ABSTRACT

PURPOSE

To explore the value of dual-layer spectral detector computed tomography (DLSCT)-derived parameters from peritumoral fat for preoperative tumor budding (TB) grade assessment in patients with colorectal adenocarcinoma (CRAC).

METHODS

This retrospective study enrolled 90 patients with pathologically confirmed CRAC who underwent preoperative DLSCT between February 2024 and July 2025. Patients were stratified into low-grade and intermediate- to high-grade TB groups based on histopathology. The effective atomic number (Z_eff), iodine concentration (IC), normalized IC concentration (NIC), virtual monoenergetic images (VMIs) at 40, 50, 60, and 70 keV in arterial phase (AP) and venous phase (VP), and the arterial enhancement fraction of intratumor and peritumoral fat were measured. Independent predictors were identified via multivariable logistic regression, and diagnostic performance was evaluated using receiver operating characteristic analysis.

RESULTS

Multiple DLSCT-derived parameters of the peritumoral fat, including VMIs (40–70 keV) and Z_eff in both AP and VP, as well as NIC in VP, were significantly higher in the intermediate- to high-TB group than in the low-TB group (all P < 0.05; AP-VMI 40 keV, P = 0.001; AP-VMI 50 keV, P = 0.002; AP-VMI 60 keV, P = 0.001; AP-VMI 70 keV, P = 0.001; AP-Z_eff, P = 0.002; VP-VMI 40 keV, P = 0.004; VP-VMI 50 keV, P = 0.001; VP-VMI 60 keV, P = 0.001; VP-VMI 70 keV, P < 0.001; VP-VMI 70 keV, P < 0.001; VP-NIC, P = 0.024; VP-Z_eff, P = 0.019), whereas other parameters derived from DLSCT were not significantly different between these two groups (P > 0.05). Multivariable analysis revealed that VMIs (40–70 keV) and Z_eff in both AP and VP, as well as NIC in VP, remained independent predictors of TB grade. Among these, VMI at 70 keV in VP exhibited the strongest discriminatory ability [area under the curve (AUC): 0.742, 95% confidence interval (CI): 0.623–0.860]. The combined model integrating these parameters yielded the best predictive performance (AUC: 0.809, 95% CI: 0.713–0.906).

CONCLUSION

DLSCT -derived quantitative parameters of peritumoral fat, particularly VMI and Z_eff values from both AP and VP, serve as independent predictors of TB grade in CRAC.

CLINICAL SIGNIFICANCE

DLSCT-derived quantitative parameters of peritumoral fat can serve as a complementary, noninvasive quantitative imaging biomarker for preoperatively predicting TB grade in CRAC.

Colorectal cancer (CRC) ranks as the third most common cancer worldwide and the second leading cause of cancer-related deaths.^{1, 2} Tumor budding (TB) is recognized as a significant prognostic factor in colorectal adenocarcinoma (CRAC) by the National Comprehensive Cancer Network and European Society for Medical Oncology guidelines.³ Defined by single cells or small clusters of up to four cells at the invasive tumor front, TB is a robust independent prognostic marker.⁴ It stratifies patients into risk categories for recurrence and mortality,⁵ making its preoperative identification crucial for treatment planning. Although histopathological biopsy remains the gold standard for TB grading, the technique is invasive and carries potential complications.⁴ Contrast-enhanced computed tomography (CECT) is a cornerstone in CRAC management, providing standardized protocols for initial staging and therapy monitoring.^{6, 7} However, CECT is often insufficient for the comprehensive preoperative prediction of CRC pathological types, necessitating the development of more advanced quantitative imaging techniques.⁸

Dual-layer spectral detector CT (DLSCT) is an advanced modality that generates material decomposition images, such as iodine- or water-based maps. With more quantitative parameters than CECT, DLSCT has proven useful in diagnosing and managing various tumors.^{6, 9-11} It demonstrates advantages in T-staging, evaluating tumor differentiation, perineural invasion (PNI), and lymphovascular invasion in CRC,^12-17 and has been applied to assess TB grade.¹⁷ Additionally, DLSCT-based radiomics and machine learning have been used to evaluate distant metastasis and microsatellite instability in CRC.^{18, 19}

In precision oncology, the peritumoral fat microenvironment (PFME) is increasingly recognized as a prognostic marker. Quantitative PFME analysis can predict histopathological markers and molecular subtypes. The mesenchymal transition of peritumoral adipose tissue is characterized by the generation of dedifferentiated adipocytes, which constitute the major stromal cell type within tumors.^20-22 A dynamic interaction exists between tumors and adipose tissue, where cancer cells induce metabolic reprogramming in adjacent adipocytes to promote progression.²³ CRC remodels the peritumoral fat through an initial phase of structural disruption followed by reorganization, ultimately leading to a brown fat-like phenotype.²⁴ These alterations directly modulate X-ray attenuation properties, which are quantified by DLSCT, providing a direct mechanistic link between spectral parameters of peritumoral fat and TB, a histopathological marker of local tumor invasiveness.¹⁷

However, the quantitative application of DLSCT for peritumoral fat assessment in CRAC TB grading has not been previously explored. Therefore, this study evaluates the utility of DLSCT quantitative parameters for the preoperative diagnosis of TB grade in patients with CRAC.

Methods

Patients

This study was approved by the Institutional Research Ethics Committee of the Third Affiliated Hospital of Sun Yat-Sen University (clinical trial number: II2025-229-01, date: 06.17.2025), and the requirement for informed consent was waived by the ethics committee due to the retrospective nature of the study. Between February 2024 and July 2025, 577 consecutive patients with pathologically confirmed CRAC who underwent preoperative spectral CT were initially enrolled. The inclusion criteria were (1) spectral CT performed within 2 weeks before surgery and (2) no prior treatment. The exclusion criteria included (1) preoperative radiotherapy or chemotherapy (n = 446), (2) CT-surgery interval exceeding 2 weeks (n = 20), (3) poor image quality due to artifacts (n = 8), and (4) mucinous or signet ring cell carcinoma (n = 13). The final cohort comprised 90 patients (24 with intermediate- to high-grade TB; 66 with low-grade TB). The entire analytical workflow of the study and the patient selection process are detailed in Figures 1 and 2, respectively.

Image acquisition

All scans were performed using a spectral CT scanner (IQon Spectral CT, Philips Healthcare, Best, The Netherlands). Patients were scanned in the supine position from the diaphragm to the pubic symphysis. The key parameters were as follows: tube voltage: 120 kV, automatic tube current modulation, pitch: 0.99, rotation time: 0.75 s, detector configuration: 64 × 0.625 mm, and slice thickness: 1.25 mm. The arterial phase (AP) was triggered at 150 Hounsfield units (HU) in the abdominal aorta, followed by a venous phase (VP) 60 seconds later.

Non-ionic contrast (iodixanol, Visipaque™, 300 mg I/mL) was administered intravenously at 1.2 mL/kg via a high-pressure syringe (3.0 mL/s), followed by a 30-mL saline flush (3.0 mL/s).

Image evaluation

Images were analyzed on an IntelliSpace Portal (Philips Healthcare) using Spectral CT Viewer software. Two radiologists (10 and 5 years of experience, respectively), blinded to the pathological results, adhered to the same standardized protocol and independently established regions of interest (ROIs). Intratumor ROIs were drawn on 70-keV virtual monoenergetic image (VMI) slices showing the largest tumor area, avoiding necrosis, vessels, and cysts. Peritumoral fat ROIs were established, with a width of ≤ 5 and a 1-mm margin from the tumor (mean: 5.5 mm², range: 2.0–12.5 mm²).²⁵

The quantitative parameters, including iodine concentration (IC) for both the tumor and the abdominal aorta or iliac vessels, effective atomic number (Z_eff), and VMIs at 40, 50, 60, and 70 keV, were generated. All measurements were performed independently by two blinded radiologists, and the mean values were used for final analysis. Normalized IC (NIC) was calculated to standardize measurements.

Pathological evaluation

Grading of TB was conducted using hematoxylin–eosin staining, assessed by two gastrointestinal pathologists (10 and 20 years’ experience, respectively) following International Tumor Budding Consensus Conference 2016 guidelines.²⁶ Disagreements were resolved by consensus. Buds were counted at 20 × magnification and normalized to buds/0.785 mm².²⁶ The grading was as follows: Bd1 (low, ≤ 4 buds), Bd2 (intermediate, 5–9 buds), and Bd3 (high, ≥ 10 buds). For the analysis, Bd1 was classified as low-grade TB, and Bd2 and Bd3 as intermediate- to high-grade TB.

Statistical analysis

All statistical analyses were performed using SPSS (version 26.0; IBM Corp., Chicago, IL, USA) and GraphPad Prism (version 9.5.1; GraphPad Software, San Diego, CA, USA). Data normality was assessed using the Shapiro–Wilk test. Normally distributed variables are presented as mean ± standard deviation, and non-normally distributed variables are presented as median and interquartile range. Group comparisons were conducted using the Mann–Whitney U test. Univariable and multivariable logistic regression analyses were performed to calculate odds ratios (ORs) with corresponding 95% confidence intervals (CIs). Predictive performance was evaluated using the area under the receiver operating characteristic curve (AUC), and AUCs were compared using the DeLong test. Multicollinearity was assessed using variance inflation factors (VIFs); variables with VIF > 10 were excluded. A backward stepwise elimination method (likelihood ratio) with a removal criterion of P > 0.05 was used to derive the final model. Interobserver agreement was assessed using the intraclass correlation coefficient (ICC), which was classified as poor (0–0.20), fair (0.21–0.40), moderate (0.41–0.60), good (0.61–0.80), or excellent (0.81–1.00). A two-sided P value < 0.05 was considered statistically significant.

Results

Demographic, clinical, and pathologic features

As shown in Table 1, 90 patients (53 men and 37 women; median age: 62 years) were enrolled in this study. All clinical characteristics (age, P = 0.51; gender, P = 0.22; location, P = 0.75; carcinoembryonic antigen, P = 0.51; carbohydrate antigen 199, P = 0.26) and qualitative parameters of CECT (cT stage, P = 0.82; cN stage, P = 0.51; c-extramural vascular invasion, P = 0.24) were not significantly different between low-grade TB and intermediate- to high-grade TB of CRAC (P > 0.05).

Among the spectral parameters, VMIs (40–70 keV) and Z_eff in both AP and VP, as well as NIC in the VP of peritumoral fat, were significantly different between low-grade TB and intermediate- to high-grade TB of CRAC (all P < 0.05; AP-VMI 40 keV, P = 0.001; AP-VMI 50 keV, P = 0.002; AP-VMI 60 keV, P = 0.001; AP-VMI 70 keV, P = 0.001; AP-Z_eff, P = 0.002; VP-VMI 40 keV, P = 0.004; VP-VMI 50 keV, P = 0.001; VP-VMI 60 keV, P = 0.001; VP-VMI 70 keV, P < 0.001; VP-NIC, P = 0.024; VP-Z_eff, P = 0.019) (Table 2). Examples of CRAC cases with low-grade TB and intermediate- to high-grade TB are shown in Figures 3 and 4.

Spectral computed tomography parameters, predictors, and logistic models for tumor budding prediction

As shown in Table 3, univariate analysis showed that VMIs at 40, 50, 60, and 70 keV, Z_eff in both AP and VP, and NIC in the VP of peritumoral fat were associated with the TB of CRAC (P < 0.05; AP-VMI 40 keV, P = 0.001; AP-VMI 50 keV, P = 0.002; AP-VMI 60 keV, P = 0.002; AP-VMI 70 keV, P = 0.001; AP-Z_eff, P = 0.002; VP-VMI 40 keV, P = 0.004; VP-VMI 50 keV, P = 0.001; VP-VMI 60 keV, P = 0.001; VP-VMI 70 keV, P < 0.001; VP-NIC, P = 0.024; VP-Z_eff, P = 0.019). In the multivariate analysis, VMI at 40, 50, 60, and 70 keV, and Z_eff in both AP and VP remained independent predictors of TB grade (OR: 1.025, 95% CI: 0.610–0.833, P = 0.001; OR: 0.987, 95% CI: 0.601–0.827, P = 0.002; OR: 0.845, 95% CI: 0.606–0.830, P = 0.002; OR: 1.166, 95% CI: 0.609–0.832, P = 0.002; OR: 8.649, 95% CI: 0.594–0.826, P = 0.002; OR: 0.910, 95% CI: 0.575–0.820, P = 0.004; OR: 1.091, 95% CI: 0.608–0.841, P = 0.001; OR: 1.069, 95% CI: 0.604–0.841, P = 0.001; OR: 1.015, 95% CI: 0.623–0.860, P = 0.001; OR: 0.433, 95% CI: 0.533–0.792, P = 0.019). The AUCs of the spectral CT parameters are presented in Table 4, ranging from 0.662 to 0.742. Among these, VP-VMI at 70 keV of peritumoral fat yielded the highest AUC of 0.742, with sensitivities of 66.67% and specificities of 81.82% for differentiating between low-grade and intermediate- to high-grade TB. Combining all spectral parameters and clinical characteristics further improved predictive ability, with an AUC of 0.809, a sensitivity of 70.80%, and a specificity of 81.80% (Figures 5 and 6, Table 4). Figure 7 presents the results from the final fitted multivariable logistic regression model.

Interobserver agreement

The interobserver agreement for CT images measured by two radiologists was excellent, with ICC values ranging from 0.657 (95% CI: 0.479–0.774) to 0.944 (95% CI: 0.914–0.963), except for the arterial enhancement fraction of peritumoral fat (Supplementary Table 1). This may be attributed to increased image noise at low keV or the challenge of consistently placing ROIs in heterogeneous peritumoral fat.

Discussion

This study identified DLSCT-derived quantitative parameters of peritumoral fat—specifically AP and VP VMI and Z_eff values—as independent predictors of TB grade in CRAC. This predictive utility is likely mediated by a mesenchymal transition of the peritumoral fat, characterized by marked metabolic rewiring (e.g., upregulation of UCP-1, TMEM26, and PON3, and downregulation of Pref-1 and adiponectin), which generates dedifferentiated adipocytes that form a key stromal component within tumors.^{20, 27}

Multiparametric DLSCT—including VMIs, IC, Z_eff, and the HU curve slope (λHU) —provides valuable insights for tumor assessment.^28-30 Iodine mapping quantitatively evaluates microcirculation via IC, offering clinical value in CRAC for staging, differentiation, PNI, and lymph node metastasis assessment.^{7, 12, 14, 31} Low-energy 40-keV VMIs provide a superior signal-to-noise ratio and contrast-to-noise ratio, improving tumor detection clarity.^{32, 33} This enhancement stems from the proximity of 40 keV to the iodine K-edge (33 keV), optimizing visualization of iodine-rich areas.^{33, 34} Tumor vascularization, derived from intrinsic angiogenesis and co-option of host vessels,³⁵ correlates with growth and metastatic potential.^{35, 36} The Z_eff quantifies material composition by representing equivalent elemental atomic numbers in X-ray attenuation^{37, 38} and can predict pathological subtypes at specific energies.³⁹

In our study, VMIs at 40, 50, 60, and 70 keV, Z_eff in both AP and VP, and NIC in the VP of peritumoral fat were significantly different between the low-grade TB and intermediate- to high-grade TB groups. This phenomenon likely results from improved visualization of contrast-enhanced lesions at lower VMI energy levels.^40-42 In our cohort, no significant correlations were found between tumor spectral parameters and TB grade, whereas peritumoral fat parameters showed strong associations. This spatial heterogeneity indicates distinct microenvironmental influences on spectral CT parameters. Our findings differ from those reported by Shao et al.,¹⁷ potentially due to divergent DLSCT protocols. Our AP scan was initiated 5 seconds after aortic enhancement reached 150 HU, potentially capturing a later, mixed AP/portal VP phase. In contrast, Shao et al. triggered scanning immediately upon detection of 100 HU. This fundamental difference in temporal triggering could result in divergent enhancement patterns of both tumor and peritumoral fat, affecting absolute parameter values and their correlation with TB. This likely accounts for the discrepancies in VP characterization observed between the studies.

In the present study, we found that some new parameters are associated with the TB grade in CRAC. Peritumoral fat spectral parameters were not reported in previous studies to predict TB grade in CRAC. In the logistic regression analysis, VMIs at 40, 50, 60, and 70 keV, Z_eff in both AP and VP, and NIC in the VP of peritumoral fat were independent predictors of the TB grade in CRAC. We propose that the TB-spectral CT correlation stems from a structural–functional cascade in peritumoral fat. The increase in both VMI attenuation and Z_eff can be attributed to several concurrent pathological processes in the peritumoral fat: fibrosis (with higher cellularity and connective tissue deposition), microvascular proliferation (increasing local blood volume), and inflammatory infiltration. These elements have intrinsically higher atomic numbers and density than pure adipose tissue. In parallel, a rise in NIC is a direct biomarker of increased iodine accumulation, signifying enhanced vascular perfusion, and potentially increased vascular permeability, likely mediated by tumor-derived angiogenic signaling. This architectural alteration directly alters X-ray absorption, producing measurable changes in spectral CT parameters, such as attenuation values and spectral λHU.⁴³

Consequently, a clinically relevant workflow may be seamlessly integrated into preoperative DLSCT staging, including performing quantitative peritumoral fat analysis, appending the results (e.g., high-grade TB probability) to the radiology report, and thereby informing the multidisciplinary team. This could impact surgical planning (e.g., margin consideration), neoadjuvant therapy stratification in rectal cancer, and postoperative prognostic stratification—especially crucial in stage II colon cancer, where TB is a high-risk feature—guiding follow-up intensity and adjuvant therapy.

This study has some limitations. First, its single-center, retrospective nature, coupled with its limited sample size, may have introduced selection bias. Moreover, the absence of internal validation (e.g., bootstrapping or cross-validation) for the logistic model necessitates future verification in larger, multicenter cohorts. Second, the study focused solely on adenocarcinomas, excluding other CRC histological variants, thereby limiting the applicability of our findings to all CRC subtypes. Third, subgroup distribution was uneven. Fourth, this study did not retrospectively collect body mass index or quantify visceral fat area. Finally, this study was restricted to a single CT vendor’s technology, and key parameters (VMI, IC, Z_eff) vary across different systems. Therefore, although the diagnostic concept is transferable, our specific thresholds are vendor sensitive, mandating multicenter studies across diverse platforms to derive generalizable standards. Future studies should also explore spectral CT’s role in evaluating tumor deposits, undertake multicenter investigations with larger sample sizes to confirm the robustness and reproducibility of these observations, and incorporate DLSCT spectral parameters into machine learning or radiomics models for automated, high-throughput analysis.

In conclusion, DLSCT-derived quantitative parameters of peritumoral fat, particularly VMI and Z_eff values from both AP and VP, serve as independent predictors of TB grade in CRAC.