Prognostic prediction model for rectal cancer based on CMS subtype indicators and SHAP-based interpretable analysis.

Area of Science:

• Radiology
• Oncology
• Medical Imaging

Background:

• Rectal cancer (RC) prognosis prediction remains challenging.
• Accurate prediction of 3-year progression-free survival (PFS) is crucial for treatment planning.
• Integrating advanced imaging techniques and molecular subtypes can improve prognostic models.

Purpose of the Study:

• To develop an interpretable radiomics model for predicting 3-year PFS in RC.
• To integrate consensus molecular subtype (CMS) indicators and Shapley additive explanations (SHAP) analysis with ultra-high b-value diffusion-weighted imaging (DWI).
• To evaluate the model's performance and identify key predictive features.

Main Methods:

• Retrospective study of 153 RC patients with 3.0T DWI (b-values 0-3000 s/mm²).
• Radiomic features extracted from stretched exponential model parametric maps.
• Two radiomic scores (Radscore1: conventional features, Radscore2: CMS surrogate-related features) developed using LASSO-Cox regression; five prognostic models constructed.

Main Results:

• Radscore2 (AUC=0.787) outperformed Radscore1 (AUC=0.751) in predicting 3-year PFS in the validation cohort.
• Model 5 (clinical factors + Radscore2) achieved the highest AUC (0.823) with good calibration.
• SHAP analysis identified key features: α_logarithm_ngtdm_Contrast and α_gradient_firstorder_TotalEnergy.

Conclusions:

• An interpretable radiomics model integrating CMS-surrogate features and SHAP analysis demonstrates good performance for 3-year PFS prediction in RC.
• This approach enhances prognostic accuracy beyond conventional methods.
• The model offers valuable insights for personalized rectal cancer management.