Survival Prediction of gastric cancer Patients based on Deep Self-supervised Learning
Binwen Zhang · 2024
Globally, the incidence and mortality of gastric cancer remain high, posing a major threat to public health. Accurate prediction of the survival of gastric cancer patients is of great significance for improving treatment effects and improving patients' quality of life. Existing studies mostly rely on traditional statistical methods, which are often based on strict preset assumptions and are difficult to capture the complex non-linear relationships between patients' clinical and demographic characteristics and survival status. This results in limited prediction accuracy. Furthermore, there is a lack of effective processing methods for the unlabeled data present in patient datasets. In medical research and clinical practice, data acquisition is costly. Therefore, based on deep learning technology, this paper conducts research on improving the survival prediction accuracy and unlabeled data processing of gastric cancer patients, and proposes a survival prediction method that combines self-supervised learning and multi-task learning frameworks. This method effectively mines survival information in unlabeled data through a self-supervised pre-training module. Furthermore, the pre-trained encoder is embedded into the multi-task learning framework to achieve simultaneous prediction of patient survival status at multiple time points and accurately achieve personalized survival prediction. Experiments were conducted on the SEER data set, and the experimental results show that the proposed model is superior to other existing studies, improving prediction accuracy and effectively improving data utilization.