Deep Learning for Early Cancer Detection: A Multi-Modal Approach Integrating Radiomics and Genomics

Abstract

Abstract—This research develops a fresh deep learning architecture that joins radiomics with genomics to achieve better
cancer finding together with classification functionalities. The research combines convolutional neural networks (CNNs)
together with transformer-based architectures for image processing of radiology pictures while adopting graph neural
networks (GNNs) with autoencoders to analyze genomic sequencing information. The developed multi-modal fusion
model processed TCGA plus TCIA public dataset information to reach 92.7% accuracy rates when predicting earlystage cancers throughout various types.