Predicting the Survival Period of Non-Small Cell Lung Cancer Based on Multi-modal Deep Learning

Non-small cell lung cancer (NSCLC) has a poor prognosis and a high mortality rate, making accurate prediction of patient survival crucial. Traditional methods for predicting cancer survival rely mainly on analysis of clinical and pathological features, but these cannot fully reveal the complexity of the disease, thereby limiting accuracy and reliability. Currently, deep learning techniques face several challenges in survival prediction, including underutilization of multimodal data, short survival prediction horizons, and inadequate prediction capabilities for short survival periods. To address these issues, a novel end-to-end multimodal fusion deep learning model called MMPLC-Net is proposed and constructed. This model utilizes convolutional neural networks to process various data modalities including clinical information, mRNA, miRNA, copy number variations, and DNA methylation, effectively extracting feature representations. By integrating these multimodal data feature representations, the model can accurately predict the survival of NSCLC patients. Experimental results demonstrate that on NSCLC datasets of squamous cell carcinoma and adenocarcinoma, the concordance index (C-index) of this survival prediction method improved from 0.718 to 0.748, and the integrated Brier score (IBS) decreased from 0.189 to 0.172, indicating its performance is significantly better than existing survival prediction networks.