Performance of virtual unenhanced images on a prototype silicon photon counting detector CT: preliminary clinical results

Abstract

Background

To compare virtual unenhanced images (VUE) from a prototype deep silicon photon-counting detector computed tomography (PCD CT) system with true unenhanced images (TUE).

Methods

Five subjects including various anatomic structures were scanned on a prototype deep silicon PCD CT at 120 kV across different phases with mA adjusted for anatomy and subject size. Five thin and three thick slices were selected from each image set at two locations to provide a variety of anatomy, including vessels, soft tissue, and adipose. Regions of interest were drawn over the anatomy on both TUE and VUE images. VUE error was defined as VUE CT number in Hounsfield units (HU) minus the TUE CT number.

Results

For all tested structures, 82% had a VUE error less than 10 HU, and 62% had a VUE error less than 5 HU. Vessels demonstrated VUE errors less than 10 HU and less than 5 HU for 70.4% and 40.7% of tested structures, respectively. Soft tissue had VUE errors less than 10 HU and 5 HU in 96.9% and 84.4%, and adipose had VUE errors less than 10 HU and 5 HU in 75% and 58.4% of tested structures.

Conclusion

With the prototype scanner and VUE algorithm, the majority of tested anatomic structures demonstrated VUE error less than 10 HU. For all but vessels, the majority of structures demonstrated VUE error less than 5 HU. With such small differences in HU between VUE and TUE images, further work may support removing TUEs from scanning protocols.

Introduction

Virtual unenhanced (VUE) images are synthetic unenhanced images derived from a contrast-enhanced dataset to simulate true unenhanced (TUE) images in computed tomography (CT). VUEs, also known as virtual non-contrast images (VNC) or virtual nonenhanced (VNE) images, have traditionally been offered with all dual-energy (DE) CT implementations regardless of spectral data acquisition method [1, 2]. There are two main use cases for VUE image sets: providing a non-contrast phase when the protocol does not contain one (e.g., routine abdominopelvic protocol), and to replace a TUE phase in protocols that commonly have them (e.g., mesenteric ischemia, active bleeder protocols urography) [3, 4]. Replacement of TUEs with VUEs offers potential dose and time savings benefits [5]. Additionally, any motion between a non-contrast phase and contrast enhanced phase in a scanning protocol would be eliminated.

VUEs have demonstrated promising performance, particularly in characterizing enhancing renal masses using a single-phase protocol [3, 6, 7], however inter-vendor variability has limited widespread clinical utility [8, 9]. Photon-counting detector (PCD) CT is an emerging technology that provides increased spatial resolution, decreased noise, increased iodine conspicuity, and prospective spectral imaging without the techniques required in current DECT implementations (e.g., requiring a special dual-layer detector, two separate tubes, or a system capable of quickly switching between two tube currents) [10, 11]. It has also been suggested that spectral resolution may be improved with PCD CT [10, 11] and current work suggests that spectral capabilities of PCD CT systems shows improvement over DECT systems [12–15], though some studies demonstrate that DECT systems are superior for certain situations (large phantoms or dense materials) [16]. This work presents a preliminary examination of the performance of a prototype deep silicon PCD CT system and associated prototype VUE algorithms. Ideally, this scanner would present similar performance to cadmium telluride based PCD CT systems such that the DECT inter-vendor variability decreases and clinical adoption of VUEs increases.

Methods

This was a prospective pilot study on human subjects scanned on an investigational deep silicon PCD-CT scanner (GE Healthcare, Waukesha WI). The study was approved by the University of Wisconsin Institutional Review Board (2022 − 1043). Eligible patients were adults of 18 years or older who had previously undergone imaging at the University of Wisconsin Hospital within the last 4 months, were not currently pregnant or lactating, were not allergic to iodine contrast agents, did not have chronic kidney disease, and were not in need of emergent care. Informed consent was obtained from all eligible subjects. This research was conducted in accordance with the International Council for Harmonisation guidelines and the Helsinki Declaration.

In this work, 5 subjects were scanned across 6 different anatomic regions (chest n = 1, head n = 1, wrist/hand n = 1, and abdomen n = 3) and 4 phases (TUE, portal venous, arterial, and late arterial) were scanned on the prototype deep silicon PCD CT with 120 kV, mA adjusted for anatomy and subject size, and 0.42 mm slice thickness. The vendor reconstructed images using a prototype VUE algorithm. Details about the scanning protocol for each subject are available in Table 1. All subjects were injected with IOHEXOL 350 mg/mL (GE HealthCare, Waukesha, WI), dosing was based on patient weight and institutional protocols.

Table 1.

Scan protocol details for all subjects

Subject Number [Body Region; Indication]	Tube Voltage [kV]	Tube Current [mA]	Pitch	Rotation Time [s]	Reconstruction Slice Thickness [mm]	Field of View [mm]	Kernel	Contrast Details: Volume, Rate, Delay	CTDIvol [mGy]
1 [Wrist; Wrist pain]	120	290	0.510	0.7	1.25, 0.50	240	Standard	92mL, 3.0mL/s, 90s (venous),	27.91
2 [Abdomen, pelvis; Recurrent urinary tract infection]	120	290	0.990	0.6	3.75, 0.80	400	Standard	69mL, 2.5mL/s, 70s (venous)	12.34
3 [Chest, Abdomen; Rib fracture]	120	290	0.990	0.8	2.50, 1.0	500	Standard	72mL, 2.5mL/s, 15s (arterial)	16.45
4 [Head; Migraines]	120	SmartmA	0.510	0.5	5.00, 0.50	240	Standard	86mL, 4.0mL/s, 20s (venous)	34.73 (TUE) 20.32 (Venous)
5 [Abdomen, pelvis; Coronary artery disease]	120	SmartmA	0.990	1.0	3.75, 0.60	300	Standard	69mL, 4.0mL/s, 25s (arterial), 70s (venous)	6.35 (TUE) 6.28 (Arterial) 6.23 (Venous)

Five consecutive slices (thin slices) and three consecutive slices (thick slices) were selected from each image set (i.e., one series of the TUE, one series of the VUE for each contrast phase (true non-contrast, portal venous, arterial, late arterial) for each subject) at two locations to encompass vessels, soft tissue, and adipose tissue. Table 2 details the exact structures used for assessment, and the number of ROIs of each type of structure over all the subjects.

Table 2.

Structures assessed for VUE error

Structure Category	Identity	Number of ROIs
Vessels	Aorta	6
	Inferior vena cava	2
	Small arteries (superior mesenteric artery)	1
	Small veins (basilic vein, cephalic vein)	5
	Sigmoid sinus	2
Soft Tissue	Lateral liver	3
	Posterior liver	3
	Spleen	3
	Psoas muscle	2
	Masseter muscle	3
	Paraspinal musculature	8
	Semispinalis capitus	1
	Flexor digitorum superficialis	2
	Abductor pollicis brevis	2
Adipose Tissue	Subcutaneous fat	20

Circular regions of interest (ROIs) were placed on contrast-enhanced phase such that the ROI was as large as possible over a structure. The ROIs were transferred from the contrast-enhanced phase to the same location in the TUE and VUE image sets (for all contrast phases) and were visually checked to account for patient motion between contrast phases. VUE was calculated for each anatomy by measuring CT number of the anatomy on each of the five slices and averaging the CT number over the slices, then subtracting the averaged CT number on the TUE from the VUE (Eq. 1).

VUE_error was defined as follows:

1

Where is the CT number in Hounsfield units (HU) on VUE and is the CT number on TUE. Published literature has used VUE_error cutoffs of 5 and 10 HU as a metric for VUE performance. We adapted this methodology in this work [17, 18].

Results

Of tested structures, 82% and 62% had a VUE_error less than 10 HU and 5 HU, respectively. For vessels, 70.4% and 40.7% showed VUE_error less than 10 HU and 5 HU, respectively. Small vessels and adipose were the most likely to have a VUE error greater than 5 or 10 HU. For soft tissue, 96.9% and 84.4% showed a VUE_error less than 10 HU and 5 HU, respectively. For adipose, 75% and 58.4% showed a VUE_error less than 10 HU and 5 HU, respectively. The average absolute VUE error was 9.8 HU for vessels, 2.6 HU for soft tissue, and 4.9 HU for adipose (Fig. 1). Clinical examples are shown in Figs. 2, 3, 4, 5 and 6, depicting TUE and VUE generated from the late arterial phase (Fig. 3, subject 5) and portal venous phase (Figs. 2 (subject 3), 3 (subject 5), 4 (subject 4), 5 (subject 1), and 6 (subject 2)). Figures 2, 3, 4, 5 and 6 demonstrate that the subtraction of iodine signal in the selected tissues produce similar CT numbers between the TUE and VUE images. Figure 7 demonstrates one subtraction image between the TUE and VUE to demonstrate a VUE_error map. This VUE_error map demonstrates that there was some movement between the contrast phases, but that the majority of the VUE_error is located in the bones (which were not assessed in this study) not the tissue.

Fig. 1.

Box and whisker plot of VUE_error for vessels (red), soft tissue (blue), and adipose (yellow) regions of interest in all clinical subjects

Fig. 2.

Subject 1 (A) True unenhanced (TUE) phase, (B) TUE with virtual unenhanced (VUE) algorithm applied acting as control, (C) portal venous phase, (D) portal venous phase with VUE algorithm applied. (WW/WL: 400/50)

Fig. 3.

Subject 2 (A) True unenhanced (TUE) phase, (B) TUE with virtual unenhanced (VUE) algorithm applied acting as control, (C) portal venous phase, (D) portal venous phase with VUE algorithm applied. Note that the aorta and portal veins are well subtracted in the portal venous phase. (WW/WL: 400/50)

Fig. 4.

Subject 3 (A) True unenhanced (TUE) phase, (B) TUE with virtual unenhanced (VUE) algorithm applied acting as control, (C) arterial phase, (D) arterial phase with VUE algorithm applied. (WW/WL: 400/50)

Fig. 5.

Subject 4 (A) True unenhanced (TUE) phase, (B) TUE with virtual unenhanced (VUE) algorithm applied acting as control, (C) portal venous phase, (D) portal venous phase with VUE algorithm applied. (WW/WL: 400/50)

Fig. 6.

Subject 5 (A) True unenhanced (TUE) phase, (B) TUE with virtual unenhanced (VUE) algorithm applied acting as control, (C) late arterial phase, (D) late arterial phase with VUE algorithm applied, (E) portal venous phase, (F) portal venous phase with VUE algorithm applied. Note that the inferior vena cava and aorta are well subtracted. (WW/WL: 200/50)

Fig. 7.

Subject 2 (A) TUE – VUE of TUE (WW/WL: 100/0) (B) TUE – VUE of portal venous phase (WW/WL: 500/0)

Discussion

This preliminary pilot study demonstrated that when using a prototype deep silicon PCD CT system, VUE images have small mean (5–10 HU) differences in CT number compared to TUE images (i.e., small VUE_error). Therefore, in a larger patient population, we expect the deep silicon PCD CT to adequately subtract iodine signal from contrast-enhanced images to produce accurate VUEs.

Previous works with PCD CT [12–14] found that VUE_errors ranged from 1 to 5 HU depending on the structure analyzed and the contrast phase. Haag et al. 2024 found that on a commercially available cadmium telluride PCD CT system VUE_errors ranged from less than 1 HU (liver arterial phase) up to 24 HU (adipose arterial phase), and errors were larger than 10 HU for the aorta and adipose tissue in both the arterial and venous phases [12]. In comparison, this work demonstrates errors less than 10 HU for both aorta and adipose in arterial and venous phases. Another group, Dudás et al. 2024, demonstrated that VUE_error on spleen, liver, and muscles was less than 1 HU in arterial phase, and less than 3 HU in venous phase [13]. For the same structures in this work, the VUE_error was only slightly larger, with 2 HU error in arterial phase and 4.5 HU in venous phase. Sartoretti et al. 2022 found that the mean VUE_error in 15 patients was 3.7 HU, which is similar to both the work from Haag et al. 2024 and Dudás et al. 2024, as well as similar to the results here for soft tissue and adipose tissue [12–14]. Therefore, this work corroborates with current literature of VUE imaging on PCD CT where the majority of VUE_error on various structures are smaller than 10 HU. Specifically, this preliminarily supports a vendor neutral performance of VUE images in the clinical setting – an issue unable to be reconciled with previous DE CT technological implementations due to the differences in the DECT technology, which affect the generalizability of results.

There are several limitations to this work. First, there were only five subjects, which limits the generalizability and statistical testing of these results. Second, this work was performed with both a prototype scanner and a prototype VUE algorithm, both of which could change upon commercial release. Future work would include phantom based VUE_error analysis and a large clinical dataset encompassing a wide range of acquired contrast-enhanced phases coupled with a reader study to increase the strength and generalizability of this work.

Despite limitations, these preliminary results on a prototype silicon based PCD CT scanner show that there is potential for VUE images to appropriately simulate TUE images. Therefore, this work suggests that future multiphasic CT protocols on this scanner may not require a dedicated TUE phase instead substituting VUE to reduce dose and time to patients, as seen with dual energy CT literature. This is particularly relevant in dose-sensitive patient populations such as pediatric, pregnant, or frequently scanned patients while leveraging the advertised benefits of PCD CT such as improved spatial resolution, noise, and iodine contrast [19].

Author contributions

All authors contributed to the study conception and design. Data collection and analysis were performed by MYan, MYue, KM, ZY, AMS, TPS, MGL, and GVT. AMS and GVT wrote the first draft. All authors read and approved the final manuscript.

Funding

Funding provided by GE HealthCare via an Institutional grant.

Data availability

All data generated or analyzed during this study are included in this published article.

Declarations

Ethics approval and consent to participate

The study was approved by the University of Wisconsin - Madison Institutional Review Board (2022 − 1043). Informed consent was obtained from all eligible subjects. Clinical Trial number: NCT06769594. Registration date: 1/6/2025.

Consent for publication

Not applicable.

Competing interests

AMS declares no conflict of interest. TPS receives research support from Canon Medical Systems and GE HealthCare, consulting fees from Alara Imaging, Imalogix, Aidoc, and Asto CT/Leo Cancer Care; royalties from Medical Physics Publishing and royalties related to intellectual property from Qaelum; founder of RadUnity. MGL spouse consultant, Elphas Bio. M.Yan, M.Yue., KM, and ZY are employees of GE HealthCare. GVT is a consultant to GE HealthCare and Canon Medical Systems.