Accurate survival prediction in oncology is critical for informing treatment strategies and improving patient outcomes. Traditional statistical models like the Cox proportional hazards model often underperform when applied to complex, high-dimensional datasets such as genomics and histopathology. This review explores the emergence of artificial neural networks (ANNs), particularly deep learning architectures, as advanced tools for survival analysis. This paper highlights multiple ANN-based models across cancer types—including breast, lung, and brain—that utilize multimodal data inputs such as gene expression, histopathology images, and longitudinal clinical records. Methods discussed include fully connected neural networks, convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and ensemble learning strategies. This paper further examines architectures such as PASNet and MDNNMD, and models trained on datasets like TCGA. Despite improved performance over conventional methods, challenges persist in clinical adoption due to interpretability, generalizability, and data privacy. Future prospects include expert systems, domain adaptation, federated learning, and multimodal integration. This paper concludes that while ANN-based survival models show great promise, interdisciplinary efforts are required to translate these innovations into scalable, secure, and explainable clinical applications.
Jiayi Yan (Wed,) studied this question.