Preoperative Noninvasive Prediction of Tertiary Lymphoid Structures in Rectal Cancer Using MRI Radiomics and
Fan Zhang1, Rong Li1, Ling Wang2
1Gansu University of Chinese Medicine, Lanzhou, China (F.Z., F.M., F.L., L.P., X.Z.).
Rationale And Objectives:
Noninvasive preoperative assessment of tertiary lymphoid structures (TLSs) status is crucial for immunotherapy in rectal cancer, yet current methods remain invasive. This study aimed to develop and validate an interpretable model for noninvasive preoperative prediction of TLSs status in rectal cancer.
Methods:
In this retrospective study, 177 rectal cancer patients were stratified into training (n=124) and testing (n=53) cohorts by TLSs status. All analyses were performed on a per-patient basis, with one lesion delineated per patient on preoperative MRI. Radiomics features were selected using ICC, RFE, and LASSO. Radiomics, clinical, and combined models were compared using ROC, calibration, DCA, NRI, and IDI analyses, with SHAP and LIME for interpretation, and overall survival was assessed in the testing cohort. Radiomics model development was performed using FAE 0.5.9, and statistical analyses were conducted using Python 3.9.21 and R 4.3.1.
Results:
A total of 177 preoperative MRI examinations were analyzed. The median age of the included patients was 62 years (interquartile range, 53-67 years), and 112 patients (63.28%) were male and 65 (36.72%) were female. NLR and age were independent TLSs predictors. The combined model achieved AUCs of 0.866 and 0.851 in training and testing cohorts, outperforming single-modality models, with favorable calibration and clinical net benefit. RadScore was the primary positive contributor, while NLR and age exerted negative contributions. High-score patients had significantly longer overall survival.
Conclusion:
The proposed interpretable combined radiomics-clinical model enables accurate noninvasive preoperative TLSs prediction and prognostic stratification in rectal cancer.