Association of CT-derived extracellular volume fraction combined with serological markers with histological grading of colorectal cancer

Abstract

This study evaluates the association between computed tomography-derived extracellular volume fraction (CT-ECV), serological markers, and histological grading in colorectal cancer, and to assess their diagnostic value in predicting poorly differentiated tumors. This retrospective study included 200 patients with pathologically confirmed primary colorectal cancer from February 2021 to May 2024. All underwent dual-phase contrast-enhanced CT and serological testing before treatment. Based on World Health Organization classification, tumors were categorized as well-differentiated (G1, n = 97), moderately differentiated (G2, n = 57), or poorly differentiated (G3, n = 46). CT-ECV and arterial phase enhancement (ΔHU) were measured. Serological markers included carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA19-9), carbohydrate antigen 125 (CA125), and neutrophil-to-lymphocyte ratio (NLR). Statistical methods included ANOVA, Spearman correlation, ordinal logistic regression, and receiver operating characteristic analysis. Among 200 patients (116 males, 84 females; mean age 62.7 ± 10.3 years), no significant age or sex differences were observed among differentiation groups. CT-ECV, ΔHU, and levels of CEA, CA19-9, CA125, NLR, and platelet-to-lymphocyte ratio increased with poorer differentiation (all P < .001). Multivariate analysis showed CT-ECV (per 5% increase, OR = 1.86), ΔHU (per 10 HU, OR = 1.51), CEA, CA19-9, and NLR were independent predictors of poor differentiation. CT-ECV alone yielded an area under the curve of 0.82; CT-ECV + NLR reached 0.87; and the full model (CT-ECV + CEA + CA19-9 + NLR) achieved an area under the curve of 0.91. CT-ECV correlated moderately with CA125 (ρ = 0.51) and NLR (ρ = 0.47). Sensitivity analyses confirmed model stability. CT-ECV is significantly associated with histological grade in colorectal cancer and moderately correlates with serological markers. Combining CT-ECV with NLR and tumor markers enhances noninvasive preoperative prediction of poorly differentiated tumors, offering clinical value for grading and treatment planning.

1. Introduction

Colorectal cancer (CRC) is a common malignant tumor of the digestive system worldwide, ranking among the top causes of cancer-related morbidity and mortality in most countries.^[1,2] With population aging and changes in lifestyle, particularly in China, the disease burden of CRC continues to rise steadily.^[3] Although advances in screening programs and comprehensive treatment strategies have improved early diagnosis and management, the overall prognosis of CRC patients remains influenced by various biological factors, among which histological differentiation is a well-recognized prognostic indicator.^[4]

Histological grading reflects not only the degree of cellular atypia and proliferative capacity but also correlates closely with clinical behavior. Well-differentiated tumors typically exhibit slower growth and a lower potential for metastasis, whereas poorly or undifferentiated tumors tend to be more aggressive, with higher recurrence rates and worse survival outcomes.^[5] Therefore, accurate preoperative assessment of tumor differentiation plays a critical role in risk stratification, therapeutic decision-making, and prognostic evaluation. Currently, tumor grading is primarily based on histopathological evaluation of biopsy specimens obtained via colonoscopy. However, this approach has limitations, including sampling bias, procedural dependency, and invasiveness, which fall short of the requirements for noninvasive and efficient assessment in the era of precision medicine.^[6]

Imaging has become an essential noninvasive tool for tumor evaluation, with contrast-enhanced computed tomography (CT) playing a well-established role in the preoperative staging and metastasis assessment of colorectal cancer. However, the ability of conventional CT to assess tumor differentiation is limited, primarily relying on visual interpretation and lacking objective quantitative parameters. This approach is susceptible to inter-observer variability and diagnostic uncertainty. In recent years, increasing attention has been given to the relationship between imaging features and changes in the tumor microenvironment, particularly angiogenesis and stromal fibrosis, which are more prominent in poorly differentiated tumors.^[7]

Extracellular volume fraction (ECV), a quantitative parameter that reflects changes in the tumor stroma, can be calculated from delayed-phase contrast-enhanced CT images. It represents the distribution of contrast agent within both intravascular and extravascular-extracellular spaces.^[8] ECV serves as an indirect indicator of microvessel density and stromal remodeling and has been proposed as a feasible imaging biomarker for quantitatively assessing tumor heterogeneity. Prior studies have demonstrated its potential value in grading tumors such as hepatocellular carcinoma and pancreatic cancer, yet its application in colorectal cancer remains underexplored.^[9]

Additionally, the interplay between inflammation and tumor progression has emerged as a key research focus. Inflammatory cells contribute to tumorigenesis by releasing chemokines and growth factors. The neutrophil-to-lymphocyte ratio (NLR), a readily available and cost-effective peripheral blood marker, reflects systemic inflammatory response and has been associated with cancer stage, metastasis, and prognosis in various malignancies.^[10] NLR is gaining recognition for its value in predicting colorectal cancer aggressiveness, although its relationship with histological differentiation has not been clearly established.

Serum tumor markers such as carcinoembryonic antigen (CEA) and carbohydrate antigen 19-9 (CA19-9) are also widely used in the diagnosis and surveillance of colorectal cancer.^[11] While their expression levels may correlate with tumor biology and differentiation, existing evidence is inconsistent, necessitating further validation and integrative analysis.

In summary, combining imaging biomarkers such as CT-derived ECV with inflammatory indicators like NLR and traditional serum tumor markers (CEA, CA19-9) may help establish a comprehensive, noninvasive assessment strategy to predict histological grading of colorectal cancer. However, current clinical research has primarily focused on these markers in isolation, and limited efforts have been made to integrate imaging and serological data for preoperative grading. This study addresses this gap by systematically evaluating the association between CT-ECV and key blood-based parameters, and by exploring the diagnostic value of their combination. By leveraging both tumor microenvironment characteristics and systemic biological responses, the proposed model offers a novel, multi-dimensional approach for assessing tumor differentiation. This integration holds promise for improving preoperative risk stratification and guiding personalized treatment decisions. Accurate preoperative grading can help guide decisions on neoadjuvant therapy – for example, poorly differentiated tumors may warrant more aggressive or combined treatment strategies – and may influence surgical planning, such as whether to expand lymph node dissection or adopt more radical resection approaches. Therefore, the aim of this study is to investigate the diagnostic performance of CT-ECV combined with serological markers in predicting tumor differentiation, thereby providing a theoretical and practical basis for enhanced noninvasive histological grading.

2. Materials and methods

2.1. Study design

This was a single-center retrospective cohort study that consecutively enrolled patients with pathologically confirmed Colorectal cancer at our institution between February 2021 and May 2024. A total of 200 eligible patients were included based on predefined inclusion and exclusion criteria. According to the World Health Organization classification of tumors of the digestive system,^[12] patients were categorized into 3 groups based on tumor histological differentiation: well-differentiated (G1, n = 97; preservation of > 95% glandular architecture), moderately differentiated (G2, n = 57; preservation of 50%–95% glandular architecture), and poorly differentiated (G3, n = 46; preservation of < 50% glandular architecture).

2.2. Inclusion and exclusion criteria

2.2.1. Inclusion criteria

Age ≥ 18 years; the patient has not received preoperative systemic therapy or preoperative radiotherapy and chemotherapy, targeted therapy, or immunotherapy (to avoid interfering with pathological grading and imaging parameters); completion of dual-phase contrast-enhanced CT (arterial and portal venous phases) and serological testing within 2 weeks before surgery; postoperative pathological diagnosis of primary Colorectal cancer, including mucinous adenocarcinoma; complete clinical, pathological (including TNM [Tumor, Node, Metastasis Classification System] staging and histological grading), and follow-up data.

2.2.2. Exclusion criteria

History of other malignancies or prior antitumor therapy; CT images not meeting the requirements for ECV analysis (e.g., motion artifacts, slice thickness > 5 mm); presence of hematologic disorders or severe hepatic/renal dysfunction that may affect interpretation of serological markers; evident chronic infection or acute inflammatory conditions at the time of testing, which could influence serological results.

2.3. Data collection

Clinical data collected included patient sex, age, tumor location, maximum tumor diameter, TNM stage, lymph node metastasis, and distant metastasis status. Tumor size was measured directly from postoperative pathological specimens, with the maximum diameter (cm) recorded. TNM staging was determined according to the 8th edition of the American Joint Committee on Cancer staging system.^[13] Lymph node and distant metastasis were confirmed through postoperative pathological examination. All clinical data were entered and verified independently by 2 researchers to ensure accuracy and completeness.

Imaging parameters were acquired using multidetector spiral CT scanners with at least 64 detector rows. All patients fasted for at least 6 hours prior to the examination. An iodinated contrast agent was administered intravenously at a dose of 1.5 mL/kg and an injection rate of 3.0 mL/s. Arterial, venous, and delayed-phase images were obtained. CT-derived extracellular volume fraction (CT-ECV) was calculated based on delayed-phase images. Regions of interest were manually drawn within the viable tumor tissue, with the spleen selected as a reference organ. Necrotic areas, calcifications, and large vessels were avoided. CT attenuation values (Hounsfield units) were independently measured by 2 experienced radiologists, and the average value was used for analysis.

The CT-ECV was calculated using the following formula^[14]:

$${\displaystyle \begin{array}{l}{\displaystyle CT-ECV\left(\mathrm{\%}\right)=(\frac{\mathrm{\Delta }H{U}_{tumor}}{\mathrm{\Delta }H{U}_{spleen}}\times (1-HCT))\times 100}\end{array}}$$

ΔHU was defined as the difference in CT attenuation values between the delayed-phase and the non-contrast phase. Hematocrit (Hct) was obtained from the complete blood count performed on the same day as the CT examination. In addition, tumor ΔHU was calculated as the difference between arterial phase and non-contrast CT values. The imaging measurements were assessed for inter-observer agreement using the intraclass correlation coefficient, which demonstrated excellent consistency (intraclass correlation coefficient = 0.92).

Serological testing was performed concurrently with the imaging examination. All patients underwent fasting venous blood collection (5 mL) in the early morning prior to surgery. After centrifugation, the serum samples were immediately analyzed. Tumor markers, including carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA19-9), and carbohydrate antigen 125 (CA125), were quantified using electrochemiluminescence immunoassay (ECLIA) on the Roche Cobas E601 analyzer. Inflammatory markers were obtained from complete blood counts performed with an automated hematology analyzer (Sysmex XN-9000), from which the NLR, platelet-to-lymphocyte ratio (PLR), and lymphocyte-to-monocyte ratio (LMR) were calculated. All serological analyses were conducted by the hospital’s central clinical laboratory, following standard quality control procedures.

2.4. Statistical analysis

Various statistical methods were employed for data analysis in this study. For continuous variables, normality was tested using the Shapiro–Wilk test. Variables with a normal distribution were expressed as mean ± standard deviation (SD) and compared between groups using 1-way analysis of variance (ANOVA). Non-normally distributed data were presented as median with interquartile range (IQR) and compared using the Kruskal-Wallis test. Categorical variables were presented as frequencies and percentages, with comparisons made using the chi-square test or Fisher exact test, as appropriate. Bonferroni correction was applied for multiple comparisons when necessary. Spearman rank correlation analysis was used to evaluate the relationships between CT-ECV and serological markers. The strength of correlation was interpreted as follows: |ρ| < 0.3, weak; 0.3 ≤ |ρ| < 0.6, moderate; |ρ| ≥ 0.6, strong.

To assess independent associations between variables and tumor differentiation, an ordinal logistic regression model (proportional odds model)^[15] was constructed. Variables with a P-value < .1 in univariate analysis were included in the multivariate model using a stepwise selection method (entry criterion P < .05, removal criterion P > .10). Model goodness-of-fit was assessed using Nagelkerke R², and the proportional odds assumption was verified using the Brant test (P > .05). Multicollinearity was checked using the variance inflation factor, with variance inflation factor > 5 indicating collinearity. Receiver operating characteristic (ROC) curve analysis was performed to evaluate the diagnostic performance of individual and combined parameters, with area under the curve (AUC) comparisons conducted using the DeLong test. All statistical tests were 2-sided, and a P-value < .05 was considered statistically significant. Statistical analyses were performed using SPSS version 26.0 (IBM Corp., Armonk).

Missing data in variables were minimal (<5%) and were assumed to be missing at random. To avoid potential bias and preserve statistical power, missing values were handled using multiple imputation based on chained equations. Five imputed datasets were generated and pooled for analysis according to Rubin rules. The results were consistent with those from complete-case analyses, confirming the robustness of the findings.

3. Results

3.1. Baseline clinical characteristics of patients with different tumor differentiation levels

There were no statistically significant differences in sex, age, or tumor location among the 3 histological differentiation groups (P > .05). However, significant differences were observed in tumor size across groups, with tumor diameter increasing as the degree of differentiation decreased (F = 6.235, P = .002). TNM staging also differed significantly among the groups: the proportion of stage III–IV disease was highest in the poorly differentiated group (G3, 63.0%), while stage I–II disease was most common in the well-differentiated group (G1, 72.2%) (χ² = 15.632, P < .001). Similarly, the incidence of lymph node metastasis increased with decreasing tumor differentiation (χ² = 14.108, P = .001), and the rate of distant metastasis was significantly higher in the poorly differentiated group (G3, 37.0%) compared to the other groups (χ² = 22.445, P < .001, Table 1).

Table 1.

Baseline clinical characteristics of colorectal cancer patients according to tumor differentiation.

Characteristics	Total (n = 200)	Well-differentiated (G1) (n = 97)	Moderately differentiated (G2) (n = 57)	Poorly differentiated (G3) (n = 46)	χ²/F	P-value
Gender, n (%)
Male	120 (60.0%)	60 (61.9%)	32 (56.1%)	28 (60.9%)	0.462	.794
Female	80 (40.0%)	37 (38.1%)	25 (43.9%)	18 (39.1%)
Age (yr, mean ± SD)	60.8 ± 10.2	61.3 ± 10.5	59.9 ± 9.8	60.4 ± 10.3	0.321	.726
Tumor location, n (%)
Ascending colon	70 (35.0%)	35 (36.1%)	18 (31.6%)	17 (37.0%)	1.825	.768
Transverse colon	30 (15.0%)	13 (13.4%)	9 (15.8%)	8 (17.4%)
Descending colon	50 (25.0%)	26 (26.8%)	14 (24.6%)	10 (21.7%)
Sigmoid colon	50 (25.0%)	23 (23.7%)	16 (28.0%)	11 (23.9%)
Tumor size (cm, mean ± SD)	4.8 ± 1.7	4.3 ± 1.5	5.0 ± 1.6	5.5 ± 1.8	6.235	.002
TNM stage, n (%)
Stage I–II	120 (60.0%)	70 (72.2%)	33 (57.9%)	17 (37.0%)	15.632	<.001
Stage III–IV	80 (40.0%)	27 (27.8%)	24 (42.1%)	29 (63.0%)
Lymph node metastasis, n (%)
Yes	120 (60.0%)	48 (49.5%)	36 (63.2%)	36 (78.3%)	14.108	.001
No	80 (40.0%)	49 (50.5%)	21 (36.8%)	10 (21.7%)
Distant metastasis, n (%)
Yes	30 (15.0%)	5 (5.1%)	8 (14.0%)	17 (37.0%)	22.445	<.001
No	170 (85.0%)	92 (94.9%)	49 (86.0%)	29 (63.0%)

SD = standard deviation, TNM = Tumor, Node, Metastasis Classification System.

3.2. Comparison of CT-ECV, ΔHU, and serological markers among patients with different tumor differentiation levels

Significant differences were observed in CT-ECV, tumor ΔHU, and serological markers (CEA, CA19-9, CA125, NLR, and PLR) among the different tumor differentiation groups (Table 2). CT-ECV was highest in the poorly differentiated group (G3: 37.2 ± 6.1%), followed by the moderately differentiated group (G2: 33.5 ± 5.7%), and lowest in the well-differentiated group (G1: 29.8 ± 5.2%) (F = 22.184, P < .001). Tumor ΔHU also increased with lower differentiation (F = 19.673, P < .001).

Table 2.

Distribution of CT-ECV, ΔHU (tumor), and serum biomarkers in colorectal cancer patients with different tumor differentiation.

Variables	Total (n = 200)	Well-differentiated (G1) (n = 97)	Moderately differentiated (G2) (n = 57)	Poorly differentiated (G3) (n = 46)	χ²/F/H	P-value
CT-ECV (%)	32.5 ± 6.4	29.8 ± 5.2	33.5 ± 5.7	37.2 ± 6.1	22.184	<.001*
ΔHU (tumor)	48.6 ± 10.5	44.7 ± 8.6	49.2 ± 9.7	54.5 ± 11.0	19.673	<.001*
CEA (ng/mL)	12.3 (5.8–27.6)	8.5 (3.2–18.7)	14.9 (7.5–32.1)	25.4 (12.6–58.7)	21.457	<.001*
CA19-9 (U/mL)	35.2 (18.4–79.5)	26.7 (14.5–52.3)	39.8 (19.6–85.9)	56.3 (28.7–110.4)	17.983	<.001*
CA125 (U/mL)	21.4 (12.5–46.8)	16.8 (10.2–34.7)	24.9 (13.6–57.3)	37.2 (18.5–69.4)	13.746	.001*
NLR	3.2 (2.1–5.4)	2.6 (1.8–3.9)	3.5 (2.4–5.8)	4.8 (3.1–7.3)	18.629	<.001*
PLR	176.3 (132.5–235.7)	154.8 (120.4–198.6)	182.7 (140.9–245.3)	223.6 (167.8–290.4)	16.302	<.001*

CA125 = carbohydrate antigen 125, CA19-9 = carbohydrate antigen 19-9, CEA = carcinoembryonic antigen, CT-ECV = CT-derived extracellular volume fraction, NLR = neutrophil-to-lymphocyte ratio, PLR = platelet-to-lymphocyte ratio.

Among serological markers, CEA, CA19-9, and CA125 levels rose as differentiation decreased. The median CEA in the G3 group was 25.4 (12.6, 58.7) ng/mL, significantly higher than in G2 and G1 (H = 21.457, P < .001). CA19-9 and CA125 also differed significantly across groups (P < .001 and P = .001, respectively).

Inflammatory markers NLR and PLR were significantly elevated in poorly differentiated tumors compared to well-differentiated ones (H = 18.629 and H = 16.302, both P < .001).

3.3. Multivariate regression analysis

Ordinal logistic regression showed that CT-ECV, ΔHU, CEA, CA19-9, and NLR were independently associated with tumor differentiation (all P < .05) (Table 3). Each 5% increase in CT-ECV was linked to an 86% higher risk of poor differentiation (OR = 1.86, 95% CI = 1.39–2.49), and each 10 HU increase in ΔHU raised the risk by 51% (OR = 1.51, 95% CI = 1.06–2.14). For serological markers, CEA (per 10 ng/mL: OR = 1.39), CA19-9 (per 50 U/mL: OR = 1.32), and NLR (per unit: OR = 1.57) were significant predictors.

Table 3.

Multivariable ordinal logistic regression analysis of factors associated with tumor differentiation grade.

Variable	β	OR (95% CI)	P-value
Imaging parameters
CT-ECV (per 5% increase)	0.62 (0.15)	1.86 (1.39–2.49)	<.001
ΔHU (tumor) (per 10 HU increase)	0.41 (0.18)	1.51 (1.06–2.14)	.022
Serum biomarkers
CEA (per 10 ng/mL increase)	0.33 (0.11)	1.39 (1.12–1.72)	.003
CA19-9 (per 50 U/mL increase)	0.28 (0.09)	1.32 (1.11–1.58)	.002
NLR (per 1 unit increase)	0.45 (0.13)	1.57 (1.21–2.03)	<.001
Clinical confounders
TNM Stage (III–IV vs I–II)	0.87 (0.22)	2.39 (1.55–3.68)	<.001
Tumor size (per 1 cm increase)	0.21 (0.10)	1.23 (1.01–1.50)	.041

Bold values indicate statistical significance.

CA125 = carbohydrate antigen 125, CA19-9 = carbohydrate antigen 19-9, CEA = carcinoembryonic antigen, CI = confidence interval, CT-ECV = computed tomography-derived extracellular volume fraction, NLR = neutrophil-to-lymphocyte ratio, OR = odds ratio, TNM = Tumor, Node, Metastasis Classification System.

Stage III–IV tumors had a 2.39-fold higher risk of poor differentiation compared to stage I–II (95% CI = 1.55–3.68), and each 1 cm increase in tumor size raised the risk by 23% (OR = 1.23, 95% CI = 1.01–1.50). The model showed good fit (Nagelkerke R² = 0.48) and met the proportional odds assumption (P = .294). CA125 and PLR were excluded due to multicollinearity with NLR and limited added value.

3.4. Diagnostic performance analysis

ROC analysis showed that CT-ECV alone had an AUC of 0.82 (95% CI = 0.75–0.88) for predicting poorly differentiated colorectal cancer (G3), outperforming traditional markers such as CEA (AUC = 0.73, P = .021) and CA19-9 (AUC = 0.71, P = .015) (Table 4). Among combined models, CT-ECV + NLR achieved the best performance (AUC = 0.87, P = .001), while the comprehensive model including CT-ECV, CEA, CA19-9, and NLR reached the highest AUC of 0.91 (P < .001).

Table 4.

Receiver operating characteristic analysis of individual and combined parameters for predicting poor differentiation.

Model	AUC (95% CI)	Sensitivity (%)	Specificity (%)	cutoff	P-value
Individual parameters
CT-ECV	0.82 (0.75–0.88)	78.3	76.6	≥34.5%	–
CEA	0.73 (0.66–0.80)	71.7	68.2	≥18.7 ng/mL	.021
CA19-9	0.71 (0.63–0.78)	67.4	72.5	≥47.2 U/mL	.015
NLR	0.76 (0.69–0.83)	73.9	74.1	≥3.8	.038
Combined models
CT-ECV + CEA	0.85 (0.79–0.91)	82.6	79.3	–	.003
CT-ECV + NLR	0.87 (0.81–0.93)	84.8	81.6	–	.001
Comprehensive model*	0.91 (0.86–0.96)	89.1	86.4	–	<.001

AUC = area under the curve, CA19-9 = carbohydrate antigen 19-9, CEA = carcinoembryonic antigen, CI = confidence interval, CT-ECV = computed tomography-derived extracellular volume fraction, NLR = neutrophil-to-lymphocyte ratio.

^*Comprehensive model includes: CT-ECV, CA19-9, and NLR.

Additional pairwise combinations were evaluated, with CT-ECV + NLR showing the best balance of diagnostic value and clinical utility. Other combinations provided limited incremental benefit or showed high intercorrelation (e.g., CEA and CA19-9, R = 0.62) and were not reported in detail.

3.5. Correlation between serological markers and CT-ECV

Spearman correlation analysis showed that all tested serological markers were significantly correlated with CT-ECV values (P < .001) (Table 5). Among tumor markers, CA125 had the strongest correlation with CT-ECV (ρ = 0.51), suggesting a possible link between stromal fibrosis and peritoneal metastasis. CEA (ρ = 0.42) and CA19-9 (ρ = 0.38) also showed moderate correlations.

Table 5.

Correlation between serum biomarkers and CT-ECV (Spearman analysis).

Serum biomarker	Correlation coefficient (ρ)	P-value	Strength of association
CEA	0.42	<.001	Moderate
CA19-9	0.38	<.001	Weak–moderate
CA125	0.51	<.001	Moderate
NLR	0.47	<.001	Moderate
PLR	0.39	<.001	Weak–moderate
LMR	−0.36	<.001	Weak–moderate

CA125 = carbohydrate antigen 125, CA19-9 = carbohydrate antigen 19-9, CEA = carcinoembryonic antigen, CT-ECV = computed tomography-derived extracellular volume fraction, LMR = lymphocyte-to-monocyte ratio, NLR = neutrophil-to-lymphocyte ratio, PLR = platelet-to-lymphocyte ratio.

^*Strength of association criteria: ρ <0.3 (weak), 0.3≤ ρ <0.6 (moderate), ρ ≥0.6 (strong).

Regarding inflammatory markers, NLR (ρ = 0.47) and PLR (ρ = 0.39) were positively correlated with CT-ECV, while LMR showed a negative correlation (ρ = −0.36). The overall strength of correlation with CT-ECV ranked as follows: CA125 > NLR > CEA > PLR > CA19-9 > LMR.

3.6. Sensitivity analysis

The combined diagnostic model (CT-ECV + CA19-9 + NLR) demonstrated robust performance across all sensitivity analyses (see Table 6). The base model yielded an AUC of 0.91 (95% CI = 0.86–0.96), with each 1-SD increase in the combined score associated with a 2.15-fold higher risk of poor differentiation (95% CI = 1.68–2.75).

Table 6.

Sensitivity analysis of the combined model (CT-ECV + CEA + CA19-9 + NLR).

Analysis type	Subgroup/adjustment	OR (95% CI)	AUC change
Base model	Total (n = 200)	2.15 (1.68–2.75)*	0.91 (0.86–0.96)*
Histology
Non-mucinous	(n = 168)	2.12 (1.63–2.76)*	0.90 (0.85–0.95)*
Mucinous	(n = 32)	2.24 (1.58–3.18)*	0.89 (0.82–0.96)*
Stage stratification
Stage I–II	(n = 120)	2.08 (1.59–2.72)*	0.88 (0.82–0.94)*
Stage III–IV	(n = 80)	2.19 (1.67–2.87)*	0.87 (0.81–0.93)*
Methodological adjustments
Multiple imputation	(Missing data)	2.13 (1.65–2.74)*	ΔAUC = −0.01*
Outlier removal	(±3SD)	2.11 (1.64–2.71)*	ΔAUC = −0.02*
Inter-observer variation
Radiologist A	(CT-ECV measurement)	2.16 (1.67–2.79)*	0.91 (0.86–0.96)*
Radiologist B	(CT-ECV measurement)	2.14 (1.66–2.76)*	0.90 (0.85–0.95)*

AUC = area under the curve, CA19-9 = carbohydrate antigen 19-9, CEA = carcinoembryonic antigen, CI = confidence interval, CT-ECV = computed tomography-derived extracellular volume fraction, NLR = neutrophil-to-lymphocyte ratio, OR = odds ratio, SD = standard deviation.

^*Statistical significance, P < .05.

Subgroup analysis showed consistent performance across tumor types: AUC = 0.90 for non-mucinous adenocarcinoma and 0.89 for mucinous types. Stratified by stage, AUCs were 0.88 (stage I–II) and 0.87 (stage III–IV). Methodological validation confirmed minimal changes: AUC dropped by only 0.01 after multiple imputation for missing data and by 0.02 after excluding outliers (±3 SD). Inter-observer agreement for CT-ECV was excellent (ICC = 0.92), and model AUCs remained stable (0.90–0.91). Calibration was good (Hosmer–Lemeshow test, P = .35), and ORs showed limited variability (2.08–2.24), supporting the model’s stability across pathological subtypes, stages, and measurement conditions.

4. Discussion

This study demonstrated that CT-derived extracellular volume fraction (CT-ECV), along with serological markers including NLR, carcinoembryonic antigen (CEA), and carbohydrate antigen 19-9 (CA19-9), was significantly associated with histological grading in colorectal cancer. CT-ECV values, inflammatory markers, and tumor marker levels all increased with poorer tumor differentiation. Multivariate analysis identified CT-ECV, ΔHU, NLR, CEA, and CA19-9 as independent predictors of poor differentiation. Notably, the combined model incorporating CT-ECV and key serological markers achieved high diagnostic accuracy (AUC = 0.91) for predicting poorly differentiated tumors, and its performance was consistent across subgroups and sensitivity analyses.

4.1. Association between CT-ECV and tumor differentiation in colorectal cancer

This study demonstrated that CT-ECV values increased significantly with decreasing tumor differentiation, indicating a positive correlation between ECV and histological grade in colorectal cancer. This supports the potential value of ECV as an imaging biomarker reflecting tumor microenvironmental changes in differentiation assessment. Biologically, poorly differentiated tumors are often characterized by enhanced angiogenesis and more prominent stromal reactions, which increase vascular permeability and extracellular matrix expansion – factors that lead to greater distribution of contrast agents in the extravascular-extracellular space.^[16] As a quantitative measure of this distribution, ECV tends to be higher in more aggressive tumors, aligning with their pathological characteristics.

Similar findings have been reported in other solid tumors. For example, Ma et al^[17] found that ECV correlates with stromal fibrosis in pancreatic cancer, suggesting its role in reflecting tissue remodeling. Our study extends this concept to colorectal cancer and provides quantitative evidence that ECV is a stronger predictor of tumor grade than traditional enhancement metrics such as ΔHU. The superior AUC of ECV in ROC analysis highlights its diagnostic robustness and potential clinical utility. Notably, ECV calculation based on delayed-phase CT offers high reproducibility and quantifiability, making it suitable for integration into routine imaging workflows.

4.2. Inflammatory mechanisms and predictive value of NLR in tumor differentiation

The NLR is a widely studied inflammation-related marker in oncology, with elevated levels linked to poor prognosis in various solid tumors. In this study, preoperative NLR levels differed significantly between well- and poorly differentiated colorectal cancers and remained an independent predictor of tumor grade in multivariate analysis.^[18] This suggests that tumor-associated chronic inflammation may contribute to tumor progression and dedifferentiation.

Mechanistically, neutrophils promote angiogenesis and tumor invasion through the secretion of pro-angiogenic factors (e.g., VEGF, IL-8) and matrix-degrading enzymes, while lymphocytes are key players in antitumor immune responses. An elevated NLR indicates both enhanced inflammatory activity and suppressed immune surveillance, correlating with more aggressive tumor behavior. Consistent with previous findings, such as the meta-analysis by Li et al,^[19] which confirmed the prognostic significance of NLR in colorectal cancer, our study expands its utility to the prediction of histological differentiation, highlighting its clinical relevance in preoperative tumor assessment.

4.3. Clinical significance of CA19-9 in relation to tumor differentiation

As indirect indicators of tumor burden and biological behavior, serum tumor markers play an important role in noninvasive monitoring. In this study, CA19-9 levels varied significantly among different differentiation groups and emerged as an independent predictor in multivariate analysis. This finding aligns with previous reports, such as that by Plemenos et al,^[20] which suggest that CA19-9 may facilitate tumor cell–endothelial adhesion and promote metastasis. Poorly differentiated tumors, which often exhibit more aggressive and invasive behavior, tend to express higher levels of CA19-9, implying that its elevation may reflect impaired differentiation and enhanced metastatic potential.^[21,22]

In contrast, although CEA showed some association with tumor grade in univariate analysis, it was not retained in the final multivariate model. This may be due to heterogeneity in its expression, interference from inflammatory conditions, or individual metabolic differences. These findings are consistent with previous studies that have reported inconsistent associations between CEA and tumor differentiation, highlighting its limitations as an independent predictor.

4.4. Clinical utility of the combined predictive model

CT-ECV, NLR, and CA19-9 reflect tumor heterogeneity from different biological dimensions – imaging, inflammation, and tumor secretion, respectively. The combined model incorporating these 3 markers demonstrated strong discriminative ability in identifying poorly differentiated colorectal cancer (AUC = 0.91), significantly outperforming any single parameter. This finding suggests that integrating multi-dimensional data enables a more comprehensive evaluation of tumor biology. Given the routine availability of contrast-enhanced CT and basic serological tests in clinical practice, the proposed model offers good feasibility and potential for clinical translation.

4.5. Clinical implications and future directions

This study proposes a novel, noninvasive preoperative grading approach by integrating imaging-based and blood-based biomarkers, with potential applications in surgical planning, adjuvant therapy selection, and prognostic evaluation for colorectal cancer patients. The model demonstrated strong robustness across multiple sensitivity analyses, supporting its adaptability in varied clinical settings.

However, several limitations should be acknowledged. First, the study was conducted at a single center with a relatively limited sample size, which may affect the generalizability of the findings. Selection bias is also possible, as only patients with complete imaging and serological data prior to surgery were included, potentially excluding cases with atypical presentations or comorbidities. These factors may influence the observed associations and diagnostic performance. Moreover, although CT-ECV is a promising imaging biomarker, its accuracy is dependent on standardized scanning protocols and precise region-of-interest placement. Similarly, serological markers such as NLR and CA19-9 can be affected by nontumor-related inflammatory or physiological conditions.

To address these limitations, future research should include larger, multicenter prospective cohorts to validate and refine the predictive model across diverse populations and clinical environments. Additionally, integrating artificial intelligence and radiomics techniques may further improve prediction accuracy, enable automation, and reduce inter-observer variability in imaging-based grading. Combining such tools with routine clinical data holds promise for the development of intelligent, real-time decision support systems in personalized cancer care.

5. Conclusion

This study confirmed that CT-derived extracellular volume fraction (CT-ECV), peripheral blood NLR, and serum CA19-9 levels are closely associated with histological grading in Colorectal cancer. All 3 were identified as independent predictors of poorly differentiated tumors. Compared to conventional imaging features, CT-ECV – a quantitative marker reflecting stromal remodeling and angiogenesis – demonstrated superior diagnostic performance in grading assessment. The multiparametric model combining ECV, NLR, and CA19-9 significantly improved the accuracy of preoperative prediction of tumor differentiation, offering a promising, noninvasive tool for individualized tumor evaluation. This model shows strong potential for clinical application and warrants further validation through large-scale prospective multicenter studies.

Author contributions

Conceptualization: Hongli Zhou, Yuan Feng, Yu Cai.

Data curation: Hongli Zhou, Sha Yang, Yu Cai.

Formal analysis: Hongli Zhou, Yuan Feng, Sha Yang, Wei Li, Yu Cai.

Investigation: Yu Cai.

Methodology: Yu Cai.

Visualization: Yu Cai.

Writing – original draft: Hongli Zhou, Yuan Feng, Sha Yang, Wei Li, Yu Cai.

Writing – review & editing: Hongli Zhou, Yuan Feng, Yu Cai.