and water (0HU) and the measured Hounsfield Unit values from the anthropomorphic phantom across two scan acquisition modes (UHRQ+ and UHRQSn), three radiation dose levels (9mGy, 6mGy, and 3mGy), and four reconstruction kernels (Br64, Bl60, Qr60, and Qr40). All measured water values were within ±4HU of the reference value. Measured air values differences more than ±10HU from the reference value are highlighted in red.

Scan Mode	Dose Level	Recon Kernel	Difference From Reference Values (HU)
			Tracheal Air	Abdominal Air	Outside Air	Water 1	Water 2
UHRQ+	HIgh	Br64	6.06	23.99	12.17	0.74	1.28
UHRQ+	HIgh	Bl60	17.23	23.96	12.92	−0.05	0.30
UHRQ+	HIgh	Qr60	−0.11	15.10	4.34	−0.62	2.62
UHRQ+	HIgh	Qr40	−9.07	4.16	0.43	1.12	1.79
UHRQ+	Med	Br64	7.52	25.73	16.80	−0.48	1.86
UHRQ+	Med	Bl60	18.22	24.63	15.54	−1.09	0.80
UHRQ+	Med	Qr60	0.87	16.74	5.81	0.38	1.00
UHRQ+	Med	Qr40	−8.33	4.56	0.62	0.21	2.43
UHRQ+	Low	Br64	18.65	27.42	19.39	−0.72	1.22
UHRQ+	Low	Bl60	17.78	24.61	16.75	−1.72	−0.07
UHRQ+	Low	Qr60	1.39	18.14	12.19	1.58	2.94
UHRQ+	Low	Qr40	−8.23	4.97	0.37	0.54	1.84
UHRQSn	HIgh	Br64	−8.80	10.54	9.64	0.49	3.17
UHRQSn	HIgh	Bl60	2.52	11.72	9.53	0.35	2.85
UHRQSn	HIgh	Qr60	−15.11	2.13	0.68	1.56	1.88
UHRQSn	HIgh	Qr40	−22.98	−7.14	−4.06	0.89	3.58
UHRQSn	Med	Br64	−7.07	12.34	10.59	1.22	2.24
UHRQSn	Med	Bl60	3.54	12.38	10.22	0.99	1.96
UHRQSn	Med	Qr60	−14.12	3.77	2.19	0.71	2.70
UHRQSn	Med	Qr40	−22.53	−7.13	−3.93	1.47	2.90
UHRQSn	Low	Br64	13.84	16.25	17.72	−1.56	1.12
UHRQSn	Low	Bl60	11.54	13.80	13.55	−2.59	0.14
UHRQSn	Low	Qr60	−8.78	7.54	6.50	0.15	1.23
UHRQSn	Low	Qr40	−20.57	−5.72	−3.71	0.42	2.37

For air measurements, only the UHRQ+ acquisition mode with the Qr40 reconstruction kernel was within ±10HU of the air reference standard (−1000HU) for all three radiation dose levels, and across all three air ROIs (outside, tracheal, and abdominal). For the UHRQSn acquisition mode, at the high radiation dose level, the outside air measurements were within ±10HU across all four reconstruction kernels. The differences from reference standard for the Qr60 and Qr40 kernels remained within ±10HU as dose decreased for outside air and abdominal air measurements but were greater than 10HU for tracheal air (Table 4).

Airway assessment

The airway measurements were made using the fixed airway in the Lung Sample A insert (Figure 1C), hence there is no reference standard to which measurement values can be compared; however, measures were generally consistent with the relative performance based on the MTF. In general, the UHRQ+ acquisition mode resulted in slightly larger airway-associated metrics (lumen area, inner diameter, outer diameter, and wall thickness) compared to UHRQSn for the equivalent reconstruction kernel (Table 5). In the Qr40 reconstructions, across both acquisition modes and all radiation doses, there was larger outer diameter (ranging from 5.04 ± 0.65mm to 5.33 ± 0.42mm) and wall thickness measurements (ranging from 1.33 ± 0.18mm to 1.49 ± 0.25mm), while all other reconstruction kernels (Br64, Bl60, Qr60) produced measurements in alignment with each other (outer diameter measurement range from 4.02 ± 0.38mm to 4.64 ± 0.53mm and wall thickness measurement range from 0.76 ± 0.11mm to 1.07 ± 0.19mm).

Table 5:

Airway measurements for a target airway in the phantom’s fixed Lung Sample A insert across two scan acquisition modes (UHRQ+ and UHRQSn), three radiation dose levels (9mGy, 6mGy, and 3mGy), and four reconstruction kernels (Br64, Bl60, Qr60, and Qr40). Measurements from the Qr40 reconstructed scans have larger outer diameter and wall thickness measurements compared to the measures from other reconstruction kernels.

Scan Mode	Dose Level	Recon Kernel	Inner Diameter (mm)		Outer Diameter (mm)		Wall Thickness (mm)		Lumen Area (mm2)
			mean	SD	mean	SD	mean	SD	mean
UHRQ+	HIgh	Br64	2.63	0.23	4.40	0.15	0.89	0.11	5.39
UHRQ+	HIgh	Bl60	2.55	0.28	4.36	0.18	0.91	0.09	5.01
UHRQ+	HIgh	Qr60	2.56	0.29	4.29	0.34	0.87	0.20	4.94
UHRQ+	HIgh	Qr40	2.37	0.23	5.04	0.35	1.34	0.18	4.34
UHRQ+	Med	Br64	2.57	0.28	4.53	0.34	0.98	0.14	5.20
UHRQ+	Med	Bl60	2.46	0.19	4.59	0.42	1.07	0.19	4.76
UHRQ+	Med	Qr60	2.62	0.29	4.38	0.67	0.88	0.25	5.32
UHRQ+	Med	Qr40	2.31	0.14	5.28	0.51	1.49	0.25	4.18
UHRQ+	Low	Br64	2.65	0.37	4.41	0.30	0.88	0.21	5.59
UHRQ+	Low	Bl60	2.58	0.33	4.51	0.24	0.97	0.12	5.27
UHRQ+	Low	Qr60	2.71	0.27	4.22	0.39	0.76	0.11	5.98
UHRQ+	Low	Qr40	2.46	0.26	5.23	0.35	1.38	0.16	4.77
UHRQSn	HIgh	Br64	2.50	0.25	4.15	0.17	0.83	0.08	4.94
UHRQSn	HIgh	Bl60	2.35	0.26	4.12	0.21	0.89	0.06	4.40
UHRQSn	HIgh	Qr60	2.46	0.31	4.07	0.27	0.80	0.13	4.67
UHRQSn	HIgh	Qr40	2.33	0.23	5.10	0.81	1.38	0.36	4.34
UHRQSn	Med	Br64	2.64	0.24	4.64	0.53	1.00	0.27	5.37
UHRQSn	Med	Bl60	2.44	0.22	4.54	0.44	1.05	0.29	4.62
UHRQSn	Med	Qr60	2.69	0.21	4.06	0.31	0.69	0.07	5.46
UHRQSn	Med	Qr40	2.34	0.12	5.04	0.65	1.35	0.33	4.26
UHRQSn	Low	Br64	2.68	0.34	4.20	0.32	0.76	0.25	5.61
UHRQSn	Low	Bl60	2.62	0.28	4.34	0.23	0.86	0.14	5.14
UHRQSn	Low	Qr60	2.27	0.36	4.02	0.38	0.88	0.17	3.94
UHRQSn	Low	Qr40	2.40	0.18	5.33	0.42	1.47	0.27	4.51

Stability of quantitative measurements as radiation dose level is reduced

To compare the measurement stability across scan modes (UHRQ+ and HURQSn) and reconstruction kernel (Br64, Bl60, Qr60, and Qr40) as the radiation dose level was decreased, the percentage measurement change was calculated. Figure 3 tabulates the quantitative percentage measurement change from high dose to low dose (high dose to medium dose data is shown in Supplemental Figure 3). The percentage change values were low for the NIST foam inserts (0% to −0.6%) and low for regions of air (−0.02% to 1.22%), demonstrating high concordance with the measures acquired at the high radiation dose level.

Figure 3:

Percentage measurement change comparing the high dose (9mGy) to the low dose (3mGy) quantitative values.

The largest change impacted the CNR and SNR measurements which decreased ~40% at low dose for UHRQSn, and ~30% for UHRQ+. Percentage measurement change was variable for the airway quantification with a tendency for larger values for airway structures at low dose (indicated by positive percentage measurement change values). Lumen area had the greatest percentage measurement change from high to low dose, across scan modes and reconstruction filters, with larger values (+3.9 to +21% change) for all but one acquisition/kernel reconstruction (UHRQSn, Qr60, −15.6% change). Wall thickness on the other hand had variable percentage measurement change across scan modes and reconstruction kernels, ranging from −12% to +9.24%. The UHRQ+ with Br64 reconstruction had the highest consistency for airway measurements at reduced acquisition radiation dose, with percentage measurement change within ±5% for wall thickness, inner and outer diameter, and airway lumen area.

DISCUSSION

The anthropomorphic phantom allows for the careful comparison of quantitative results across multiple PCD-CT scan acquisition settings. This study has focused on the comparison of ultra-high resolution modes (with and without tin filtering), and a range of radiation dose levels. We also compared results across a range of reconstruction kernels that have been used in previous PCD-CT studies: Br64 (15), Bl60 (5, 6, 16), Qr60 (17), Qr40 (2, 16, 18). Importantly, our results indicate that for the use of the UHRQ+ mode for quantitative lung assessment, the quantitative density and airway measurements were highly stable as radiation dose was reduced from 9mGy to 3mGy, with the expected measurement changes occurring in CNR, SNR and consequently in in-plane MTF (19).

Qualitative visual inspection of the image data revealed UHRQSn provided small improvements to images generated using the highest radiation dose levels, particularly with less beam hardening at the apex of the lung. However, at lower doses, UHRQ+ provided clearer boundary visualization of small objects. In particular, the Qr60 reconstruction kernel was found to have the best compromise for the parameter settings studied between image noise and delineation of small structures (i.e. the three parenchymal holes illustrated in Figure 2). This was supported quantitatively by the higher CNR, SNR and MTF values for the UHRQ+ acquisition mode. As expected, the sharper kernels have the best MTF performance, while the softest kernel has the best CNR and SNR performance, thus kernel selection at high dose should depend on the task priority.

Data was reconstructed to maximize the spatial resolution possible with the ultra-high resolution scan modes, including both the expanded in-plane matrix (1024 × 1024) and 0.2mm in-plane detector size and slice thickness (0.15mm). With these setting we found the UHRQ+ acquisition mode with Qr40 reconstruction kernel was the only acquisition mode and kernel combination that had water, outside air, abdominal air and tracheal air measurement within empirical tolerances of ±4HU of 0HU for water, and ±10HU of −1000HU for air across all three radiation dose levels. This deviation from −1000HU for higher frequency kernels may be due to the combined impact of the very small slice thickness and iterative reconstruction (QIR 3). In another study, similar protocol settings, but using thicker reconstruction slice thickness (>0.5mm) with QIR 2 and QIR 4, found air measurements in the anthropomorphic phantom did not have such pronounced deviations from −1000HU (20). The standard deviation for outside air measurements was lower, compared to tracheal and abdominal air, in both the UHRQ+ and UHRQSn (but to a lesser degree) scan modes. This finding mirrors that from Malkus et al. who reported an approximately factor of two decrease in standard deviation in outside air compared to within a water phantom (21). Our results indicate further refinement of reconstruction algorithms may be required for quantitative assessment at ultra-high resolution with 0.2mm slice thickness.

In this study we did not find superior quantitative assessment concordance at low radiation dose acquisition with UHRQSn acquisition mode. This indicates, that at the doses explored in this study, there is minimal improvement for beam hardening and image quality through tin filtering, beyond what is achieved through the energy discrimination from PCD-CT technology. Spectral shaping with tin filtration resulted in a trend to higher density (HU) values and smaller airway values compared to UHRQ+ mode at high radiation dose levels. Higher HU measurements with UHRQSn, especially for high density materials, is expected due to the adjusted mean energy with spectral shaping compared to non-tin filtered and use of polychromatic reconstruction. As an aside, this is also why URHQSn mode is not recommended for contrast-enhanced CT. When looking at the percentage measurement change from high radiation dose to low radiation dose, there was not increased measurement stability at low dose with tin filtration. However, it is possible that the UHRQSn has some image quality advantages at ultra-low radiation dose levels (<3mGy) not examined in this study. Another consideration for the comparison of the scan modes is acquisition time. The UHRQSn acquisition at highest and medium radiation dose levels had a smaller pitch (0.35 and 0.45 respectively, compared to 1.0 for other protocols), and hence slightly longer acquisition time (~10sec). For the UHRQSn mode, the scanning platform did not allow increased CARE keV IQ (which adjusts the reference mAs) to a level that matched the highest and medium dose levels, and hence the pitch had to be adjusted to achieve matched CTDI_vols. While not evaluated in the anthropomorphic phantom, it is possible that human volunteer or patient scanned in vivo over the longer acquisition time could result in decreased image quality from cardiac motion.

This study does have some limitations. The focus was on the evaluation of quantitative measurement relevant to the lung using an anthropomorphic phantom, which has been utilized in other studies looking at quantitative lung imaging CT protocol development and comparison (22, 23). However, phantom data does not capture some of the complexity of human subjects, such as cardiac motion and/or variations in body mass index. Future work will be a translational study to validate findings in a biological model, as we have previously performed (11, 24), before translating the protocol to quantitative lung assessment in humans. In this study, we did not explore the impact of different iterative reconstruction levels. We selected to focus on one quantum iterative reconstruction level (QIR 3) based on prior work by Sartoretti et al. illustrating QIR 3 performed best for image quality and sharpness by expert readers(25), and other work in the literature focusing on QIR 3 (5, 6, 26). The reconstruction kernels selected focused on those from prior publications, future work will explore the Qr64 reconstruction kerel for airway evaluation.

Results from this study indicate use of the UHRQ+ mode for quantitative lung assessment. With PCD-CT technology, the quantitative density and airway measurements were highly stable as radiation dose was reduced from 9mGy to 3mGy, with the only substantial measurement changes occurring in CNR, SNR and consequently in-plane MTF. Ideally, two reconstructions would be produced for quantitative lung assessment: (1) Qr40 used for density assessment due to accuracy of air measurement across regions and high CNR and SNR, and (2) Br64 used for airway evaluation as, despite a slightly lower MTF spatial frequency cutoff than the Bl60 kernel, the Br64 kernel produced more consistent airway wall, diameter and lumen area measurements as dose was lowered from 9mGy to 3mGy.