Reproduce “Robust Multimodal Survival Prediction with the Latent Differentiation Conditional Variational AutoEncoder” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Junjie Zhou, Jiao Tang, Yingli Zuo, Peng Wan, Daoqiang Zhang, Wei Shao

Researchers at Nanjing University of Aeronautics and Astronautics developed the Conditional Latent Differentiation Variational AutoEncoder (LD-CVAE) for robust multimodal cancer survival prediction, specifically addressing scenarios where genomic data is unavailable during inference. This model achieved superior C-index values, improving performance by 1.0% to 3.9% over state-of-the-art methods on several TCGA cancer datasets, while effectively generating functionally diverse genomic features from histopathological images.

Key insights

  • The VIB-Trans effectively balances WSI compression with relevance to survival prediction, enabling robust conditioning for genomic reconstruction from massive image data.
  • Latent differentiation within the LD-VAE allows for the generation of functionally diverse genomic embeddings, better reflecting the heterogeneous biological roles of genes.
  • Combining Product-of-Experts and a Wasserstein distance-based alignment loss provides a robust mechanism to learn joint multimodal latent distributions, crucial for effective inference when one modality is missing.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.