Artificial intelligence in multimodal data analysis for cancer survival prediction

Artificial intelligence (AI) has been introduced to meet the demand for more precise predictions of cancer patient survival by simultaneously interpreting multiple types of input data. These data can originate from a broad spectrum of modalities, where a modality refers to one of several distinct forms in which data can be represented or observed. Using AI, researchers have investigated how to model the interactions among different data modalities as a promising approach to multimodal fusion in cancer prognosis. The goal is to build models capable of reliably handling and integrating these heterogeneous data sources to improve the accuracy and interpretability of patient outcome predictions. Nonetheless, current integration techniques encounter two major obstacles that must be overcome before survival prediction can be carried out: data alignment and data fusion. This chapter surveys existing alignment methods and fusion strategies within a pipeline designed to predict cancer survival. The chapter begins by describing the search strategy followed to identify the relevant literature to be analyzed here. Then, multimodal machine learning is introduced with the three key data types investigated in this study. we review 31 recent multimodal studies related to alignment and fusion using deep learning, covering the period from 2015 to 2025. Among these, 18 studies focus on alignment techniques followed by fusion strategies, while another 13 investigate fusion strategies independently of alignment. This chapter illustrates how synchronizing the data modalities before applying fusion can improve both the performance and the interpretability of the models. The synthesis of these studies uncovers methodological challenges and suggests future directions to develop more effective and clinically relevant AI-based survival prediction frameworks.