Prediction of Lymphovascular Invasion in Rectal Cancer Using Deep Learning Models Based on Multi-Parametric MRI

Motivation: Lymphovascular invasion (LVI) of rectal cancer is an independent risk factor for poor prognosis. However, achieving an accurate preoperative diagnosis using MRI remains challenging. Goal(s): A deep learning model was constructed based on multi-parameter MRI to accurately predict the LV1 status of rectal cancer before surgery. Approach: The largest tumor layer and its upper and lower layers were selected as input for the deep learning network to construct the DW1-DL, T2-FS-DL, T1CE-DL, and combined-DL models, followed by external validation. Results: All models demonstrated strong predictive performance, with the combined-DL model achieving the highest AUC(0.878~0.971). Impact: This study enhances preoperative diagnosis of lymphovascular invasion in rectal cancer using deep learning and multi-parameter MRI, leading to potential improved treatment strategies, reduced unnecessary surgeries, and better patient outcomes.