MTSC2: Multi-task Survival Prediction on Colorectal CT Images with Clinical Information

Colorectal cancer (CRC) is a highly malignant tumor, and accurate survival prediction is crucial in determining the most suitable treatment options. Recent studies have demonstrated that deep learning-based survival models have the potential to outperform traditional radiomics-based models. However, the complexity and variation of the colorectal focal area make it challenging for current deep-learning models to effectively capture the survival prediction-related information. To tackle this challenge, we propose an end-to-end multi-task survival prediction model that incorporates clinical information and segments the region of interest (ROI) of preoperative computed tomography (CT) scans as an auxiliary task. Our model achieves a concordance index of 0.7275 that outperforms other existing methods for CRC survival prediction on our in-house dataset and can help improve the accuracy and efficiency of survival prediction for patients with colorectal cancer.