Quantitative dual-energy computed tomography parameters as imaging biomarkers for assessing the histopathologic grade of colorectal cancer Authors: Tiantian Ma, Jianhua Liu, Meng Wu, Haijia Yu, Siwen Liu, Lihua Gao Published: 2026 ABSTRACT AbstractBackgroundColorectal cancer (CRC) represents a major global healthcare burden. Tumor differentiation grade is a critical prognostic factor for evaluating patients with CRC. The objective of this study was to examine the value of dual-energy computed tomography (DECT) quantitative parameters as imaging biomarkers for the preoperative prediction of histopathologic CRC grade.MethodsA total of 117 patients with pathologically confirmed CRC who underwent preoperative contrast-enhanced DECT were enrolled and divided into low-grade (n=74) and high-grade (n=43) groups. Monoenergetic 74-keV images from the arterial phase (AP), venous phase (VP), and delayed phase (DP) were reconstructed as conventional 120 kVp-like computed tomography (CT) images. Lesion CT attenuation values, iodine concentration (IC), normalized iodine concentration (NIC), effective atomic number (Zeff), and extracellular volume (ECV) were measured in three phases by two radiologists, and clinical data were collected. Intergroup comparisons and Spearman correlation analyses were conducted to determine the associations between spectral parameters and the histopathologic grade of CRC, and multivariable logistic regression analyses were applied to identify independent correlates. Three types of predictive models were established: a baseline model incorporating clinical variables and CT attenuation values, a single spectral parameter model, and combined models integrating clinical features with different spectral parameters. Receiver operating characteristic (ROC) curves and the DeLong test were used to compare the area under the curve (AUC) between the models. Bootstrap internal validation with 1,000 resamples was performed on the optimal model.ResultsSerum carcinoembryonic antigen (CEA) concentration showed significant differences between the low-grade group and high-grade group (P=0.009), whereas triphasic 120 kVp-like CT attenuation values showed no statistical differences (P>0.05). For the AP, only IC differed significantly (P=0.011); for the VP and DP, IC, NIC, Zeff, and ECV differed significantly (P<0.05), with ECV showing the strongest correlation with histopathologic grade in CRC (r=0.650: P<0.001). Multivariable analysis identified VP IC [odds ratio (OR) =4.601; P=0.046] and ECV (OR =1.492; P<0.001) as independent correlates of high histopathologic grade in CRC, with AUCs of 0.749 and 0.889, respectively. The integrated model combining clinical indicators with VP and DP spectral parameters achieved the best performance for predicting high histopathologic grade in CRC [AUC =0.904, 95% confidence interval (CI): 0.845–0.962; bias‑corrected AUC =0.904, 95% CI: 0.837–0.954], with an accuracy, sensitivity, and specificity of 85.5%, 88.4%, and 83.8%, respectively. The DeLong test revealed that this integrated model had significantly better diagnostic performance than did the baseline model (AUC =0.538; P<0.001), whereas no statistically significant difference was observed between the integrated model and the VP + DP dual-phase model (P=0.504).ConclusionsDECT quantitative parameters have potential clinical value for predicting the histopathologic grade of patients with CRC. They may serve as imaging biomarkers and provide valuable information for understanding tumor biological behavior.Keywords: Colorectal cancer (CRC), histopathologic grading, dual-energy computed tomography (DECT), spectral parameters, imaging biomarkers Introduction Despite advances being achieved in its treatment, colorectal cancer (CRC) remains a major global healthcare challenge, as it ranks as the third most common cancer and the second deadliest malignancy, representing a significant burden in terms of cancer morbidity and mortality worldwide (1,2). Although screening programs, molecular profiling, and targeted therapeutics have greatly optimized the clinical management of patients with CRC, preoperative histopathologic tumor grading is still critical for individualized prognostic stratification. The histopathologic classification of CRC is determined by the proportion of gland-forming tumor components. Tumors are classified as well-, moderately, or poorly differentiated lesions (3). Well- and moderately differentiated tumors share similar indolent biological behaviors and have lower proliferation and metastatic potential. Therefore, according to the World Health Organization (WHO) grading system, they are grouped into a unified low-grade category, while poorly differentiated cancers constitute the high-grade group (4). This latter category is characterized by prominent cellular atypia, active proliferation, and a poor clinical prognosis. Poorly differentiated cancers may exhibit higher malignant potential, represent an independent adverse prognostic factor for stage II colon cancer (5), and have a tendency to recur (6). Preoperative radiotherapy or chemoradiotherapy may improve survival in patients with this type of disease yet is associated with notable toxicity (7,8). Given the differences in prognostic outcomes between low- and high-grade tumors, accurate preoperative tumor grading is critical in clinical practice. Pathological grading of tumor differentiation largely depends on invasive biopsy, but this is prone to sampling bias and fails to fully capture intralesional tumor heterogeneity. Preoperative contrast-enhanced conventional computed tomography (CT) is routinely applied to examine patients with CRC, yet it lacks sufficient accuracy for evaluating tumor differentiation status. Through material decomposition, dual-energy CT (DECT) can provide quantitative data beyond conventional CT scanning, which may improve the detection of subtle tissue variations (9,10). By quantifying iodine density, generating virtual monoenergetic images, and mapping spectral decay curves, this technology has the potential to overcome some of the sampling limitations associated with invasive biopsies. Studies have found a close association between CRC differentiation grade and key histological features, including microvessel density (MVD), extracellular matrix (ECM) composition, and the extent of fibrosis (11-13). Extracellular volume (ECV) reflects the combined volume of the extravascular extracellular space and the intravascular compartment. Its measurement allows for the combined assessment of stromal fibrosis and MVD and thus the comprehensive characterization of the tumor microenvironment (14). Indeed, previous studies have linked ECV to CRC T stage (15) and neurovascular invasion (16). Normalized iodine concentration (NIC) also has the capacity to differentiate high-grade from low-grade CRC (17), and it has demonstrated performance comparable to that in differentiating CRC grades when predicting the grade of hepatocellular carcinoma (18) and renal clear cell carcinoma (19). These findings support the use of DECT-derived quantitative parameters as noninvasive imaging biomarkers. However, few studies have examined the application of DECT parameters for the preoperative grading of CRC. A recent comprehensive review of the research on spectral CT in CRC (20) indicated that the majority of relevant work in this field consists of small, single-center studies that lack a standardized methodology; moreover, there has been little research conducted on using quantitative DECT metrics to predict histopathologic grade. Therefore, we performed a study with the following aims: (I) to determine whether four core DECT parameters, including iodine concentration (IC), NIC, effective atomic number (Zeff), and ECV, can serve as noninvasive imaging biomarkers for the preoperative assessment of CRC histopathologic grade and (II) to further develop multiparameter combined models to improve preoperative grading performance. We present this article in accordance with the TRIPOD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1-2641/rc). Methods Patients Patients who received surgery with a subsequent pathological confirmation of CRC from September 2024 to July 2025 at the Second Hospital of Jilin University were enrolled in the study. All included patients underwent contrast-enhanced DECT of the abdomen before surgery. The following inclusion criteria were used to retrospectively select eligible patients from our medical records: (I) histopathologically confirmed CRC with a definite differentiation grade, (II) DECT scanning performed within 2 weeks prior to surgery, and (III) hematocrit (HCT) measurement obtained within 3 days before or after the DECT examination. Meanwhile, the following exclusion criteria were applied: (I) lesions not visible on CT images, (II) history of iodinated contrast media allergy or other contraindications to contrast-enhanced imaging, (III) images of unsatisfactory quality for postprocessing analysis due to severe artifacts, (IV) incomplete clinical data, and (V) anticancer treatment administered before the DECT examination. Patients were stratified based on their postoperative pathological results. Given the comparable biological behaviors between moderately and well-differentiated tumors, these two subtypes were consolidated into a low-grade group. Meanwhile, poorly differentiated tumors constituted the high-grade group. Figure 1 presents a flowchart of the patient enrollment process. Flowchart of participant selection based on eligibility criteria. CRC, colorectal cancer; DECT, dual-energy computed tomography. This retrospective study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments and was approved by the Institutional Review Board of the Second Hospital of Jilin University. The requirement for informed consent was waived, as all patient data were fully anonymized, no extra interventions or patient contacts were conducted, and the study posed no physical or psychological risks to participants. All clinical and DECT imaging data were retrospectively retrieved from our hospital’s archives. Clinical features Clinical data were extracted from medical records, including demographic characteristics, serum tumor biomarkers, HCT, and colonoscopic pathological findings. Demographic data included age, sex, height, weight, and body mass index (BMI). Serum tumor biomarkers comprised carcinoembryonic antigen (CEA) and carbohydrate antigen 19-9 (CA19-9). Postoperative pathological reports of the resected specimens provided data regarding tumor location, vascular invasion status, and histological grade. CT examination protocol Contrast-enhanced abdominal scans were acquired in three phases. All images were acquired with a spectral CT system (SOMATOM Force, Siemens Healthineers, Erlangen, Germany). Patients were asked to do the following: (I) abstain from taking heavy metal-containing medications for 1 week prior to scanning, (II) avoid gastrointestinal contrast examinations within the preceding week, (III) fast for 4–6 hours before the scan, and (IV) remove all metallic objects or interfering materials from the scan field. The scan range was established according to anatomical landmarks, commencing at the diaphragmatic dome and concluding at the pubic symphysis. Arterial phase (AP) scanning was initiated 10 seconds after a trigger threshold of 100 Hounsfield units (HU) was reached. This protocol was preceded by the intravenous administration of 75 mL of contrast medium (injection rate 4.0 mL/s) and a 40-mL saline flush, both delivered by an automated injector. Venous phase (VP) and delayed phase (DP) images were subsequently acquired at 60 and 120 seconds after contrast injection, respectively. The key scanning parameters were configured as follows: tube voltage was set within a range of 100–150 kV, while tube currents for X-ray tube A and X-ray tube B were 240 mAs and 120 mAs, respectively. Images were reconstructed with a 512×512 matrix and subsequently analyzed on a dedicated workstation. Image analysis and data measurement A multiphasic DECT scan was performed to obtain quantitative measurements of IC, NIC, Zeff, and ECV. Regions of interest (ROIs) were manually drawn to cover the maximum cross-sectional area of the tumor. During ROI delineation, necrotic areas without enhancement on all CT phases were strictly excluded. All quantitative parameters were independently measured by two radiologists with more than 5 years of experience in abdominal imaging who were blinded to the pathological results. The average values of the measurements from the two observers were calculated and used for subsequent statistical analysis. Interobserver measurement consistency was assessed via the intraclass correlation coefficient (ICC) and Bland-Altman analysis, with the latter providing the mean bias and 95% limits of agreement. The copy and paste function of the workstation was used to ensure that the shape and size of the ROI were relatively consistent across all three phases. IC and Zeff values were recorded within each tumor ROI. For signal normalization, the IC within the abdominal aorta or ipsilateral iliac artery at the same anatomical level was also measured. The NIC was calculated as follows: NIC = (IC lesion/IC aorta/iliac artery) (1). Meanwhile, the ECV was calculated according to the following formula (with the NIC measured in the DP): ECV (%) = (1 – HCT) × NIC × 100% (2). In spectral monoenergetic imaging, CT attenuation values decrease continuously with increasing kiloelectron volt (keV) levels. Previous studies (21,22) have shown that 74-keV monoenergetic images yield CT values comparable to those from conventional 120-kVp polychromatic CT images. Therefore, this study used 74-keV monoenergetic images to represent conventional 120-kVp CT and to compare the diagnostic efficacy between DECT and conventional CT in differentiating the pathological grades of CRC. Statistical analysis Statistical analyses were performed with SPSS software version 27.0 (IBM Corp., Armonk, NY, USA). Continuous variables with a normal distribution are presented as the mean ± standard deviation (SD), whereas nonnormally distributed data are reported as the median and interquartile range (IQR). Categorical variables are summarized as frequencies and percentages. Between-group comparisons were performed via the independent-samples t test for normally distributed continuous variables, the Mann-Whitney U test for nonnormally distributed continuous data, and the chi-squared or Fisher exact test for categorical variables. A two-sided P value <0.05 was considered statistically significant for all analyses. Spearman correlation analysis was used to evaluate the associations between spectral parameters (IC, NIC, Zeff, and ECV) and the pathological grade of CRC. The strength of correlation was classified as negligible (|r| =0.00–0.19), weak (|r| =0.20–0.39), moderate (|r| =0.40–0.69), strong (|r| =0.70–0.89), or very strong (|r| =0.90–1.00). Variables with P<0.05 in the univariable analysis were included in a multivariable logistic regression model to adjust for potential confounders and identify independent correlates of CRC histopathologic grade. The diagnostic performance of all single parameters and combined models was assessed via receiver operating characteristic (ROC) curve analysis, with pairwise area under the curve (AUC) comparisons performed via the DeLong test. To account for potential optimistic bias attributed to the relatively limited sample size, we performed bootstrap-based internal validation for the final optimal prediction model. Briefly, 1,000 resamples were randomly generated from the original cohort with replacement. An identical prediction model was refitted in each bootstrap replicate, and the corresponding AUC was calculated to establish an empirical sampling distribution. Based on this distribution, the bias-corrected AUC and its 95% CI were estimated via the bootstrap percentile method. The SD and SE of the bootstrap-generated AUC values were also computed. Additionally, a series of diagnostic metrics were calculated, including accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (LR+), and negative likelihood ratio (LR–). The Youden index was applied to full study cohort to determine the optimal cutoff value for each parameter and model. Results Patients The study enrolled 117 patients, with ages ranging from 33 to 85 years (mean ± SD 65.0±11.1 years). All participants were stratified into two groups based on the WHO histological classification criteria: a low-grade group (n=74) and a high-grade group (n=43). Among the 43 patients in the high-grade group, the histological subtypes were distributed as follows: poorly differentiated adenocarcinoma, 86.0% (n=37); mucinous adenocarcinoma, 11.6% (n=5); and signet-ring cell carcinoma, 2.3% (n=1). Clinical characteristics Serum CEA concentration differed significantly between the low-grade group (median 3.45 ng/mL, IQR, 2.39–13.69 ng/mL) and high-grade group (median 9.31 ng/mL, IQR, 3.82–36.13 ng/mL) (P=0.009); however, there were no significant differences in age (P=0.568), gender (P=0.862), height (P=0.438), weight (P=0.922), BMI (P=0.707), CA19-9 level (P=0.611), HCT (P=0.207), tumor location (P=0.154), or vascular invasion status (P=0.108). The distribution of clinical characteristics across the study groups is summarized in Table 1. Table 1Comparison of clinical characteristics between the low- and high-grade groupsCharacteristicsLow-grade group (n=74; 63.2%)High-grade group (n=43; 36.8%)t/Z/X2P valueAge (years)66 (57, 72) [33–83]67 (58, 74) [35–85]−0.5710.568Sex0.0300.862 Female27 (36.5)15 (34.9) Male47 (63.5)28 (65.1)Height (m)1.68 (1.58, 1.72) [1.46–1.80]1.69 (1.65, 1.70) [1.53–1.85]−0.7750.438Weight (kg)66.92±12.84 [37.00–101.00]66.70±9.60 [47.50–90.00]−0.0980.922BMI (kg/m²)24.25±3.64 [15.01–32.98]23.99±3.40 [16.44–33.99]−0.3770.707CEA (ng/mL)3.45 (2.39, 13.69) [0.93–1,500]9.31 (3.82, 36.13) [1.44–1,500]−2.6290.009CA19-9 (U/mL)13.51 (6.57, 33.11) [2.00–1,200]12.37 (4.90, 48.88) [2.00–1,200]−0.5090.611Hematocrit (%)38.97±6.73 [22.60–53.40]37.40±5.88 [22.70–47.40]−1.2700.207Tumor location2.0300.154 Right hemicolon23 (31.1)19 (44.2) Left hemicolon51 (68.9)24 (55.8)Vascular invasion2.5830.108 Negative49 (66.2)22 (51.2) Positive25 (33.8)21 (48.8)Data are presented as n (%), median (interquartile range) [range] or mean ± standard deviation [range]. BMI, body mass index; CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen. Comparison of clinical characteristics between the low- and high-grade groups Data are presented as n (%), median (interquartile range) [range] or mean ± standard deviation [range]. BMI, body mass index; CA19-9, carbohydrate antigen 19-9; CEA, carcinoembryonic antigen. CT120 kVp-like attenuation values in the AP, VP and DP Comparisons between the high-grade and low-grade CRC groups across the AP, VP, and DP revealed no statistically significant differences in ACT120 kVp-like, VCT120 kVp-like, or DCT120 kVp-like values (P>0.05). Detailed results are presented in Table 2. Table 2Comparison of 120 kVp-like CT values between the low- and high-grade groupsAttenuation values (HU)Low-grade group (n=74; 63.2%)High-grade group (n=43; 36.8%)t/ZP valueACT120 kVp-like79.22±15.06 [38.88–112.13]79.80±14.33 [42.72–105.59]0.2060.837VCT120 kVp-like73.16 (68.09, 80.41) [45.21–93.10]75.89 (66.66, 82.34) [45.39–90.03]−0.5090.611DCT120 kVp-like69.50±9.37 [40.69–87.65]70.88±10.88 [43.86–92.36]0.7230.471Data are presented as median (interquartile range) [range] or mean ± standard deviation [range]. ACT120 kVp-like, arterial-phase computed tomography attenuation values measured on 74-keV monoenergetic (120 kVp-like) images; CT, computed tomography; DCT120 kVp-like, delayed-phase computed tomography attenuation values measured on 74-keV monoenergetic (120 kVp-like) images; HU, Hounsfield unit; VCT120 kVp-like, venous-phase computed tomography attenuation values measured on 74-keV monoenergetic (120 kVp-like) images. Comparison of 120 kVp-like CT values between the low- and high-grade groups Data are presented as median (interquartile range) [range] or mean ± standard deviation [range]. ACT120 kVp-like, arterial-phase computed tomography attenuation values measured on 74-keV monoenergetic (120 kVp-like) images; CT, computed tomography; DCT120 kVp-like, delayed-phase computed tomography attenuation values measured on 74-keV monoenergetic (120 kVp-like) images; HU, Hounsfield unit; VCT120 kVp-like, venous-phase computed tomography attenuation values measured on 74-keV monoenergetic (120 kVp-like) images. Comparison of quantitative DECT parameters The interobserver agreement for all DECT parameters across the AP, VP, and DP was considered excellent, with ICC values ranging from 0.964 to 0.997. The Bland-Altman plots presented in Figure 2 show that nearly all data points fell within the 95% limits of agreement. Bland-Altman plots for interobserver agreement of all DECT quantitative parameters (IC, NIC, and Zeff) across the AP, VP, and DP. Most points lie within the 95% limits of agreement, indicating good reproducibility. AP, arterial phase; DECT, dual‑energy computed tomography; DP, delayed phase; IC, iodine concentration; NIC, normalized iodine concentration; SD, standard deviation; VP, venous phase; Zeff, effective atomic number. Statistical analysis revealed significant intergroup differences for IC between the low‑ and high‑grade groups during the AP, with higher values observed in the high-grade group (2.25±0.54 mg/mL, range, 1.21–3.41 mg/mL) than in the low-grade group (1.96±0.62 mg/mL, range, 0.47–3.72 mg/mL) (P=0.011). All quantitative parameters were significantly lower in the low-grade group than in the high-grade group during the VP, including IC (low-grade: median 1.62 mg/mL, IQR, 1.37–2.04 mg/mL, range, 0.79–3.33 mg/mL; high-grade: median 2.00 mg/mL, IQR, 1.80–2.23 mg/mL, range, 1.31–3.10 mg/mL; P<0.001), NIC (low-grade: median 0.35, IQR, 0.30–0.42, range, 0.20–0.66; high-grade: median 0.43, IQR, 0.37–0.49, range, 0.25–0.86; P<0.001), and Zeff (low-grade: 8.42±0.21, range, 7.98–8.86; high-grade: 8.51±0.23, range, 7.92–8.87; P=0.029). All quantitative parameters demonstrated significant intergroup differences during the DP, with lower values observed in the low-grade group than in the high-grade group, including IC (low-grade: median 1.42 mg/mL, IQR, 1.21–1.62 mg/mL, range, 0.78–2.17 mg/mL; high-grade: median 1.74 mg/mL, IQR, 1.55–1.90 mg/mL, range, 1.11–3.01 mg/mL; P<0.001), NIC (low-grade: median 0.47, IQR, 0.39–0.52, range, 0.30–0.82; high-grade: median 0.61, IQR, 0.52–0.70, range, 0.40–0.87; P<0.001), and Zeff (low-grade: 8.36±0.17, range, 7.95–8.70; high-grade: 8.45±0.20, range, 8.09–9.08; P=0.012). A statistically significant intergroup difference was identified for ECV, with higher values in the high-grade group (38.49±6.54 %; range 23.75–52.03 %) than in the low-grade group (28.07±5.34 %; range 17.17–43.18 %) (P<0.001). Comparisons of these parameters are presented in Table 3 and Figure 3, and representative CT images of different histopathologic grades are shown in Figures 4,5. Table 3Comparison of DECT parameters between the low- and high-grade groupsParametersLow-grade group (n=74; 63.2%)High-grade group (n=43; 36.8%)t/ZP valueAP IC (mg/mL)1.96±0.62 [0.47–3.72]2.25±0.54 [1.21–3.41]2.5860.011AP NIC0.14 (0.11, 0.18) [0.04–0.49]0.16 (0.13, 0.20) [0.08–0.45]−1.9450.052AP Zeff8.58±0.31 [7.70–9.19]8.63±0.26 [7.99–9.19]0.7970.427VP IC (mg/mL)1.62 (1.37, 2.04) [0.79–3.33]2.00 (1.80, 2.23) [1.31–3.10]−4.483<0.001VP NIC0.35 (0.30, 0.42) [0.20–0.66]0.43 (0.37, 0.49) [0.25–0.86]−3.929<0.001VP Zeff8.42±0.21 [7.98–8.86]8.51±0.23 [7.92–8.87]2.2160.029DP IC (mg/mL)1.42 (1.21, 1.62) [0.78–2.17]1.74 (1.55, 1.90) [1.11–3.01]−4.449<0.001DP NIC0.47 (0.39, 0.52) [0.30–0.82]0.61 (0.52, 0.70) [0.40–0.87]−6.162<0.001DP Zeff8.36±0.17 [7.95–8.70]8.45±0.20 [8.09–9.08]−2.5040.012ECV (%)28.07±5.34 [17.17–43.18]38.49±6.54 [23.75–52.03]9.358<0.001Data are presented as median (interquartile range) [range] or mean ± standard deviation [range]. AP, arterial phase; DECT, dual-energy computed tomography; DP, delayed phase; ECV, extracellular volume; IC, iodine concentration; NIC, normalized iodine concentration; VP, venous phase; Zeff, effective atomic number. Comparison of DECT parameters between the low- and high-grade groups Data are presented as median (interquartile range) [range] or mean ± standard deviation [range]. AP, arterial phase; DECT, dual-energy computed tomography; DP, delayed phase; ECV, extracellular volume; IC, iodine concentration; NIC, normalized iodine concentration; VP, venous phase; Zeff, effective atomic number. Comparison of spectral parameters between the low- and high-grade CRC groups. (A) IC values in the AP, VP, and DP. (B) NIC values in the AP, VP, and DP. (C) Zeff values in the AP, VP, and DP. (D) ECV measurements. Horizontal lines indicate error bars. *, P<0.05; ***, P<0.001; AP, arterial phase; CRC, colorectal cancer; DP, delayed phase; ECV, extracellular volume; IC, iodine concentration; NIC, normalized iodine concentration; ns, not significant; VP, venous phase; Zeff, effective atomic number. Representative case of a 64-year-old female with pathologically confirmed high-grade rectal adenocarcinoma (J,K). White circles denote ROIs (ECV =38.8%). (A) Arterial phase. (B) AP IC =3.20 mg/mL and NIC =22%. (C) AP Zeff =8.70. (D) Venous phase. (E) VP IC =2.00 mg/mL and NIC =40%. (F) VP Zeff =8.35. (G) Delayed phase. (H) DP IC =2.20 mg/mL and NIC =61%. (I) DP Zeff =8.45. AP, arterial phase; DP, delayed phase; ECV, extracellular volume; IC, iodine concentration; NIC, normalized iodine concentration; ROI, region of interest; VP, venous phase; Zeff, effective atomic number. Representative case of a 60-year-old male with pathologically confirmed low-grade adenocarcinoma of the ascending colon (J,K). White circles denote ROIs (ECV =32.27%). (A) Arterial phase. (B) AP IC =1.73 mg/mL and NIC =12%. (C) AP Zeff =8.66. (D) Venous phase. (E) VP IC =1.50 mg/mL and NIC =38%. (F) VP Zeff =8.33. (G) Delayed phase. (H) DP IC =1.63 mg/mL and NIC =61%. (I) DP Zeff =8.35. AP, arterial phase; DP, delayed phase; ECV, extracellular volume; IC, iodine concentration; NIC, normalized iodine concentration; ROI, region of interest; VP, venous phase; Zeff, effective atomic number. Spearman correlation analysis (Table 4) indicated significant correlations between histological grade of CRC and most spectral parameters, with ECV exhibiting the strongest association (r=0.650; P<0.001), followed by NIC in the DP (r=0.572) and IC in the VP (r=0.416). Neither NIC nor Zeff in the AP showed a significant correlation with histological grade (P>0.05). Table 4Correlations between DECT parameters and histopathologic gradeParametersLow-grade group (n=74; 63.2%)High-grade group (n=43; 36.8%)r valueP valueArterial phase IC (mg/mL)1.96±0.62 [0.47–3.72]2.25±0.54 [1.21–3.41]0.2450.008 NIC0.14 (0.11, 0.18) [0.04–0.49]0.16 (0.13, 0.20) [0.08–0.45]0.1810.051 Zeff8.58±0.31 [7.70–9.19]8.63±0.26 [7.99–9.19]0.0710.444Venous phase IC (mg/mL)1.62 (1.37, 2.04) [0.79–3.33]2.00 (1.80, 2.23) [1.31–3.10]0.416<0.001 NIC0.35 (0.30, 0.42) [0.20–0.66]0.43 (0.37, 0.49) [0.25–0.86]0.365<0.001 Zeff8.42±0.21 [7.98–8.86]8.51±0.23 [7.92–8.87]0.2090.024Delayed phase IC (mg/mL)1.42 (1.21, 1.62) [0.78–2.17]1.74 (1.55, 1.90) [1.11–3.01]0.413<0.001 NIC0.47 (0.39, 0.52) [0.30–0.82]0.61 (0.52, 0.70) [0.40–0.87]0.572<0.001 Zeff8.36±0.17 [7.95–8.70]8.45±0.20 [8.09–9.08]0.2330.012 ECV (%)28.07±5.34 [17.17–43.18]38.49±6.54 [23.75–52.03]0.650<0.001Data are presented as median (interquartile range) [range] or mean ± standard deviation [range]. DECT, dual-energy computed tomography; ECV, extracellular volume; IC, iodine concentration; NIC, normalized iodine concentration; Zeff, effective atomic number. Correlations between DECT parameters and histopathologic grade Data are presented as median (interquartile range) [range] or mean ± standard deviation [range]. DECT, dual-energy computed tomography; ECV, extracellular volume; IC, iodine concentration; NIC, normalized iodine concentration; Zeff, effective atomic number. Evaluation and selection of quantitative DECT parameters To eliminate confounding factors, a multivariable logistic regression was performed, adjusting for all variables that showed significance in the univariable analysis. Multivariable analysis identified the independent correlates of histopathologic grade to be VP IC [odds ratio (OR) 4.601; 95% CI: 1.029–20.580; P=0.046] and ECV (OR 1.492; 95% CI: 1.195–1.862; P<0.001), while the other parameters were excluded from the final model as nonsignificant confounders (Table 5). Table 5Independent correlates of histopathologic CRC gradeVariablesbStandard errorWaldOR (95% CI)P valueVP IC (mg/mL)1.5260.7643.9884.601 (1.029–20.580)0.046ECV (%)0.4000.11312.4981.492 (1.195–1.862)<0.001CI, confidence interval; CRC, colorectal cancer; ECV, extracellular volume; IC, iodine concentration; OR, odds ratio; VP, venous phase. Independent correlates of histopathologic CRC grade CI, confidence interval; CRC, colorectal cancer; ECV, extracellular volume; IC, iodine concentration; OR, odds ratio; VP, venous phase. Diagnostic efficacy of single parameters and multiparameter models This study initially evaluated the diagnostic performance of clinical indicators and DECT parameters in predicting the histopathologic grade of CRC (Table 6 and Figure 6A). Among the individual parameters, ECV demonstrated a relatively higher AUC (0.889; 95% CI: 0.826–0.952), followed by VP IC (0.749; 95% CI: 0.662–0.837), whereas serum CEA showed a comparatively lower AUC (0.646; 95% CI: 0.545–0.747). Table 6Diagnostic efficacy of clinical and DECT parametersVariablesAUC (95% CI)Youden indexCutoff valueSensitivity (%)Specificity (%)Clinical characteristic CEA (ng/mL)0.646 (0.545–0.747)0.3183.67579.152.7DECT parameters VP IC (mg/mL)0.749 (0.662–0.837)0.5111.6793.058.1 ECV (%)0.889 (0.826–0.952)0.62636.0172.190.5AUC, area under the curve; CEA, carcinoembryonic antigen; CI, confidence interval; DECT, dual-energy computed tomography; ECV, extracellular volume; IC, iodine concentration; VP, venous phase. Diagnostic efficacy of clinical and DECT parameters AUC, area under the curve; CEA, carcinoembryonic antigen; CI, confidence interval; DECT, dual-energy computed tomography; ECV, extracellular volume; IC, iodine concentration; VP, venous phase. ROC curves for the prediction of CRC histopathologic grade. (A) Single-parameter ROC curves. (B) Multiparameter combined model ROC curves. (C) Bootstrap internal validation (1,000 resamples) of the optimal combined model (Clinical + VP + DP), with the mean ROC curve shown as a solid blue line and the individual iteration curves in light gray. “Clinical” refers to serum CEA. The combined models were constructed as follows: Clinical + HU = a baseline model incorporating clinical variables and triphasic 120 kVp-like CT attenuation values; VP + DP = VP IC + ECV; Clinical + VP = CEA + VP IC; Clinical + DP = CEA + ECV; Clinical + VP + DP = CEA + VP IC + ECV. AUC, area under the curve; CEA, carcinoembryonic antigen; CI, confidence interval; CRC, colorectal cancer; CT, computed tomography; DP, delayed phase; ECV, extracellular volume; IC, iodine concentration; ROC, receiver operating characteristic; VP, venous phase. We subsequently compared the performance of single parameters and the combined multiparameter models (Table 7 and Figure 6B). It was found that the integrated models outperformed the single parameters. Specifically, the combined model incorporating clinical data and dual‑phase spectral parameters from the VP and DP (Clinical + VP + DP) achieved the best overall performance among all tested combinations, with an AUC of 0.904 (95% CI: 0.845–0.962), an accuracy of 85.5%, a sensitivity of 88.4%, and a specificity of 83.8%. Bootstrap internal validation (1,000 resamples; Figure 6C) yielded a bias‑corrected AUC of 0.904 (SD =0.03; SE =0.001; 95% CI: 0.837–0.954), and the Hosmer-Lemeshow test indicated acceptable model fit (χ²=7.4; P=0.494). Table 7Diagnostic performances of the combined parametersParametersVP + DPClinical + VPClinical + DPClinical + VP + DPAUC (95% CI)0.902 (0.843–0.961)0.750 (0.662–0.838)0.891 (0.828–0.953)0.904 (0.837–0.954)Cutoff value0.2980.2850.5330.299Accuracy (%)84.671.084.685.5Sensitivity (%)88.493.072.188.4Specificity (%)82.458.191.983.8PPV (%)74.556.383.876.0NPV (%)92.493.585.092.5LR+5.022.228.905.46LR−0.140.120.300.14“Clinical” refers to serum CEA. The combined models were constructed as follows: VP + DP = VP IC + ECV; Clinical + VP = CEA + VP IC; Clinical + DP = CEA + ECV; Clinical + VP + DP = CEA + VP IC + ECV. AUC, area under the curve; CEA, carcinoembryonic antigen; CI, confidence interval; DP, delayed phase; IC, iodine concentration; ECV, extracellular volume; LR+, positive likelihood ratio; LR−, negative likelihood ratio; NPV, negative predictive value; PPV, positive predictive value; VP, venous phase. Diagnostic performances of the combined parameters “Clinical” refers to serum CEA. The combined models were constructed as follows: VP + DP = VP IC + ECV; Clinical + VP = CEA + VP IC; Clinical + DP = CEA + ECV; Clinical + VP + DP = CEA + VP IC + ECV. AUC, area under the curve; CEA, carcinoembryonic antigen; CI, confidence interval; DP, delayed phase; IC, iodine concentration; ECV, extracellular volume; LR+, positive likelihood ratio; LR−, negative likelihood ratio; NPV, negative predictive value; PPV, positive predictive value; VP, venous phase. Pairwise AUC comparisons with the DeLong test further revealed that the AUC of VP IC was significantly lower than that of ECV (P=0.006), the VP + DP dual‑phase model (P<0.001), the three‑variable combined model (P<0.001), and the DP-CEA combined model (P=0.005). In contrast, no statistically significant differences were observed between VP IC and either CEA alone or the VP-CEA combined model (P>0.05). Of note, the VP + DP dual‑phase model and the three‑variable combined model exhibited no statistically significant difference (P=0.504), suggesting that adding serum CEA to the dual‑phase imaging model did not provide clear additional diagnostic benefit. To preliminarily investigate whether spectral parameters provide incremental diagnostic value relative to conventional CT attenuation values, we constructed a baseline model by integrating serum CEA and triphasic 120 kVp-like CT attenuation values (ACT120 kVp-like, VCT120 kVp-like, and DCT120 kVp-like). Although these conventional CT parameters showed no intergroup differences in the univariable analysis, they were included in the baseline model to serve as a reference for verifying the incremental value of spectral parameters based on conventional CT metrics. We further compared the diagnostic performance of this baseline model with that of the combined model incorporating clinical variables and spectral parameters. The results indicated that the combined model achieved a significantly higher AUC than did the baseline model (0.904 vs. 0.538; ΔAUC =0.366; P<0.001). Discussion Our results suggest that integrating clinical features with DECT-derived quantitative parameters may facilitate the preoperative evaluation of CRC histopathologic grade. Univariable analysis revealed that serum CEA and multiple DECT parameters—particularly those acquired in the VP and DP—were significantly associated with tumor grade, with higher values being observed in poorly differentiated tumors, consistent with previous reports (23,24). The combined model incorporating clinical data and dual-phase (VP + DP) spectral parameters yielded the optimal predictive performance (AUC =0.904), markedly outperforming the baseline model combining CEA with triphasic conventional CT attenuation values (AUC =0.538; ΔAUC =0.366; P<0.001). This finding indicates that spectral parameters capture additional physiological information beyond CT attenuation measurements: HU values reflect only lesion density, whereas measuring the IC enables precise quantification of tissue perfusion and vascular supply. This difference may explain the superior discriminatory performance of spectral parameters in predicting tumors’ histopathologic grade. Although preoperative CEA level is a recognized prognostic marker for resectable rectal cancer (25), its capacity to differentiate CRC grade was limited in our cohort (AUC =0.646), and adding CEA to the imaging‑based model provided no incremental diagnostic benefit. This result may be explained by CEA’s limited sensitivity in early‑stage or poorly differentiated CRC, and its specificity is often compromised by nonneoplastic confounders such as chronic inflammation, smoking, and organ dysfunction. Within the tumor microenvironment, tumor progression and invasion critically depend on angiogenesis. MVD, defined as the microvessel count in regions with the highest vascular density, is a key indicator of angiogenic capacity (26). In CRC, higher-grade tumors exhibit rapid angiogenesis, elevated MVD, and hypervascularity, which together promote tumor invasiveness (27). Pigment epithelium-derived factor and vascular endothelial growth factor are vital in the colorectal adenoma-carcinoma sequence, and they show inverse and positive correlations, respectively, with tumor differentiation grade, with both promoting MVD elevation (28). Moreover, ECM remodeling is critical for tumor progression, as the ECM modulates cell proliferation and differentiation (29). Matrix stiffening further facilitates tumor invasion and metastasis (30), ultimately resulting in poorer patient prognosis (31). ECV may act as a composite biomarker reflecting both MVD and stromal fibrosis, potentially offering a comprehensive characterization of the tumor microenvironment. Previous research has demonstrated strong agreement between imaging-derived ECV and histologically quantified ECM from biopsy specimens (32), and this fact may facilitate the noninvasive assessment of intralesional ECM. In our cohort, patients with high-grade CRC tended to exhibit elevated ECV, which was positively correlated with histopathologic grade. This finding suggests that ECV could assist in stratifying CRC according to tumor grade. One mechanism underlying this relationship is the tendency of poorly differentiated tumors to develop extensive yet immature neovasculature featuring fragile, thin-walled vessels. Such vascular alterations may increase blood flow and vascular permeability, possibly promoting contrast extravasation into the interstitium and contributing to higher ECV measurements. Furthermore, the different grades of CRC exhibit distinct cytological atypia, architectural patterns, and tissue organization (33). When a tumor has high cellular density and abundant ECM protein secretion, this results in altered contrast agent kinetics, particularly prolonged retention within the extracellular space. This pathophysiological mechanism improves the ability of DP imaging to identify high-grade tumors. During tumorigenesis, the accumulation of genetic alterations drives uncontrolled cellular proliferation, resulting in increased cellular density and a marked reduction in the proportion of normal glandular structures (34). Consequently, malignancies of different pathological grades lead to variations in ECV values by altering cellular composition within the tumor microenvironment and inducing ECM remodeling. Interestingly, while the DP is theoretically optimal for capturing contrast agent retention in the extracellular space, only ECV emerged as an independent correlate in our study, whereas other parameters during DP, such as IC and NIC, did not. This can be attributed to the fact that ECV integrates information regarding both tissue perfusion and extracellular compartment volume, rendering it more sensitive to the stromal alterations typical of high-grade tumors. In contrast, the simple IC in the DP is influenced by both vascular and extravascular components, and its signal may be diluted by residual intravascular contrast or affected by individual variations in clearance kinetics (35). The VP, on the other hand, reflects a balance between arterial supply and early interstitial extravasation, and VP IC may capture the peak of tumor microvascular perfusion before significant contrast washout occurs. Therefore, the lack of an independent association for other DP parameters does not contradict the proposed mechanism; rather, it highlights that ECV may be a more specific biomarker for ECM remodeling than raw IC (36,37). Additionally, we identified VP IC as a significant independent correlate of histopathologic grade and thus may be a quantitative imaging biomarker. Patients with high-grade CRC exhibited elevated VP IC values compared to those with low-grade disease, a finding consistent with the study by Chen et al. (38). Since iodine-based agents serve as the primary contrast medium for tumor vasculature during contrast-enhanced scans, IC values derived from spectral CT imaging directly quantify the degree of lesion blood supply. As tumors become less differentiated, they typically develop a more abundant blood supply and greater neovascularization, resulting in elevated IC values on DECT. In principle, elevated IC values correspond to greater perfusion within the lesion (39). Hemodynamic patterns differ distinctly by phase: AP enhancement is driven by large-vessel supply, whereas VP enhancement arises from a combination of capillary perfusion within the tumor microcirculation and subsequent extravasation into the interstitial space. Previous studies have suggested that VP spectral parameter provide good diagnostic efficacy for predicting lymph node metastasis in patients with CRC (40). However, the diagnostic utility of IC measurement is limited, as it neither fully captures the hemodynamic profile and iodine uptake within tumor tissue nor remains robust against variations in scanning protocols and interpatient physiological differences. The utilization of the NIC parameter partially mitigates these technical and physiological confounders, thereby improving comparability between different groups. However, NIC was not identified as an independent correlate of histopathologic grade in our study. Possible explanations for this discrepancy include variations in CT scanning protocols, patient characteristics, and inconsistent ROI delineation. The focus in future studies will therefore be the optimization of experimental design by addressing these variables. Moreover, consistent with previous work (23,24), our statistical analysis identified no significant intergroup differences for certain clinical characteristics, including HCT. Most notably, our study evaluated the predictive performance of combined parameters. Previous studies have identified several individual spectral parameters associated with tumor grade. Akashi et al. (41) reported lower apparent diffusion coefficient values in poorly differentiated rectal adenocarcinoma; Yang et al. (23) found that AP-derived IC and NIC improved grading accuracy; and Chen et al. (42) observed higher Zeff, AP-derived NIC, and VP-derived NIC in high-grade CRC. However, none of these previous studies examined whether combining these parameters could further enhance diagnostic performance. In contrast, our findings suggest that integrating clinical features with DECT-derived parameters—particularly VP IC and ECV—may outperform individual parameters, possibly due to their complementary information reflecting distinct pathophysiological features of the tumor microenvironment, and they may offer incremental value beyond single-parameter assessments. Patients with higher-grade CRC tend to have poorer long-term survival than do those with lower-grade tumors, although clinical outcomes vary. In this regard, noninvasive DECT could serve as a source of auxiliary quantitative data to aid in preoperative tumor grading. The combined model incorporating CEA, VP IC, and ECV in our work appears to be capable of noninvasively stratifying tumor grades before surgery. Such an imaging strategy may help tailor postoperative surveillance schemes and stratify prognostic risks and could lower the reliance on diagnostic biopsies to some extent. Several notable limitations to this study should be acknowledged. First, we employed a single‑center design, and multicenter trials are needed to fully verify our findings. Second, the high-grade group included three histological subtypes with divergent biological features that could have caused analytical heterogeneity. Subgroup analyses could not be performed due to the insufficient sample size, and larger cohorts in future work are required for subtype-specific validation. Third, a small sample size carries a risk of model overfitting, and cutoffs generated from the whole cohort may overestimate sensitivity and specificity. Although bootstrap internal validation was conducted, external validation in larger independent cohorts is still necessary to confirm model robustness and generalizability. Conclusions DECT‑derived quantitative parameters have potential clinical value in identifying the histopathologic grade of CRC. They may serve as imaging biomarkers and provide valuable information for patients with CRC, offering important clues for understanding the biological behavior of tumors. Supplementary The article’s supplementary files as Acknowledgments We would like to thank our colleagues at the Institute of Radiology for their help and support in this study. Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The retrospective study was approved by the Institutional Review Board of the Second Hospital of Jilin University, and the requirement for informed consent was waived, as all patient data were fully anonymized, no extra interventions or patient contacts were conducted, and the study posed no physical or psychological risks to participants. Footnotes Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1-2641/rc Funding: This study was supported by the National Natural Science Foundation of China (grant No. 2071981), the Program of Youth Science and Technology Innovation and Entrepreneurship Outstanding Talents (Team) of Jilin Province, China (grant No. 20230508063RC), the Excellent Youth Training Foundation of Jilin University, China (grant No. 419080520665), the Innovation and Entrepreneurship Talent Funding Program of Jilin Province, China, and the Health Special Project of the Finance Department of Jilin Province, China and the Program of Key Research and Development—Medical and Health Field (Industrial Key Technologies) of Jilin Provincial Department of Science and Technology, China (No. 20260204050YY). Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1-2641/coif). The authors have no conflicts of interest to declare. Data Sharing Statement Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1-2641/dss References Associated Data Supplementary Materials The article’s supplementary files as Data Availability Statement Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1-2641/dss