Predicting Recurrence in Locally Advanced Rectal Cancer Using Multitask Deep Learning and Multimodal MRI

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Artificial Intelligence in Medicine

Background:

• Locally advanced rectal cancer (LARC) treatment involves neoadjuvant chemoradiotherapy (nCRT), but predicting disease-free survival (DFS) remains challenging.
• Accurate prognostic prediction is crucial for tailoring treatment strategies and improving patient outcomes.

Purpose of the Study:

• To develop and validate a deep multitask network, MultiRecNet, for fully automatic prediction of DFS in nCRT-treated LARC patients.
• To evaluate the performance of MultiRecNet using multimodal MRI data and clinical information.

Main Methods:

• A retrospective study collected data from 445 LARC patients treated with nCRT across three centers.
• MultiRecNet was developed to perform simultaneous segmentation, classification, and survival prediction tasks.
• Multimodal MRI (T2, ADC, D_app, K_app) and clinical data were used as inputs.

Main Results:

• The best MultiRecNet model achieved a Dice similarity coefficient (DSC) of 0.72 for tumor segmentation.
• The model demonstrated high accuracy in classifying recurrence or metastasis at 3 years (AUC = 0.97) and predicting DFS (C-index = 0.92) in the internal test set.
• The model maintained strong performance for survival prediction in the external test set (C-index = 0.81).

Conclusions:

• The MultiRecNet-based model provides fully automated, end-to-end prognostic prediction for LARC patients post-nCRT.
• This deep learning approach shows significant potential for improving DFS prediction accuracy and guiding clinical decision-making.