Multiparametric MRI-Based Integrated Analysis of Clinical, Radiomics, Deep Learning, and Machine Learning for Predicting Tumor Proliferation and Prognosis in Locally Advanced Rectal Cancer.

RATIONALE AND

OBJECTIVES

This study aimed to develop and validate a predictive model integrating clinical, radiomics, deep learning (DL), and machine learning (ML) from multiparametric magnetic resonance imaging (MRI) for predicting tumor cell proliferation status and prognosis in patients with locally advanced rectal cancer (LARC).

MATERIALS AND

METHODS

A total of 384 LARC patients from three centers (January 2016-August 2022) were retrospectively enrolled. Radiomics and DL features were extracted from multisequence MRI (T2WI, DWI, T1WI, and contrast-enhanced T1WI). Twelve ML algorithms across 107 combinations were used to construct models. The optimal algorithm combination for radiomics-DL (RDL) model construction was determined by the highest average area under the curve (AUC) across cohorts. The resulting predictive probability (RDL-score) was integrated with clinical risk features to construct a combined model, visualized as a nomogram. Model performance was evaluated using AUC, DeLong test, calibration curves, and decision curve analysis (DCA). The prognostic value of the nomogram for recurrence-free survival (RFS) was assessed by Kaplan-Meier analysis.

RESULTS

The Stepglm[both] + glmBoost combination achieved the highest predictive performance for tumor cell proliferation status. The nomogram integrating the RDL-score, maximal tumor thickness, and carcinoembryonic antigen (CEA) demonstrated excellent discrimination, with AUCs of 0.939, 0.884, and 0.895 in the training, internal validation, and external validation cohorts, respectively. Furthermore, the nomogram effectively stratified patients by RFS in all cohorts (all log-rank p < 0.01).

CONCLUSION

This multiparametric MRI-based nomogram integrating clinical data, radiomics, DL, and ML techniques demonstrated robust performance in predicting Ki-67 expression and stratifying prognosis in LARC patients, supporting personalized management in LARC.