Nomogram based on MRI images and clinical data for differentiating mucinous from non-mucinous rectal adenocarcinoma

Abstract

Background

Preoperative differentiation between rectal mucinous adenocarcinoma (MAC) and non-mucinous adenocarcinoma (NMAC) remains a clinical challenge. This study aimed to develop and validate a nomogram incorporating baseline clinical characteristics and magnetic resonance imaging (MRI) features to distinguish MAC from NMAC.

Methods

This retrospective study included clinical baseline characteristics, laboratory parameters, and MRI features of patients with MAC and NMAC from two medical centers. Relevant variables were identified using univariate logistic regression analysis. Separate models based on clinical and imaging features were developed and subsequently integrated into a nomogram. Model performance was evaluated using receiver operating characteristic (ROC) curves and the area under the curve (AUC), and decision curve analysis (DCA) was conducted to assess clinical utility.

Results

Data from 221 patients (NMAC = 160, MAC = 61) from Center 1 were collected for this study. Data from 76 patients (NMAC = 54, MAC = 22) from Center 2 were used as an external validation cohort to verify the robustness of the models. We developed three models: a clinical model, an imaging feature model, and a nomogram. The nomogram integrating both clinical and imaging features demonstrated the best performance, with an AUC of 0.937 (95% CI, 0.894–0.979) in the training cohort and 0.882 (95% CI, 0.793–0.971) in the validation cohort. In the validation cohort, the nomogram achieved a sensitivity of 0.869, specificity of 0.925, and accuracy of 0.909. Furthermore, calibration curves confirmed good agreement between the predicted and observed outcomes.

Conclusions

The nomogram integrating clinical characteristics with MRI features enables efficient and practical differentiation between rectal MAC and NMAC, providing a valuable reference for individualized treatment decisions.

Background

Rectal cancer is the third most common cancer and the fourth leading cause of cancer-related death worldwide [1]. In recent years, both the incidence and mortality of rectal cancer have been rising in China, ranking third and fifth, respectively, among all malignant tumors [2]. Mucinous rectal adenocarcinoma (MAC) accounts for approximately 10–15% of all rectal cancer cases [3–5] and represents a distinct histopathological subtype of rectal cancer. The diagnostic criteria, as defined by the World Health Organization (WHO), require that extracellular mucin constitutes more than 50% of the tumor components [6]. Histopathological features are characterized by tumor epithelial cells surrounding extracellular mucin pools [7]. Population-based studies have confirmed that mucinous adenocarcinoma is an independent adverse prognostic factor in patients with rectal cancer [8].

Existing studies have shown that the local recurrence rate of MAC is higher than that of non-mucinous adenocarcinoma (NMAC), and it has a poorer prognosis, even when confined to the muscularis propria, making local resection unsuitable [9, 10]. For patients with locally advanced rectal cancer, the standard treatment regimen involves neoadjuvant chemoradiotherapy (CRT) followed by surgical intervention [11]. It has been reported that, compared with NMAC, MAC exhibits poorer histological response to preoperative CRT, lower pathological complete response (pCR) rates, less significant downstaging of tumor (T) and lymph node (N) stages, and more limited tumor regression [12, 13]. This variability in treatment response underscores the urgent need to develop non-invasive, precise diagnostic tools. Improved recognition of MAC would enable clinicians to tailor therapeutic strategies more precisely. Patients with MAC can benefit from intensified multimodal approaches [14], including preoperative CRT dose escalation, extended neoadjuvant-to-surgery intervals to enhance tumor regression, or more extensive resection with comprehensive lymphadenectomy [15, 16].

Currently, preoperative endoscopic biopsy is routinely performed as part of standard clinical evaluation; however, due to its limited sampling depth and range, it typically captures only the superficial portion of the tumor. As a result, it is difficult to accurately assess mucin content based on the small volume of biopsy specimens, making reliable identification of MAC challenging [17, 18]. Studies have reported that the diagnostic accuracy of MAC at initial biopsy is as low as 5%–8% [12, 13]. Although numerous studies have shown significant differences between MAC and NMAC in terms of biomarkers—such as BRAF and KRAS mutations, microsatellite instability (MSI), and DNA mismatch repair status—testing for these markers usually requires invasive biopsy and costly genetic analyses [6, 14, 19, 20]. In contrast, clinical baseline characteristics and imaging features are non-invasive and readily accessible in clinical practice. Previous studies have noted that MAC typically appears as high signal intensity on T2-weighted images (T2WI) [6, 14, 18, 21]. Additionally, apparent diffusion coefficient (ADC) values have been identified as potential indicators of tumor aggressiveness in rectal cancer [22, 23]. Moreover, compared with NMAC, MAC has been reported to present with more aggressive clinical features, including larger primary tumors, deeper invasion, and a higher likelihood of metastasis [24, 25]. However, previous research has not integrated multiple factors into a comprehensive and practical model for accurately distinguishing MAC from NMAC.

Based on this, our study aims to develop a simple and effective model for distinguishing MAC from NMAC by retrospectively analyzing the clinical baseline characteristics and MRI imaging features of rectal cancer patients prior to surgery, thereby assisting clinicians in optimizing treatment strategies.

Methods

Patients

The study was approved by the Ethics Committee of XXX (Ethics number: SWYX: NO.2024-652). As this study was a retrospective analysis of pre-existing imaging data that neither involved patient identifiers nor affected clinical management, the requirement for informed consent was waived by the institutional ethics committee.

This retrospective study included rectal cancer patients who underwent surgical treatment from September 2017 to September 2023 at two centers: xxx (Center 1) and xxx (Center 2). Inclusion criteria were as follows: (1) Histopathologically confirmed MAC or NMAC following surgical resection. (2) Underwent standard contrast-enhanced pelvic MRI within two weeks prior to surgery. Exclusion criteria included: (1) History of neoadjuvant therapy. (2) Presence of signet ring cell carcinoma or mixed histology involving signet ring cell components. (3) Presence of other primary malignancies. (4) Poor image quality that impeded analysis, such as significant artifacts, incomplete scan coverage, or missing sequences. (5) Incomplete clinical data. (6) Patients who underwent palliative surgery. (7) Patients with recurrent rectal cancer.

In this study, we adhered to the histological criteria for mucinous adenocarcinoma as defined by the World Health Organization (WHO) [26]. Tumors were classified as MAC only if mucinous components accounted for at least 50% of the tumor volume. Those with focal mucinous features or mucinous components comprising less than 50% were categorized as NMAC. Distant metastases were excluded preoperatively through thoracoabdominal imaging examinations.

Fig. 1 shows the participant selection flowchart.

Fig. 1.

Flowchart of patient inclusion process

Data collection

Baseline clinical characteristics were obtained from patients’electronic medical records. Clinical characteristics included patient sex, age at diagnosis, tumor size (measured as the longest diameter on the axial T2WI slice showing the maximal tumor cross-section), tumor location, baseline TNM stage (Staged according to the TNM classification system of the American Joint Committee on Cancer [6]), presence of peritoneal involvement, and selected laboratory parameters. The laboratory parameters included carcinoembryonic antigen (CEA), carbohydrate antigen 19–9 (CA 19–9), platelet count (PLT), albumin (ALB), and absolute lymphocyte count (ALC) and absolute neutrophil count (ANC). (Tumor location was defined as the distance from the tumor’s inferior boundary to the anal verge (DTAV), measured on the standard sagittal T2WI sequence used for rectal tumor localization to ensure clear anatomical delineation. The anal verge was defined on sagittal T2WI as the inferior margin of the external anal sphincter, and DTAV was measured along the central axis of the rectal lumen [27]. Tumors with DTAV ≤ 5 cm were classified as low rectal cancer, those with DTAV of 5–10 cm as mid-rectal cancer, and those with DTAV of 10–15 cm as high rectal cancer.)

Preoperative baseline MRI images of MAC and NMAC patients, acquired within two weeks prior to surgery, were retrieved from the Picture Archiving and Communication System (PACS).

MRI protocols

All rectal MRI examinations were performed on a 1.5-Tesla system (Magnetom Avanto or Aera, Siemens Healthineers, Germany) using a dedicated pelvic phased-array surface coil. The scanning protocol was designed for comprehensive anatomical and functional assessment of rectal tumors, including high-resolution T2-weighted imaging in oblique axial and sagittal planes (oblique axial TR/TE = 4000/100 ms; sagittal TR/TE = 5800/80 ms; slice thickness 4–5 mm; FOV 24–26 × 24–26 cm; matrix adjusted according to patient size) to visualize tumor morphology, boundary, mucin distribution, rectal wall layers, and local anatomy. Diffusion-weighted imaging (DWI, b-values = 0 and 800 s/mm²) was used to assess water diffusion in tumor tissue, with ADC maps automatically generated for quantitative analysis. Dynamic contrast-enhanced T1-weighted imaging (3D VIBE) was acquired before and at multiple time points after intravenous Gd-DTPA administration (approximately 20–25 s, 55–60 s, and 150–180 s) to evaluate tumor vascularity and enhancement patterns. All imaging parameters followed standard clinical protocols and consensus recommendations for rectal cancer MRI to ensure diagnostic quality and comparability.

MRI acquisition and image segmentation

MRI scans were independently assessed by two gastrointestinal radiologists, each possessing more than five years of experience in rectal cancer diagnosis. Tumor segmentation was performed using ITK-SNAP software (Scientific Resource Identifier: SCR_017341; official website: http://www.itksnap.org), with the radiologists manually delineating the tumor regions as regions of interest (ROIs). In this study, “disagreement” between the two reviewers was defined as the presence of differences in any of the following aspects: (1) discrepancies in the delineation of the primary tumor boundary or in the inclusion/exclusion of adjacent anatomical structures (e.g., rectal lumen or perirectal fat); (2) inconsistencies in the axial slice selected for ROI placement; (3) disagreement regarding the exclusion of non-viable tumor components, including necrotic, cystic, hemorrhagic areas, or major intratumoral vessels. When any such disagreement occurred, a senior radiologist with 30 years of diagnostic experience re-evaluated the images and provided the final decision. Regions of interest (ROIs) were measured on axial T1-weighted images (T1WI) and T2-weighted images (T2WI) for the tumor, obturator internus or piriformis muscles, and urine. Similar ROI measurements were obtained on both pre- and post-contrast axial fat-saturated T1-weighted scans, with urine excluded. ROIs were systematically placed on the axial slice demonstrating the maximal tumor cross-section, with strict exclusion of necrotic, cystic, or hemorrhagic components and major intratumoral vessels, and only viable tumor tissue was included. For every imaging sequence per patient, corresponding ROIs were positioned at anatomically matched levels for both tumor and reference tissues to ensure measurement consistency. This standardized approach was uniformly applied across conventional sequences, DWI, and ADC maps. Urine ROIs were drawn at the base of the bladder. Tumor-to-muscle and tumor-to-urine signal intensity (SI) ratios were derived from T1- and T2-weighted images. For post-contrast sequences, enhancement ratios were determined by dividing the SI of the post-contrast scan by that of the corresponding pre-contrast scan for both tumor and muscle. For tumors with heterogeneous components (e.g., mixed mucinous and solid areas), a whole-tumor ROI was delineated to encompass the entire visible tumor boundary, while strictly avoiding the rectal lumen and adjacent perirectal fat [28].

Additional baseline MRI tumor features were evaluated independently by two gastrointestinal radiologists, each with more than five years of rectal cancer experience. To verify measurement accuracy, a randomly selected subset of cases was reassessed by a senior radiologist with three decades of diagnostic expertise. The evaluated MRI features included:（1）Mesorectal fascia (MRF):Tumor involvement of the MRF was defined as the presence of tumor, malignant lymph nodes, tumor deposits, or tumor-invaded vessels within 1 mm of the mesorectal fascia on MRI, or any direct invasion beyond the fascia. (2）Extramural venous invasion（EMVI）:EMVI was considered present when tumor signal was observed within vessels beyond the muscularis propria [29].

Models construction and validation

Potential risk factors were first screened and evaluated using univariate logistic regression, with variables meeting the criterion of p < 0.05 included for further analysis. Multivariate logistic regression was then performed based on the selected factors. Clinical and imaging feature models were constructed separately using R software (RRID: SCR_001905; version 4.3.2, https://www.r-project.org/). Covariate diagnostics were conducted for variables in each model. Finally, variables with p < 0.05 in both models were integrated to develop a nomogram.

In addition, The discriminatory power of each model was evaluated using receiver operating characteristic (ROC) curves. AUC values were compared across models with the DeLong test. Model calibration was assessed with the Hosmer–Lemeshow goodness-of-fit test, and calibration curves were generated to examine the agreement between predicted and observed outcomes. Model performance was further evaluated by calculating sensitivity, specificity, and accuracy. Clinical utility was assessed using decision curve analysis (DCA). To validate model robustness, patients from Center 2 were used as an external validation cohort. Baseline characteristics of the two centers are summarized in Table 1, with no significant differences observed for most variables.

Table 1.

Baseline characteristics of patients in the training and validation cohorts

Variable	Training cohort (Center 1)	External validation cohort (Center 2)		Training cohort (Center 1)		External validation cohort (Center 2)
	(n = 221)	(n = 76)	P	NMAC (n = 160)	MAC (n = 61)	P	NMAC (n = 54)	MAC (n = 22)	P
Outcome, n (%)			0.822
NMAC	160 (72.40)	54 (71.05)
MAC	61 (27.60)	22 (28.95)
CEA (ng/ml)	4.75 (2.32, 11.62)	3.58 (2.28, 8.42)	0.447	3.55 (1.99, 7.19)	8.96 (5.25, 15.19)	< 0.001	3.30 (2.05, 8.75)	4.33 (2.33, 7.77)	0.705
CA199 (U/ml)	12.63 (8.41, 22.87)	13.95 (7.58, 23.43)	0.753	12.10 (7.87, 21.52)	15.15 (10.80, 27.05)	0.030	14.45 (7.71, 25.58)	12.85 (6.42, 20.18)	0.478
ALB（g/L）	39.50 (36.10, 42.40)	40.05 (37.50, 43.20)	0.099	39.25 (35.80, 42.18)	40.20 (37.20, 43.90)	0.033	41.55 (37.73, 43.30)	39.35 (37.52, 40.05)	0.173
PLT (109/L)	236.00 (194.00, 282.00)	239.00 (199.75, 296.25)	0.337	238.50 (194.00, 289.25)	236.00 (194.00, 277.00)	0.747	236.00 (197.50, 311.25)	239.00 (210.25, 273.75)	0.963
ANC (109/L)	3.73 (2.68, 4.96)	3.40 (2.82, 4.36)	0.146	3.76 (2.72, 5.51)	3.53 (2.55, 4.22)	0.058	3.45 (2.82, 4.20)	3.29 (2.85, 4.38)	0.950
ALC (109/L)	1.61 (1.24, 1.92)	1.52 (1.10, 1.86)	0.167	1.62 (1.23, 1.93)	1.61 (1.25, 1.84)	0.888	1.53 (1.08, 1.85)	1.50 (1.22, 1.85)	0.680
DWI	155.57 (108.00, 310.08)	132.21 (84.31, 295.97)	0.054	156.50 (109.48, 308.08)	142.00 (106.50, 345.00)	0.674	135.25 (92.11, 277.12)	120.55 (69.09, 337.97)	0.443
ADC	921.33 (785.00, 1115.00)	996.70 (826.08, 1364.33)	0.009	872.91 (772.95, 1012.89)	1115.00 (974.45, 1357.00)	< 0.001	938.20 (793.58, 1090.91)	1511.64 (1266.08, 2168.15)	< 0.001
Tumor-to-muscle SI ratio on T1WI	1.06 (0.96, 1.18)	1.04 (0.93, 1.15)	0.153	1.06 (0.98, 1.17)	1.07 (0.91, 1.22)	0.441	1.06 (0.93, 1.16)	1.03 (0.93, 1.11)	0.313
Tumor-to-urine SI ratio on T1WI	2.08 (1.74, 2.37)	2.17 (1.96, 2.56)	0.020	2.14 (1.91, 2.38)	1.86 (1.55, 2.25)	0.002	2.22 (1.92, 2.74)	2.15 (1.99, 2.41)	0.647
Tumor-to-muscle SI ratio on T2WI	3.23 (2.60, 4.50)	4.16 (3.15, 5.12)	< 0.001	2.90 (2.46, 3.60)	4.80 (3.69, 6.73)	< 0.001	3.60 (2.85, 4.85)	4.90 (3.90, 6.43)	0.005
Tumor-to-urine SI ratio on T2WI	0.44 (0.35, 0.62)	0.48 (0.42, 0.65)	0.016	0.39 (0.33, 0.50)	0.61 (0.49, 0.77)	< 0.001	0.47 (0.37, 0.58)	0.64 (0.46, 0.88)	0.002
The contrast-enhancement ratio of tumor	2.58 (2.24, 2.93)	2.78 (2.51, 3.29)	0.003	2.63 (2.27, 2.91)	2.52 (1.90, 3.15)	0.619	2.72 (2.50, 3.12)	3.22 (2.70, 3.68)	0.035
The contrast-enhancement ratio of muscle	1.42 (1.24, 1.56)	1.45 (1.27, 1.62)	0.217	1.42 (1.25, 1.55)	1.44 (1.24, 1.59)	0.810	1.48 (1.32, 1.63)	1.34 (1.26, 1.58)	0.262
Sex, n (%)			0.411			0.755			0.272
Male	87 (39.37)	34 (44.74)		64 (40.00)	23 (37.70)		22 (40.74)	12 (54.55)
Female	134 (60.63)	42 (55.26)		96 (60.00)	38 (62.30)		32 (59.26)	10 (45.45)
Location, n (%)			0.337			0.744			0.760
Upper	67 (30.32)	17 (22.37)		50 (31.25)	17 (27.87)		12 (22.22)	5 (22.73)
Middle	100 (45.25)	41 (53.95)		73 (45.62)	27 (44.26)		28 (51.85)	13 (59.09)
Lower	54 (24.43)	18 (23.68)		37 (23.12)	17 (27.87)		14 (25.93)	4 (18.18)
Size, n (%)			0.887			0.007			0.056
< 5 cm	130 (58.82)	44 (57.89)		103 (64.38)	27 (44.26)		35 (64.81)	9 (40.91)
≥5 cm	91 (41.18)	32 (42.11)		57 (35.62)	34 (55.74)		19 (35.19)	13 (59.09)
Age, n (%)			0.889			0.003			0.778
< 50	98 (44.34)	33 (43.42)		61 (38.12)	37 (60.66)		24 (44.44)	9 (40.91)
≥50	123 (55.66)	43 (56.58)		99 (61.88)	24 (39.34)		30 (55.56)	13 (59.09)
Tumor invasion, n (%)			0.773			< 0.001			0.010
TIT2	86 (38.91)	31 (40.79)		78 (48.75)	8 (13.11)		27 (50.00)	4 (18.18)
T3T4	135 (61.09)	45 (59.21)		82 (51.25)	53 (86.89)		27 (50.00)	18 (81.82)
Presence of lymph node metastasis, n (%)			0.884			0.003			0.051
NO	101 (45.70)	34 (44.74)		83 (51.88)	18 (29.51)		28 (51.85)	6 (27.27)
YES	120 (54.30)	42 (55.26)		77 (48.12)	43 (70.49)		26 (48.15)	16 (72.73)
Presence of distant metastasis, n (%)			0.587			< 0.001			0.160
NO	180 (81.45)	64 (84.21)		144 (90.00)	36 (59.02)		48 (88.89)	16 (72.73)
YES	41 (18.55)	12 (15.79)		16 (10.00)	25 (40.98)		6 (11.11)	6 (27.27)
Presence of peritoneal involvement, n (%)			0.921			< 0.001			0.010
NO	155 (70.45)	54 (71.05)		126 (79.25)	29 (47.54)		43 (79.63)	11 (50.00)
YES	65 (29.55)	22 (28.95)		33 (20.75)	32 (52.46)		11 (20.37)	11 (50.00)
EMVI, n (%)			0.125			0.799			0.514
NO	85 (38.64)	37 (48.68)		61 (38.12)	24 (40.00)		25 (46.30)	12 (54.55)
YES	135 (61.36)	39 (51.32)		99 (61.88)	36 (60.00)		29 (53.70)	10 (45.45)
MRF, n (%)			0.004			0.153			0.531
NO	100 (45.45)	49 (64.47)		77 (48.43)	23 (37.70)		36 (66.67)	13 (59.09)
YES	120 (54.55)	27 (35.53)		82 (51.57)	38 (62.30)		18 (33.33)	9 (40.91)

CEA, carcinoembryonic antigen; CA199, carbohydrate antigen 19–9; ALB, albumin; PLT, platelet count; ANC, absolute neutrophil count; ALC, absolute lymphocyte count; DWI, diffusion-weighted imaging; ADC, apparent diffusion coefficient; EMVI, extramural venous invasion; MRF, mesorectal fascia; MAC, mucinous rectal adenocarcinoma; NMAC, non-mucinous adenocarcinoma

Statistical analysis

Statistical analyses and figure generation were performed using IBM SPSS Statistics (RRID: SCR_019096; version 25.0; SPSS, Inc., Chicago, IL, USA) and GraphPad Prism (RRID: SCR_002798; version 9.5; GraphPad, San Diego, CA, USA). Continuous variables with a normal distribution were compared using independent-samples t tests and are presented as mean ± standard deviation. Non-normally distributed continuous variables were analyzed using the Mann–Whitney U test to assess intergroup differences, with results reported as median (interquartile range). Categorical variables were compared using the chi-square test or Fisher’s exact test, and are presented as counts (n) and percentages (%). All statistical tests were two-sided, and a p value < 0.05 was considered statistically significant.

Results

Patients

A total of 297 rectal cancer patients meeting the inclusion and exclusion criteria were enrolled from two clinical centers, including 83 patients with postoperative pathology-confirmed mucinous adenocarcinoma and 214 patients with non-mucinous adenocarcinoma. Patients from Center 1 (n = 221; 160 NMAC and 61 MAC) were assigned to the training cohort, while patients from Center 2 (n = 76; 54 NMAC and 22 MAC) were designated as the external validation cohort to assess the generalization of the model.

Clinical and MRI characteristics

Each patient enrolled in the study was classified as either MAC or NMAC based on postoperative pathology. In the training cohort, significant differences (p < 0.05) were observed between the two groups in terms of age, tumor size, depth of tumor invasion (T stage), presence of lymph node metastasis (N stage), distant metastasis (M stage), and peritoneal involvement. No significant differences (p > 0.05) were found between the groups regarding sex, tumor location, CEA, CA19-9, PLT, ANC, ALC, or ALB. Results of univariate and multivariate logistic regression analyses of baseline clinical characteristics in patients with MAC and NMAC are presented in Table 2.

Table 2.

Univariate and multivariate logistic regression analysis of clinical features

univariate logistic analysis			multivariate regression analyses
Characteristics	P	OR (95%CI)	Characteristics	P	OR (95%CI)
CEA (ng/ml)	0.150	1.01 (1.00 ~ 1.03)
CA199 (U/ml)	0.461	1.00 (1.00 ~ 1.01)
ALB（g/L）	0.394	1.00 (1.00 ~ 1.01)
PLT (109/L)	0.729	1.00 (1.00 ~ 1.00)
ANC (109/L)	0.085	0.87 (0.75 ~ 1.02)
ALC (109/L)	0.692	0.91 (0.58 ~ 1.43)
Sex
Male		Ref
Female	0.755	1.10 (0.60 ~ 2.02)
Age (year)			Age
< 50		Ref	< 50		Ref
≥50	0.003*	0.40 (0.22 ~ 0.73)	≥50	0.005*	0.37 (0.18 ~ 0.74)
Size			Size
< 5 cm		Ref	< 5 cm		Ref
≥5 cm	0.007*	2.28 (1.25 ~ 4.15)	≥5 cm	0.008*	2.63 (1.29 ~ 5.35)
Location
Upper		Ref
Middle	0.815	1.09 (0.54 ~ 2.20)
Lower	0.458	1.35 (0.61 ~ 2.99)
Tumor invasion			Tumor invasion
T1T2		Ref	T1T2		Ref
T3T4	< 0.001*	6.30 (2.82 ~ 14.10)	T3T4	0.002*	4.07 (1.65 ~ 10.00)
Presence of lymph node metastasis			Presence of lymph node metastasis
No			No		1.00 (Reference)
Yes	0.003*	2.58 (1.37 ~ 4.84)	Yes	0.195	1.64 (0.78 ~ 3.45)
Presence of distant metastasis			Presence of distant metastasis
No		Ref	No
Yes	< 0.001*	5.48 (2.70 ~ 11.12)	Yes	0.016*	2.76 (1.21 ~ 6.34)
Presence of peritoneal involvement			Presence of peritoneal involvement
No		Ref	No
Yes	< 0.001*	4.21 (2.24 ~ 7.93)	Yes	0.014*	2.51 (1.20 ~ 5.23)

*p < 0.05; OR, odds ratio; CI, confidence interval

MRI analysis demonstrated that patients with MAC exhibited significantly higher tumor-to-muscle SI ratio on T2WI, tumor-to-urine SI ratio on T2WI, and ADC values compared with those with NMAC (p < 0.05). In contrast, no significant differences were observed between the two groups in tumor-to-muscle SI ratio on T1WI, tumor-to-urine SI ratio on T1WI, the contrast-enhancement ratio of tumor and muscle, involvement of the MRF, or the presence of EMVI (p > 0.05). Results of the univariate and multivariate logistic regression analyses of MRI features are summarized in Table 3.

Table 3.

Univariate and multivariate logistic regression analysis of imaging features

univariate logistic analysis			multivariate regression analyses
Characteristics	P	OR (95%CI)	Characteristics	P	OR (95%CI)
DWI	0.246	1.00 (1.00 ~ 1.00)
ADC	< 0.001*	1.01 (1.01 ~ 1.01)	ADC	< 0.001*	1.01 (1.01 ~ 1.01)
Tumor-to-muscle SI ratio on T1WI	0.259	0.46 (0.12 ~ 1.77)
Tumor-to-urine SI ratio on T1WI	0.082	0.65 (0.39 ~ 1.06)
Tumor-to-muscle SI ratio on T2WI	< 0.001*	2.34 (1.79 ~ 3.06)	Tumor-to-muscle SI ratio on T2WI	< 0.001*	1.88 (1.43 ~ 2.49)
Tumor-to-urine SI ratio on T2WI	< 0.001*	77.15 (14.24 ~ 417.95)	Tumor-to-urine SI ratio on T2WI	0.030*	6.77 (1.21 ~ 37.96)
The contrast-enhancement ratio of tumor	0.450	1.13 (0.82 ~ 1.55)
The contrast-enhancement ratio of muscle	0.503	0.79 (0.39 ~ 1.59)
EMVI
No		Ref
Yes	0.799	0.92 (0.50 ~ 1.70)
MRF
No		Ref
Yes	0.154	1.55 (0.85 ~ 2.84)

*p < 0.05; OR, odds ratio; CI, confidence interval

Performance of clinical and imaging models

Univariate and multivariate logistic regression analyses were performed on clinical and imaging parameters to construct two predictive models: a clinical model and an imaging model. The clinical model included age, tumor size, depth of tumor invasion, presence of peritoneal involvement, and presence of distant metastasis. The imaging model incorporated tumor-to-muscle SI ratio on T2WI, tumor-to-urine SI ratio on T2WI, and ADC value. In the training cohort, the AUCs of the clinical and imaging models were 0.826 (95% CI:0.765~0.887) and 0.877 (95% CI:0.822~0.931), respectively; in the external validation cohort, the corresponding AUCs were 0.726 (95% CI:0.599~0.853) and 0.830 (95% CI:0.730~0.931). Figure 2 illustrates the ROC curves of the two models.

Fig. 2.

The ROC curves of the clinical model, and imaging model in the training cohort (a) and the external validation cohort (b)

Construction and evaluation of the nomogram

Variables with p < 0.05 in the clinical and imaging models were used to construct a nomogram, including age, tumor size, depth of tumor invasion, presence of peritoneal involvement, presence of distant metastasis, tumor-to-muscle SI ratio on T2WI, tumor-to-urine SI ratio on T2WI, and ADC value. The resulting nomogram is shown in Fig. 3.

Fig. 3.

Nomogram based on clinical and imaging features

The nomogram demonstrated an AUC of 0.937（95% CI:0.894~0.979） in the training cohort and 0.882（95% CI:0.793~0.971） in the external validation cohort (Fig. 4). DeLong’s test demonstrated that the nomogram outperformed both the clinical and imaging models in terms of AUC in both cohorts (p < 0.05). Hosmer–Lemeshow tests indicated good calibration in the training cohort (p = 0.236) and external validation cohort (p = 0.148), which was further confirmed by calibration curves (Fig. 5).

Fig. 4.

ROC curves of nomogram in training cohort and external validation cohort. (a) ROC curve of nomogram in the training cohort. (b) ROC curve of nomogram in the external validation cohort

Fig. 5.

Calibration curves of nomogram in training cohort and external validation cohort. (a) calibration curve of nomogram in the training cohort. (b) calibration curve of nomogram in the external validation cohort

Additionally, decision curve analysis (DCA) of the clinical model, imaging model, and nomogram is presented in Fig. 6. The nomogram yielded the highest net benefit, followed by the imaging model. The accuracy, specificity, and sensitivity of all models are summarized in Table 4.

Fig. 6.

The decision curve analysis for all models in training cohort (a) and external validation cohort (b)

Table 4.

Diagnostic performance of each model in the training and validation cohorts

Cohort	Models	AUC	Accuracy	Specificity	Sensitivity
Training cohort	Clinical model	0.826(95% CI:0.765~0.887)	0.755	0.736	0.803
	Imaging feature model	0.877(95% CI: 0.822～0.931)	0.846	0.875	0.770
	Nomogram	0.937(95% CI: 0.894～0.979)	0.909	0.925	0.869
External validation cohort	Clinical model	0.726(95% CI: 0.599～0.853)	0.645	0.611	0.727
	Imaging feature model	0.830(95% CI: 0.730～0.931)	0.803	0.852	0.682
	Nomogram	0.882(95% CI: 0.793～0.971)	0.842	0.833	0.864

AUC, area under the curve; CI, confidence interval

Discussion

In this study, models were developed based on baseline clinical characteristics and MRI features to differentiate MAC from NMAC. Among them, the combined model incorporating both clinical and MRI features demonstrated the best performance, achieving AUCs of 0.937 and 0.882 in the training and validation cohorts, respectively, and exhibited good accuracy and feasibility. In a previous study, Ge et al. [30]demonstrated that CT-based radiomic features could effectively differentiate between MAC and NMAC preoperatively, achieving an AUC of 0.93 in the primary cohort with maintained strong performance in external validation. Distinct from their CT-based approach, our study incorporated clinical baseline characteristics with MRI-based imaging features, using MRI rather than CT as the primary imaging modality. Previous studies have shown that MRI outperforms CT in assessing organ invasion and tumor staging [31].

In this study, we first developed a model based on baseline clinical characteristics. The clinical model achieved AUCs above 0.7 in both the training and external validation cohorts; however, it exhibited relatively low specificity, at 0.736 in the training cohort and 0.611 in the external validation cohort. Decision curve analysis (DCA) also suggested limited clinical benefit. An imaging model based on MRI features was then constructed. Compared with the clinical model, the imaging model demonstrated higher AUC, accuracy, and specificity; however, its sensitivity was lower than that of the clinical model. To improve diagnostic performance, variables with p < 0.05 from both models were combined to create a simplified nomogram. This nomogram showed excellent performance, achieving an AUC of 0.937 in the training cohort, surpassing both the clinical and imaging models. Sensitivity, specificity, and overall accuracy all exceeded 0.80, and the model maintained strong performance in the external validation cohort, with an AUC of 0.882. In addition to AUC evaluation, DeLong’s test confirmed that the nomogram significantly outperformed the other two models (p < 0.05). Calibration curves indicated good agreement between predicted and observed outcomes in both cohorts, and DCA curves demonstrated that the nomogram provided the highest clinical benefit among the three models.

The nomogram was constructed based on eight relevant variables (p < 0.05) derived from both the clinical and imaging feature models. The imaging-related factors included tumor-to-muscle SI ratio on T2WI, tumor-to-urine SI ratio on T2WI, and ADC value. In this study, standardized relative quantitative parameters were employed to minimize inter-individual and cross-equipment variability. We utilized both tumor-to-muscle and tumor-to-urine SI ratios as low-contrast and high-contrast references, respectively. This dual-reference strategy effectively addressed the limitations of approaches relying on a single reference tissue, thereby improving the accuracy and reliability of diagnostic performance. Our findings indicate that MAC exhibits high signal intensity on T2WI, whereas NMAC typically demonstrates intermediate signal intensity. This difference in signal characteristics can be attributed to the abundant mucin content within MAC, as high T2 signal intensity corresponds to large pools of intratumoral mucin [28]. The presence of T2-hyperintense mucin within a tumor on baseline MRI has been identified as an independent biomarker of poor prognosis and unfavorable treatment response [6, 12]. However, hyperintense areas on T2WI are not specific to MAC and may also be seen in other fluid-containing lesions, such as cysts, effusions, or necrotic tumors [32, 33]. Therefore, in our nomogram, we incorporated the ADC as an additional imaging feature. Our results showed that MAC exhibits a higher ADC value compared to NMAC, which can be explained by the presence of characteristic extracellular mucin pools in MAC. These pools reduce resistance to water molecule diffusion, in contrast to the densely packed tumor cells and fibrous stroma typically seen in NMAC, resulting in an elevated ADC value [33, 34].

In the nomogram, the clinical baseline feature model included several factors that were significantly associated with MAC (p < 0.05), including age, tumor size, tumor invasion depth, presence of peritoneal involvement, and occurrence of distant metastasis. Due to its molecular mutation-driven aggressiveness and the diffusion-promoting effect of its mucinous microenvironment, MAC tends to progress more rapidly and is more commonly observed in younger patients [35, 36]. Huang et al. reported that Chinese patients under 50 years of age have a higher risk of developing MAC [37]. The characteristic extracellular mucin pools in MAC are highly hydrophilic and form a gel-like matrix that facilitates expansile tumor growth, often resulting in larger primary lesions compared to NMAC [38]. Our findings indicate that MAC is often diagnosed at an advanced stage (T3 or T4). This may be attributed to its lower microvascular density and abnormal vascular architecture, which render lymphatic and hematogenous dissemination less detectable at early stages. Furthermore, MAC typically exhibits an exophytic growth pattern, expanding outward from the bowel lumen. As a result, it is less likely to cause luminal narrowing or obstruction in the early stages, leading to delayed symptom onset and diagnosis at a more advanced stage [5, 14, 39]. Additionally, the mechanical shearing effect of mucin may facilitate cancer cell detachment along tissue planes, promoting direct peritoneal dissemination and increasing the risk of peritoneal metastasis [35, 40]. Previous studies have confirmed that MAC exhibits a greater tendency for rapid invasion and dissemination compared to NMAC [5, 32, 41], resulting in a higher incidence of distant metastases—findings consistent with our study.

In this study, we constructed a simple and practical nomogram based on easily accessible clinical and imaging variables. By assigning point values to each relevant factor, the nomogram enhances the accuracy of distinguishing MAC from NMAC. This nomogram facilitates personalized treatment planning by accurately stratifying patients according to risk. Beyond preoperative risk assessment, this tool may also provide valuable guidance for follow-up strategies, postoperative surveillance, and personalized therapeutic decision-making, thereby potentially improving patient outcomes. Future studies are warranted to further validate this model in larger, multicenter, and prospective cohorts. In addition, integrating advanced imaging biomarkers or molecular biological markers may further enhance predictive accuracy and help to assess its practical utility across diverse clinical settings.

Several limitations of this study should be acknowledged. First, as with most retrospective studies, although data were collected from multiple institutions, the overall sample size remained relatively limited. Given the relatively low incidence of MAC, the number of MAC cases was inevitably small, which may limit the generalizability of the findings. Although the proposed nomogram demonstrated good performance in both the training cohort and the external validation cohort, further validation in larger, multicenter studies is warranted to confirm its robustness. Future studies with larger sample sizes and prospective cohort designs are needed to validate our results and to further evaluate the applicability of the model across a broader range of clinical settings. Second, the retrospective nature of this study may introduce potential selection bias. To mitigate this issue, strict inclusion and exclusion criteria were applied to minimize the impact of potential bias on the study results. Third, although MRI images were acquired using different scanners, which may have introduced variability in image quality, all included images met diagnostic requirements. Moreover, standardized image preprocessing procedures were employed to minimize the influence of inter-scanner differences on image analysis.

Conclusions

In summary, we developed and validated a simple and effective nomogram that integrates clinical baseline characteristics and MRI features to differentiate rectal mucinous adenocarcinoma from non-mucinous adenocarcinoma. This model serves as a valuable reference for clinicians in making timely diagnoses and formulating individualized treatment strategies.

Abbreviations

MAC

Mucinous rectal adenocarcinoma

NMAC

Non-mucinous adenocarcinoma

MRI

Magnetic resonance imaging

CT

Computed tomography

WHO

World Health Organization

CRT

Neoadjuvant chemoradiotherapy

pCR

Pathological complete response

MSI

Microsatellite instability

T2WI

T2-weighted images

T1WI

T1-weighted images

DWI

Diffusion-weighted imaging

ADC

Apparent diffusion coefficient

CEA

Carcinoembryonic antigen

CA 19–9

Carbohydrate antigen 19–9

PLT

Platelet count

ALB

Albumin

ALC

Absolute lymphocyte count

ANC

Absolute neutrophil count

PACS

Picture Archiving and Communication System

ROI

Regions of interest

SI

Signal intensity

MRF

Mesorectal fascia

EMVI

Extramural venous invasion

AUC

Area under the curve

ROC

Receiver operating characteristic

DCA

Decision curve analysis.

Author contributions

SZ-W and HY-W contributed to the design and implementation of the concept; ZP-W, KH, ML and MJ-X contributed to the collection of the data and discussion of the results; CY-Q, YC-M, JH-L and WT-F contributed to the analysis and interpretation of the data; HY-W contributed to the reviewing and editing of the final manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by the Key Technology Research and Development Program, China (2022YFC2408400).

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Ethical approval was obtained from the Ethics Committee of Shandong Provincial Hospital Affiliated to Shandong First Medical University, China (NO. 2024–652). In addition, patient informed consent was waived due to the retrospective nature of this study, which was under the permission of the Ethics Committee of Shandong Provincial Hospital Affiliated to Shandong First Medical University. All methods were carried out in accordance with the Declaration of Helsinki. We confirmed that all methods were performed in accordance with relevant guidelines and regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.