PathoGen-X: A Cross-Modal Genomic Feature Trans-Align Network for Enhanced Survival Prediction from Histopathology Images

Precise survival prediction is crucial for personalized cancer treatment, yet genomic data, which often outperforms pathology data as a predictor, remains costly and difficult to obtain. We introduce PathoGen-X, a cross-modal genomic feature translation and alignment network designed to enhance survival prediction using histopathology images. This deep learning framework combines genomic and imaging data during training but only requires imaging data at testing. PathoGen-X leverages transformer-based networks to align and translate image features into the genomic feature space, strengthening imaging data with genomic insights. Unlike other methods, PathoGen-X translates and aligns features without mapping them to a common latent space and requires fewer paired samples. Validated on the TCGA-BRCA, TCGA-LUAD, and TCGA-GBM datasets, PathoGen-X achieves high performance in survival prediction, highlighting the potential of enriched imaging models for accessible cancer prognosis.

Recommended citation: Krishna A., Kurian N. C., Patil A., Parulekar A. and Sethi A. (2024). PathoGen-X: A Cross-Modal Genomic Feature Trans-Align Network for Enhanced Survival Prediction from Histopathology Images. https://arxiv.org/abs/2411.00749
Download Paper