Diagnostic accuracy of dual-layer spectral computed tomography virtual monoenergetic imaging with multiplanar reformation for T-staging of colorectal cancer

Abstract

BACKGROUND

Accurate preoperative T staging is essential for determining optimal treatment strategies in colorectal cancer (CRC). Low-keV virtual monoenergetic images (VMIs) have been shown to enhance lesion conspicuity. This study aimed to assess the diagnostic value of dual-layer spectral computed tomography (CT)-derived VMIs, in combination with multiplanar reformation (MPR) and evaluation of peritumoral fat stranding (PFS), for improving the accuracy of T staging in CRC.

AIM

To assess the diagnostic performance of dual-layer spectral CT (DLSCT) VMIs, particularly at low energy levels, and their integration with personalized MPR for preoperative T staging of CRC.

METHODS

In this retrospective study, 157 patients with pathologically confirmed CRC (mean age: 63.5 ± 12.1 years) underwent DLSCT within 1 week before surgery. VMIs ranging from 40 keV to 70 keV (at 10 keV intervals) and conventional polyenergetic images (PEIs) were reconstructed. Objective image quality parameters, including image noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR), were quantified, alongside subjective image quality scores using a 5-point Likert scale. Interobserver agreement was evaluated using κ statistics. Taking histopathology as the reference standard, the diagnostic accuracy of T staging (T1-2 vs T3-4) was compared across PEIs and VMIs, both with and without MPR and PFS.

RESULTS

Low-keV VMIs (40-70 keV) demonstrated significantly higher SNR and CNR than PEIs (all P < 0.001). Notably, 40-keV VMIs achieved noise levels comparable to PEIs (8.17 ± 3.63 vs 8.53 ± 2.90; P = 0.673). Subjective image quality peaked at 40-50 keV VMIs (Likert scores 4.85-4.88 vs 3.97 for PEIs; P < 0.001), supported by excellent interobserver agreement (κ = 0.812-0.913). The combination of 40-50 keV VMIs with MPR yielded the highest T staging accuracy (94.27%) compared to axial PEIs (70.7%), with a sensitivity and specificity of 83.87% and 96.83%, respectively (Youden index = 0.81; P < 0.05). While PFS enhanced staging accuracy on PEIs (up to 77.07% with MPR), it provided no significant additional benefit for VMIs.

CONCLUSION

DLSCT VMIs at 40-50 keV significantly enhanced image quality and improved preoperative T staging accuracy of CRC when combined with MPR. These findings underscored the clinical value of low-keV spectral imaging in tailoring individualized treatment strategies.

Key Words: Colorectal cancer; Dual-layer spectral computed tomography; Multiplanar reformation; Peritumoral fat stranding; T staging; Virtual monoenergetic images

Core Tip: Accurate preoperative T staging is critical for guiding individualized therapeutic strategies in colorectal cancer (CRC). However, conventional computed tomography demonstrates limited diagnostic performance, with reported accuracies ranging from 60.6% to 89%. This study demonstrated that low-energy (40-50 keV) virtual monoenergetic images (VMIs), when combined with multiplanar reformation (MPR), could substantially improve T staging accuracy to as high as 94.27%. While peritumoral fat stranding further enhanced staging accuracy on polyenergetic images, MPR alone achieved an accuracy of 77.07%. These findings supported the incorporation of low-keV VMIs and MPR into routine preoperative imaging protocols to optimize CRC staging and management.

INTRODUCTION

Colorectal cancer (CRC) ranks as the third most commonly diagnosed malignancy worldwide. In 2022 alone, over 1.9 million new cases were reported, with approximately 900000 deaths, figures that underscore significant regional disparities in both epidemiology and clinical outcomes[1]. Accurate preoperative T-staging is pivotal in guiding optimal treatment strategies, including surgical resection, neoadjuvant therapy, or organ-preserving approaches. However, nearly one-quarter of patients with locally advanced CRC are mismanaged due to inaccurate staging[2,3]. The 8^th edition of the American Joint Committee on Cancer (AJCC) TNM classification emphasizes the critical role of precise T staging in risk stratification, prognostic assessment, and the formulation of personalized treatment plans[4].

Medical imaging plays an indispensable role in determining the T-stage of CRC. Each modality offers distinct advantages. Endoscopic ultrasound, which allows detailed visualization of bowel wall layers, is commonly employed for rectal cancer staging[5]. Colonoscopic high-frequency mini-probe ultrasonography demonstrates superior accuracy compared to conventional computed tomography (CT) in the local staging of colonic cancer. Regarding nodal assessment, 12-MHz ultrasonography outperforms both CT and 20-MHz ultrasonography in diagnostic accuracy[6]. Magnetic resonance imaging (MRI) is considered the gold standard for rectal cancer due to its superior soft tissue contrast, achieving a sensitivity of 87% and specificity of 75% for identifying T1-2 and T3-4 tumors. Moreover, MRI effectively assesses tumor penetration depth, nodal involvement, and extramural vascular invasion[7]. In contrast to the high accuracy of MRI in staging rectal cancer, the diagnostic accuracy for T3-stage colon cancer remains relatively low, ranging from 51% to 71%[8]. CT, although lower in soft tissue resolution, remains widely utilized owing to its accessibility and cost-effectiveness. The combined use of MRI and CT for preoperative T-staging has demonstrated superior accuracy in patients with CRC, compared to either modality used alone[9]. CT is particularly valuable for evaluating extramural tumor spread and detecting distant metastases[10]. High-performance CT can contribute to accurate preoperative tumor staging in colon cancer[11].

Traditional CT staging typically relies on axial contrast-enhanced images[11]. One of the most informative imaging signs for assessing tumor infiltration depth and differentiating between T1-2 and T3-4 stages is peritumoral fat stranding (PFS), characterized by fat infiltration or nodularity in the mesorectal region following tumor invasion through the bowel wall. Multiple studies have demonstrated a strong correlation between the presence of PFS and pathological T3-T4 staging, as well as poor prognosis[12-14]. The application of multiplanar reformation (MPR) techniques, which enable coronal and sagittal reconstructions, can significantly enhance spatial resolution. Incorporating MPR into CT interpretation has been shown to improve T-staging accuracy to as high as 89%, facilitating a more precise assessment of tumor invasion and lymph node involvement[15-19].

Dual-layer spectral CT (DLSCT) represents a novel advancement in spectral imaging technology[20]. By simultaneously acquiring both high- and low-energy photon data, DLSCT enables the reconstruction of virtual monoenergetic images (VMIs) across a wide energy range, typically spanning from 40 keV to 200 keV. Low-keV VMIs, particularly at 40 keV, are known to enhance iodine contrast, accentuate differences in tissue attenuation, and improve the conspicuity of lesions, thereby facilitating more precise delineation between malignant and normal tissues[21,22].

Given its ability to enhance contrast and anatomical detail, this technology holds considerable promise beyond CRC, with potential applications across a wide range of gastrointestinal malignancies, especially in early detection, precise staging, and treatment response evaluation[23].

Despite these advantages, the combined use of low-keV VMI and MPR in the context of CRC T-staging has been insufficiently explored. The synergistic integration of enhanced soft-tissue contrast from VMI, improved anatomical resolution from MPR and the visualization of PFS, a key imaging biomarker associated with advanced T-stage, may improve the diagnostic accuracy of CRC. This study aimed to evaluate the diagnostic utility of DLSCT-based VMIs combined with MPR for accurate T staging in CRC, thereby providing a more robust imaging strategy for clinical decision-making.

MATERIALS AND METHODS

Participants

A retrospective study was performed involving 157 patients with pathologically confirmed CRC who were treated at Nantong First People’s Hospital between November 2023 and December 2024. The study protocol received approval from the institutional ethics committee (Approval No. 2025KT124). Inclusion criteria were: (1) No known allergy to iodinated contrast agents; (2) Preoperative contrast-enhanced spectral CT scan of the abdomen conducted within 1 week prior to surgery; (3) No history of neoadjuvant chemotherapy or other preoperative treatments; and (4) Complete clinical records and definitive postoperative pathological staging. Exclusion criteria encompassed: (1) Severe allergy to iodine contrast or impaired renal function (estimated glomerular filtration rate < 40 mL/minute/1.73 m²); (2) Missing or incomplete clinical, surgical, or pathological data; (3) Inadequate image quality preventing accurate assessment; (4) Significant cardiopulmonary dysfunction or coexistence of other malignancies; and (5) Lesions too small to allow reliable measurement or placement of regions of interest (ROIs) on CT images.

Imaging protocol

All patients underwent spectral CT imaging using the IQon Spectral CT scanner (Philips Healthcare, Netherlands), with the scanning range extending from the diaphragmatic dome to the inferior border of the pubic symphysis. The DLSCT acquisition parameters were as follows (Table 1): Tube voltage set at 120 kVp; automatic tube current modulation (mAs); detector collimation of 64 mm × 0.625 mm; gantry rotation time of 0.4 seconds per rotation; and a pitch of 1. An intravenous injection of the nonionic iodinated contrast agent Iohexol (350 mg iodine/mL; Jiangsu Hengrui Medicine Co., Ltd., China) was administered using a high-pressure injector at a rate of 3.0 mL/second, with a dosage of 1.5 mL/kg body weight. Bolus tracking was employed by placing an ROI in the abdominal aorta, initiating arterial phase acquisition when attenuation reached 150 Hounsfield units (HU), followed by venous phase imaging after a 40-second delay. Spectral base images were reconstructed with a slice thickness and reconstruction interval of 1 mm.

Table 1 Dual-layer spectral computed tomography parameters.

Scan parameters and contrast agent injection protocol	Details
Scan parameters
Range	Diaphragmatic dome to the inferior pubic symphysis
Tube voltage (kV)	120
Tube current	Automatic tube current modulation
Detector collimation (mm)	64 × 0.625
Gantry rotation speed (second)	0.4
Pitch	1.0
Contrast agent
Concentration (mg/mL)	350
Dose (mL/kg)	1.5
Rate (mL/s)	3.0
Monitoring
Monitoring method	Bolus tracking
Monitoring level	Abdominal aorta
Monitoring threshold (HU)	150
Scan phases	Arterial phase, venous phase (delayed by 40 seconds)

HU: Hounsfield units.

Image processing

Spectral base images from the venous phase were used to generate both VMIs and polyenergetic images (PEIs) via dedicated post-processing software (IntelliSpace Portal 10.0, Philips Healthcare)[24]. VMIs ranging from 40 keV to 70 keV were reconstructed at 10-keV intervals using a spectral reconstruction algorithm (Spectral level 3, Philips Healthcare). In contrast, PEIs were reconstructed with an iterative reconstruction algorithm (iDose⁴, Philips Healthcare) employing a standard soft tissue kernel. MPR was performed on both VMIs and PEIs at 40-70 keV. Oblique axial (OA) planes were reconstructed perpendicular to the tumor’s longitudinal axis mainly for T-staging, while coronal planes were reconstructed parallel to it for assisting in T-staging. The reformation parameters included a slice thickness and interval of 2 mm each[19].

Objective image evaluation

For each patient, image analysis was performed at the level of the largest tumor cross-section. Circular ROIs were manually placed within the colorectal tumor, avoiding tumor margins, bowel wall vessels, necrotic regions, and areas of peritumoral inflammation. A circular ROI with a diameter of approximately 8 mm and an area of 50 mm² was placed within the subcutaneous fat layer. If the thickness of the subcutaneous fat layer was less than 8 mm, the ROI diameter was adjusted to two-thirds of the measured thickness. The ROI size and location were standardized and replicated across PEI and all VMIs (40-70 keV) to ensure consistency. Each measurement was repeated three times and averaged. Circular ROIs placed within the subcutaneous fat determined the SD, which served as a quantitative surrogate for image noise. Signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were calculated using the following formulas: SNR = HU_tumor/SD_tumor; and CNR = (HU_tumor - HU_fat)/SD_fat.

Subjective image evaluation

Two senior radiologists, with 10 and 15 years of experience in abdominal imaging, independently reviewed all images for objective image evaluation, blinded to clinical and pathological data. Image quality and T-stage were assessed using standardized evaluation criteria. Further methodological details on subjective image quality assessment and T-staging are as follows: Two physicians independently evaluated subjective image quality and T-staging, and interobserver agreement for both assessments was calculated. Discordant T-stage assessments were re-reviewed until consensus was reached, after which diagnostic performance was compared against the pathological gold standard. Image quality was rated on a 5-point Likert scale[25] as follows: Score 5 (excellent) indicated lesions sharply defined with outstanding contrast, minimal image noise, no artifacts, and high diagnostic confidence; Score 4 (good) indicated clearly delineated lesions with excellent contrast, low image noise, mild artifacts, and no change in confidence; Score 3 (moderate) referred to adequately visualized lesions with good contrast, moderate noise and artifacts, decreased confidence but diagnosis still feasible; Score 2 (poor) described difficult lesion identification, suboptimal contrast, strong noise, severe artifacts, and degraded confidence rendering diagnosis questionable; and Score 1 (non-diagnostic) denoted barely visible lesions with poor contrast, very strong noise, severe artifacts, making diagnosis impossible.

T-staging criteria

T-staging was performed in accordance with the 8^th edition of the AJCC TNM classification system[4]. The definitions were as follows: T1, tumor invasion limited to the submucosa; T2, invasion into the muscularis propria; T3, extension beyond the muscularis propria into the subserosa or pericolorectal tissues; T4a, penetration of the visceral peritoneum; and T4b, direct invasion into adjacent organs or structures. Corresponding spectral CT staging criteria[19] were adapted as: T1, lesion confined to the submucosa with smooth bowel wall contours (Figure 1A); T2, involvement of the muscularis propria without evidence of pericolic fat stranding (Figure 1B); T3, irregular bowel wall accompanied by nodular or infiltrative spread into the perienteric fat (Figure 1C); and T4, invasion into the visceral peritoneum or neighboring anatomical structures (Figure 1D). Three diagnostic approaches were evaluated against pathological staging (pT1-2 vs pT3-4): (1) Morphological assessment using spectral CT criteria; (2) Identification of PFS, with a positive finding indicating T3-4; and (3) A composite method combining both criteria, with the higher stage taken as the final judgment.

Figure 1 Computed tomography staging criteria. A: A 73-year-old male patient with histological stage T1 tumor. The 40 keV oblique axial (OA) virtual monoenergetic image in venous phase showed a sigmoid colon mucosal mass (white arrow) surrounding 1/3 of the lumen with clearly visible normal bowel wall muscularis propria appearing (orange arrow) with relatively low density and without peritumoral fat stranding (PFS) indicative of T1 stage disease; B: A 53-year-old female patient with histological stage T2 tumor. The 40 keV OA virtual monoenergetic image in venous phase showed an upper rectal tumor (orange arrow) surrounding 2/3 of the lumen with a smooth outer border of thickened rectal wall and without PFS indicative of T2 stage disease; C: A 65-year-old male patient with histological stage T3 tumor. The 40 keV OA virtual monoenergetic image in venous phase demonstrated a sigmoid colon tumor with an irregular border and (orange arrow) spiculations extending into the peri-rectal fat, along with PFS sign (white arrow) - findings indicative of stage T3 disease; D: A 63-year-old male patient with histological stage T4 tumor. The 40 keV OA virtual monoenergetic image in venous phase demonstrated a tumor at the hepatic flexure surrounding the entire lumen (white arrow) with infiltration into peritumoral fat and invasion of adjacent liver (orange arrow) indicative of T4 stage disease.

Statistical analysis

All statistical analyses were conducted using SPSS software (version 26.0; IBM Corp., Chicago, IL, United States). Continuous variables with normal distributions are presented as mean ± SD, whereas non-normally distributed variables are expressed as median ± interquartile range. Homogeneity of variance was first assessed for objective image quality parameters. When variance was homogeneous, one-way ANOVA followed by least significant difference post hoc tests was applied; otherwise, the Kruskal-Wallis test and Wilcoxon rank-sum test were used for multiple and pairwise comparisons, respectively. Subjective image quality ratings were analyzed using non-parametric tests. Interobserver agreement for both image quality assessment and T-staging was quantified using weighted κ and standard κ statistics. Agreement was categorized as poor (< 0.4), moderate (0.4-0.6), good (0.6-0.8), or excellent (0.8-1.0). Diagnostic performance metrics, including accuracy, sensitivity, specificity, and the Youden index, were compared between VMIs (40-70 keV) and PEIs using the McNemar test. A two-tailed P value < 0.05 was considered statistically significant.

RESULTS

Patients

A total of 157 patients (59 females and 98 males; mean age, 63.5 ± 12.1 years) were included in this study, all of whom had surgically and histopathologically confirmed CRC (Figure 2). Detailed demographic and clinical characteristics are summarized in Table 2.

Figure 2  Patient recruitment flowchart.

Table 2 Patient demographic characteristics.

Characteristic	Percentage/cohort details
Sex, n (%)
Male	98 (62.4)
Female	59 (37.6)
Age (mean ± SD, year)	63.5 ± 12.1
Pathological stage, n (%)
T1	14 (8.9)
T2	17 (10.8)
T3	118 (75.2)
T4	8 (5.1)
Differentiation grade, n (%)
G1	9 (5.7)
G2	142 (90.4)
G3	6 (3.8)
Tumor location
Rectum	87 (55.4)
Sigmoid colon	31 (19.7)
Descending colon	11 (7.0)
Splenic flexure	1 (0.6)
Transverse colon	7 (4.5)
Hepatic flexure	1 (0.6)
Ascending colon	19 (12.1)

Image quality evaluation (objective and subjective)

Significant differences were observed in image noise, SNR, and CNR between the 40-70 keV VMIs and the PEIs (all P < 0.001). Notably, the noise level of the 40 keV VMIs (8.17 ± 3.63) was statistically comparable to that of the PEIs (8.53 ± 2.90, P = 0.673), but noise levels decreased progressively with rising keV levels. No statistically significant difference was found between adjacent VMI energy levels (P > 0.05). Compared to PEIs, VMIs at all energy levels demonstrated significantly enhanced SNR and CNR (SNR: 10.73 ± 3.26 to 18.13 ± 7.09 vs 8.41 ± 2.45, P < 0.001; CNR: 27.89 ± 12.11 to 45.69 ± 22.76 vs 23.09 ± 8.95, P < 0.001). Both SNR and CNR decreased as the keV increased (all P < 0.05; Tables 3 and 4).

Table 3 Subjective and objective image quality assessments of 40-70 keV virtual monoenergetic images and polyenergetic images in colorectal cancer evaluation (n = 157).

Subjective/objective	PEIs	40 keV VMIs	50 keV VMIs	60 keV VMIs	70 keV VMIs	Z value	P value
Noise	8.53 ± 2.90	8.17 ± 3.63	7.27 ± 3.18	6.83 ± 2.73	6.57 ± 2.58	74.521	< 0.001
SNR	8.41 ± 2.45	18.13 ± 7.09	14.81 ± 5.09	12.60 ± 4.05	10.73 ± 3.26	427.420	< 0.001
CNR	23.09 ± 8.95	45.69 ± 22.76	37.85 ± 18.12	31.58 ± 13.60	27.89 ± 12.11	275.540	< 0.001
Subjective scores	3.97 ± 0.42	4.88 ± 0.33	4.85 ± 0.36	4.69 ± 0.50	4.11 ± 0.49	376.086	< 0.001

Objective image quality parameters (noise/contrast-to-noise ratio/signal-to-noise ratio): Reported as median ± interquartile range due to non-normal distribution. Subjective scores: Reported as mean ± SD. CNR: Contrast-to-noise ratio; SNR: Signal-to-noise ratio; PEI: Polyenergetic image; VMI: Virtual monoenergetic image.

Table 4 Intergroup comparisons of subjective and objective image quality parameters for 40-70 keV virtual monoenergetic images and polyenergetic images in colorectal cancer evaluation (n = 157).

Subjective/objective		PEIs	40 keV VMIs	50 keV VMIs	60 keV VMIs	70 keV VMIs
Noise	PEI	-	0.673	< 0.001	< 0.001	< 0.001
	40 keV VMIs	-	-	0.054	< 0.001	< 0.001
	50 keV VMIs	-	-	-	0.873	0.075
	60 keV VMIs	-	-	-	-	1.000
	70 keV VMIs	-	-	-	-	-
SNR	PEIs	-	< 0.001	< 0.001	< 0.001	< 0.001
	40 keV VMIs	-	-	< 0.001	< 0.001	< 0.001
	50 keV VMIs	-	-	-	< 0.001	< 0.001
	60 keV VMIs	-	-	-	-	< 0.001
	70 keV VMIs	-	-	-	-	-
CNR	PEIs	-	< 0.001	< 0.001	< 0.001	< 0.001
	40 keV VMIs	-	-	0.002	< 0.001	< 0.001
	50 keV VMIs	-	-	-	0.001	< 0.001
	60 keV VMIs	-	-	-	-	< 0.001
	70 keV VMIs	-	-	-	-	-
Subjective scores	PEIs	-	< 0.001	< 0.001	< 0.001	0.376
	40 keV VMIs	-	-	1.000	0.033	< 0.001
	50 keV VMIs	-	-	-	0.120	< 0.001
	60 keV VMIs	-	-	-	-	< 0.001
	70 keV VMIs	-	-	-	-	-

CNR: Contrast-to-noise ratio; SNR: Signal-to-noise ratio; PEI: Polyenergetic image; VMI: Virtual monoenergetic image.

Subjective image quality scores also varied significantly across the different energy levels (P < 0.001). The highest scores were consistently achieved at 40 keV and 50 keV VMIs, surpassing those of PEIs and higher-energy VMIs (4.85 ± 0.36 to 4.88 ± 0.33 vs 3.97 ± 0.42 to 4.11 ± 0.49, P < 0.001; Tables 3 and 4).

Interobserver agreement analysis

Cohen’s κ values for interobserver agreement between the two radiologists assessing image quality across 40-70 keV VMIs and PEIs were 0.913, 0.812, 0.812, 0.766, and 0.789, respectively (all P < 0.001). The agreement was classified as excellent (κ ≥ 0.81) for PEIs, 40 keV, and 50 keV VMIs, and as good (κ = 0.61-0.80) for 60 keV and 70 keV VMIs. These findings highlighted the consistently high interrater reliability, particularly at lower keV levels, which are crucial for improving the conspicuity of colorectal lesions. Additional details are presented in Table 5.

Table 5 Interobserver agreement analysis of subjective quality scores between two radiologists for 40-70 keV virtual monoenergetic images and polyenergetic images (n = 157).

Subjective quality scores			Doctor B
			3	4	5	κ	P value
Doctor A	PEIs	3	14	0	0	0.913	< 0.001
		4	2	130	2
		5	0	0	9
	40 keV VMIs	3	0	0	0	0.812	< 0.001
		4	0	15	2
		5	0	4	136
	50 keV VMIs	3	0	0	0	0.813	< 0.001
		4	0	18	2
		5	0	5	132
	60 keV VMIs	3	2	1	0	0.766	< 0.001
		4	1	32	5
		5	0	9	107
	70 keV VMIs	3	11	0	0	0.789	< 0.001
		4	0	107	4
		5	0	11	24

SNR: Signal-to-noise ratio; PEI: Polyenergetic image; VMI: Virtual monoenergetic image.

Interobserver agreement for T-staging assessments

The Cohen’s κ values reflecting interobserver agreement between the two radiologists for T staging assessment across PEIs and 40-70 keV VMI MPR reconstructions were as follows: 0.891 for axial PEIs, 0.899 for OA PEIs, 0.943 for 40 keV OA VMIs, 0.924 for 50 keV OA VMIs, 0.843 for 60 keV OA VMIs, and 0.906 for 70 keV OA VMIs (all P < 0.001). All κ values met or exceeded the threshold for excellent agreement (κ ≥ 0.81), underscoring strong reproducibility in T-stage assessments across both PEIs and VMIs. Notably, OA VMIs demonstrated consistently high agreement, further affirming their reliability in clinical evaluation. Additional details are provided in Table 6.

Table 6 Interobserver agreement for T-staging assessments between two radiologists for 40-70 keV virtual monoenergetic images and polyenergetic images (n = 157).

	Images	T staging	Doctor B
			T1	T2	T3-4	κ	P value
Doctor A	Axial PEIs	T1	2	0	0	0.891	< 0.001
		T2	1	29	3
		T3-4	0	2	123
	OA PEIs	T1	2	0	0	0.899	< 0.001
		T2	1	33	2
		T3-4	0	3	116
	40 keV OA VMIs	T1	15	0	0	0.943	< 0.001
		T2	1	14	2
		T3-4	0	0	125
	50 keV OA VMIs	T1	13	0	0	0.924	< 0.001
		T2	2	15	2
		T3-4	0	0	125
	60 keV OA VMIs	T1	4	1	0	0.843	< 0.001
		T2	2	21	3
		T3-4	0	2	124
	70 keV OA VMIs	T1	3	0	0	0.906	< 0.001
		T2	1	27	3
		T3-4	0	1	122

PEI: Polyenergetic image; VMI: Virtual monoenergetic image; OA: Oblique axial.

Diagnostic efficacy of preoperative T-staging in CRC

The comparison between subjectively assessed imaging and postoperative pathological staging (pT stage) revealed that the diagnostic accuracy of axial PEIs for CRC was 70.7% (sensitivity 38.71%, specificity 78.57%). Evaluation based solely on the presence of PFS achieved an accuracy of 68.79% (sensitivity 54.84%, specificity 72.22%), while the combination of morphological assessment and fat stranding yielded an identical accuracy of 70.7% (sensitivity 38.71%, specificity 78.56%).

With the use of MPR for OA reconstruction, diagnostic performance improved. For OA PEIs, morphological evaluation achieved an accuracy of 76.43% (sensitivity 45.16%, specificity 84.13%), PFS showed 73.89% (sensitivity 80.65%, specificity 72.22%), and the combined evaluation further enhanced accuracy to 77.07% (sensitivity 45.16%, specificity 84.92%).

The highest diagnostic performance was observed with OA 40 and 50 keV VMIs, yielding an accuracy of 94.27% (sensitivity 83.87%, specificity 96.83%; Figures 3 and 4) and a Youden index of 0.81. This was significantly superior to axial PEIs, OA PEIs, and 60-70 keV VMIs. Diagnostic accuracy declined progressively with increasing keV levels. When relying solely on the PFS sign, the accuracy of 40 and 50 keV VMIs dropped to 73.25% and 76.43%, respectively. Complete data are presented in Table 7.

Figure 3 T-staging in axial polyenergetic image, oblique axial polyenergetic image, and 40-70 keV oblique axial virtual monoenergetic images for T3 stage by pathology. An upper rectal cancer patient with axial polyenergetic image (PEI), oblique axial (OA) PEI, and 40-70 keV oblique virtual monoenergetic images (VMIs), was proved to be T3 by pathology. A: Axial PEI showed a smooth outer border of the thickened rectal wall without peritumoral fat stranding (PFS), suggesting stage T2; B-F: Oblique PEI (B) and 70-40 keV OA VMIs (C-F) demonstrated a smooth outer border of the thickened rectal wall with spiculation extending into the perirectal fat (orange arrow) but without PFS, suggesting stage T3. Compared to axial PEI (A), the lesion was better visualized at 40-50 keV due to increased lesion attenuation and hyper-enhancement at the anterior rectal wall. Among all images, the 40-50 keV OA VMIs (E and F) performed best.

Figure 4 T-staging in axial polyenergetic image, oblique axial polyenergetic image and 40-70 keV oblique axial virtual monoenergetic images for T2 stage by pathology. A middle rectal cancer patient with axial polyenergetic image (PEI), oblique PEI and 40-70 keV oblique axial (OA) virtual monoenergetic images (VMIs), was proved to be T2 by pathology. A: Axial PEI showed a smooth outer border of the thickened rectal wall with peritumoral fat stranding (PFS; white arrow), suggesting stage T3; B-F: Oblique PEI (B) and 70-40 keV OA VMIs (C-F) demonstrated the lesion located in the left anterior wall (orange arrow) of the rectal wall and the outer border of the thickened rectal wall was smooth. No spiculations extending into the perirectal fat and PFS was found, suggesting stage T2. Compared to axial PEI (A), the lesion was better visualized at 40-50 keV VMIs due to increased lesion attenuation and hyper-enhancement. Among all images, the 40-50 keV OA VMIs (E and F) performed best.

Table 7 Diagnostic performance of 40-70 keV virtual monoenergetic images and polyenergetic images in preoperative staging of T1-2 and T3-4 colorectal cancer.

Images	CT T-staging	Pathological staging		Accuracy (%)	Sensitivity (%)	Specificity (%)	Youden index
		T1-2 (n = 31)	T3-4 (n = 126)
Axial PEIs	T1-2	12	19	70.7	38.71	78.56	0.17
	T3-4	27	99
PFS	T1-2	17	14	68.79	54.84	72.22	0.27
	T3-4	35	91
Axial PEIs and PFS	T1-2	12	19	70.7	38.71	78.56	0.17
	T3-4	27	99
OA PEIs	T1-2	14	17	76.43	45.16	84.13	0.29
	T3-4	20	106
PFS	T1-2	25	6	73.89	80.65	72.22	0.53
	T3-4	35	91
OA PEIs and PFS	T1-2	14	17	77.07	45.16	84.92	0.30
	T3-4	19	107
40 keV OA VIMs	T1-2	26	5	94.27	83.87	96.83	0.81
	T3-4	4	122
PFS	T1-2	25	6	73.25	80.65	71.43	0.52
	T3-4	36	90
40 keV OA VMIs and PFS	T1-2	25	6	93.63	80.65	96.83	0.77
	T3-4	4	122
50 keV OA VMIs	T1-2	26	5	94.27	83.87	96.83	0.81
	T3-4	4	122
PFS	T1-2	25	6	76.43	80.65	76.19	0.57
	T3-4	36	96
50 keV OA VMIs and PFS	T1-2	25	6	93.63	80.65	96.83	0.77
	T3-4	4	122
60 keV OA VMIs	T1-2	18	13	84.08	58.06	90.48	0.49
	T3-4	12	114
PFS	T1-2	25	6	73.25	80.65	71.43	0.52
	T3-4	36	90
60 keV OA VMIs and PFS	T1-2	18	13	84.08	58.06	90.48	0.49
	T3-4	12	114
70 keV OA VMIs	T1-2	16	15	80.25	51.61	87.30	0.39
	T3-4	16	110
PFS	T1-2	25	6	73.25	80.65	71.43	0.52
	T3-4	36	90
70 keV OA VMIs and PFS	T1-2	16	15	80.25	51.61	87.30	0.39
	T3-4	16	110

PEI: Polyenergetic image; VMI: Virtual monoenergetic image; PFS: Peritumoral fat stranding; OA: Oblique axial.

DISCUSSION

In this study, we systematically evaluated both subjective and objective image quality parameters of PEIs and 40-70 keV VMIs, alongside interobserver consistency in image quality and T-stage assessments. Furthermore, we compared the diagnostic accuracy of T staging using PEIs and 40-70 keV VMIs, with or without MPR and PFS, against pathological staging. Our findings revealed that 40-50 keV VMIs demonstrated noise levels comparable to those of PEIs, while achieving significantly superior SNR, CNR, and subjective image quality scores. Notably, the 40 keV and 50 keV OA VMIs exhibited excellent interobserver agreement for both image quality and T-stage evaluations, achieving a diagnostic accuracy of 94.27% (Youden index: 0.81). While the addition of PFS improved T staging accuracy for both axial and oblique PEIs, it conferred no substantial benefit for 40-70 keV VMIs.

Virtual monoenergetic imaging, a derivative of DLSCT, enhances lesion conspicuity, particularly for small lesions. In our cohort, noise levels at 40-50 keV (8.17 ± 3.63) were statistically indistinguishable from those of PEIs (8.53 ± 2.90, P = 0.673). Additionally, image noise declined steadily with increasing keV levels (up to 70 keV), remaining consistently below that of PEIs. This could be attributed primarily to the use of a DLSCT system, which enabled fully time- and position-matched projection-domain reconstruction of low- and high-energy data VMIs, thereby reducing beam hardening artifacts and convolution kernel-related artifacts[20,26,27], suppressing anticorrelated noise between the two datasets[28,29], and further diminishing image noise through model-based iterative reconstruction algorithms[30]. This approach fundamentally differs from traditional dual-source dual-energy CT platforms (e.g., Siemens systems)[31], which perform image domain reconstruction. These findings are consistent with those reported by Arico' et al[22] and El Kayal et al[32]. Given the K-edge of iodine at 33.2 keV, photon energies around 40 keV maximized iodine attenuation, optimizing lesion conspicuity by enhancing the contrast between the tumor and normal bowel wall. Consequently, 40 keV VMIs achieved superior SNR and CNR compared to PEIs, with performance declining at 70 keV (CNR: 27.89 ± 12.11), consistent with prior studies by Arico' et al[22] and Jia et al[33].

Consequently, the superior tumor delineation achieved by low keV VMIs was reflected in significantly higher subjective image quality scores (4.88 ± 0.33 and 4.85 ± 0.36 for 40 keV and 50 keV, respectively), compared to 70 keV (4.11 ± 0.49) and PEIs (3.97 ± 0.42). Interobserver agreement was also excellent for PEIs and 40-50 keV VMIs (κ = 0.812-0.913), but dropped markedly at 60-70 keV (κ = 0.766-0.789), likely due to diminished contrast resolution at higher energy levels. Despite the high consistency in PEI image quality assessments, their diagnostic accuracy for T staging remained limited at 70.7%, underscoring their inferiority to VMIs in reliably staging colorectal tumors.

Accurate preoperative T staging of CRC is pivotal for guiding individualized treatment plans and estimating patient prognosis. Current staging techniques include morphological analysis, detection of PFS, quantification of imaging biomarkers[34], identification of extramural vascular invasion[17], and integration of multiparametric spectral data[35-37]. In our study, axial PEIs yielded a T staging accuracy of 70.7% (sensitivity: 38.71%; specificity: 78.57%), which improved with OA MPR reconstruction to 76.43% (sensitivity: 45.16%; specificity: 84.13%). These findings aligned with prior reports demonstrating staging accuracies ranging from 60.6% to 87.1%[38-40], although slightly below the 81.6%-89% reported in higher-performing studies[15,16,18,19,41-43]. The use of MPR markedly enhances visualization of tumor invasion depth and its relationship with surrounding pericolonic fat across multiple planes, overcoming the limitations of conventional axial PEIs and substantially improving the comprehensiveness of tumor invasion assessment. This benefit has been validated in previous literature[44].

Importantly, implementing 40-50 keV VMIs markedly improved T-staging accuracy, increasing it from 80.25% (sensitivity: 51.61%; specificity: 87.30%) to 94.27% (sensitivity: 83.87%; specificity: 96.83%; Youden index: 0.81). This improvement was largely attributed to more precise differentiation of early-stage tumors (T1-T2), with correct identification increasing from 12/31 to 26/31 cases. Enhanced tumor tissue exhibited distinct attenuation characteristics compared to the normal bowel wall. VMIs at 40-50keV, predominantly governed by the photoelectric effect, significantly increased contrast between tumor tissue and normal bowel wall, creating a clear attenuation gradient. This facilitated superior visualization of tumor boundaries, mucosal disruption, and muscular layer involvement compared to conventional CT. Notably, enhanced tumor tissue achieved optimal iodine contrast at 40 keV VMIs, closely aligning with the iodine K-edge (33.2 keV)[26,45]. The synergistic effect of 40-50 keV VMIs combined with MPR enabled detailed morphological evaluation of tumor invasion, compensating for the inherent limitations of conventional axial CT[33,46] and ultimately improving the accuracy of CT-based staging. However, it is worth noting that the dichotomization of T-stages into T1-2 vs T3-4 may have led to a slight overestimation of diagnostic accuracy.

PFS may indicate either direct tumor infiltration[24] or inflammatory changes in fat metabolism[47]. While integrating MPR with PEIs improved diagnostic accuracy from 70.7% to 77.07%, this enhancement was not observed with VMIs. This discrepancy may arise from differences in attenuation characteristics of peritumoral fat and tumor tissues at varying energy levels in VMIs. Tissue attenuation at lower energies is predominantly governed by the photoelectric effect. Due to its lower electron density and effective atomic number, peritumoral fat absorbs more low-energy photons, resulting in relatively lower attenuation values[27,48]. Conversely, tumor tissues exhibit a nonlinear increase in attenuation as keV decreases, amplifying the contrast between tumor and peritumoral fat. PEIs show higher attenuation values in peritumoral fat compared to low-keV VMIs, leading to relatively reduced contrast with tumor tissue. Consequently, subtle fat stranding sometimes visible on PEIs becomes less conspicuous on low-keV VMIs owing to these attenuation properties. In contrast, low-keV VMIs provide superior delineation of tumor margins and infiltration depth, particularly via OA morphological evaluation. Thus, the inherent advantages of low-keV imaging outweigh the potential additional value of PFS. Moreover, PFS may be confounded by inflammatory changes, as previously reported[49]. Additionally, PFS has been correlated with circumferential tumor involvement, lesion length, positive resection margins, and lymphovascular invasion[14,50]. Therefore, T-staging can be assessed by observing the extent of tumor invasion on low-keV OA VMIs, rather than relying solely on the presence of PFS.

This study demonstrated an improved T-staging accuracy of 94.27%. Compared to axial PEIs, the number of correctly identified T1-2 stage cases increased from 12/31 (38.71%) to 26/31 (83.87%), while correctly staged T3-4 cases rose from 99/126 (78.57%) to 122/126 (96.83%) following the application of 40 keV OA VMIs. According to the 2024 NCCN guidelines for CRC[2,3], surgical resection is recommended for T1-2 stages, whereas radical surgery combined with adjuvant therapy, including preoperative chemoradiotherapy for CRC patients, is advised for T3-4 stages to enhance resection rates and survival outcomes. Improved T-staging accuracy facilitates more precise therapeutic decision-making, such as avoiding unnecessary neoadjuvant therapy in T1-2 patients and ensuring timely preoperative chemoradiotherapy for T3-4 patients, ultimately improving surgical outcomes[51]. Notably, the 5-year survival rate for T3 stage CRC patients increases from 73.0% to 77.2% with accurate staging[52]. Moreover, precise preoperative T staging reduces complication rates related to intraoperative vascular and adjacent tissue injury[17] and serves as a critical prognostic indicator, with reported 5-year survival rates for non-metastatic colon cancer of 96% for T1, 89% for T2, 75% for T3, and 79% for T4[53]. Furthermore, T stage strongly correlates with recurrence risk, as evidenced by 5-year local disease-free survival rates of 84.0% for T4 vs 98.6% for T3[54]. Previous studies have consistently confirmed that accurate T staging aids clinical decision-making, minimizes complications, and improves prognosis[51-54]. Nevertheless, the clinical implications of enhanced T-staging accuracy using 40 keV OA VMIs, particularly regarding treatment planning, intraoperative complications, and long-term prognosis, warrant further prospective investigation.

Limitations of this study should be acknowledged. First, this was a retrospective, single-center investigation, and thus, the generalizability of our findings requires validation through larger, multicenter studies. Second, this study focused on T-staging interpretation using OA VMIs and did not incorporate quantitative parameters such as spectral curve slope. Future research should investigate the predictive value of parameters including 40-70 keV interval slopes, iodine concentration, normalized iodine concentration, electron density, and effective atomic number for staging, prognosis, and neoadjuvant therapy efficacy, thereby advancing the multiparametric evaluation system. Third, MRI, the established gold standard for rectal cancer staging, was not employed for comparative validation. Fourth, the clinical implications of improved T-staging accuracy using 40 keV OA VIMs, particularly regarding treatment planning, intraoperative complications, and prognosis enhancement, still require further prospective studies.

CONCLUSION

In conclusion, 40-50 keV VMIs derived from DLSCT significantly enhanced both subjective and objective image quality in the preoperative T staging of CRC. Compared with PEIs, 40-50 keV VMIs demonstrated superior noise characteristics, higher SNR and CNR, excellent interobserver agreement (κ ≥ 0.81), and outstanding diagnostic accuracy (94.27%, Youden index: 0.81). These findings established low-keV VMIs as the optimal imaging choice for evaluating colorectal neoplasms. Although MPR reconstruction improved PEI performance in OA views (accuracy: 77.07%), VMIs at 40-50 keV consistently outperformed both PEIs and higher-energy VMIs, particularly in the accurate delineation of early-stage tumors (T1-2).