A review of cancer data fusion methods based on deep learning

Abstract

With advancements in modern medical technology, an increasing amount of cancer-related information can be acquired through various means, such as genomics, proteomics, imaging, and pathology. However, these datasets come from diverse sources and possess heterogeneity and complexity in terms of data types, formats, and quality, which pose challenges for cancer diagnosis, treatment, and prognosis evaluation. Data fusion has emerged as an effective approach to address data heterogeneity and enhance data information value. Integrating cancer data of different origins and types can improve diagnostic accuracy and deepen our profound understanding of cancer. Currently, data fusion methods based on deep learning have gained considerable attention in cancer research. This study presents a comprehensive review of the most recent research and development trends of deep learning-based data fusion in cancer, with emphasis on the advancements of various data fusion methods based on heterogeneous data types (including specific methodologies, their pros, and cons), which offer substantial support for enhancing the precision of diagnosing and treating cancer. Furthermore, we present an overview of prevalent cancer data types and fusion approaches and analyze the general modeling methodologies based on deep learning. We further discuss the challenges and future directions, aiming to provide assistance and guidance for researchers endeavoring to devise deep learning solutions in the sphere of cancer research.