Towards modelling hazard factors in unstructured data spaces using gradient-based latent interpolation

TL;DR

This paper introduces a novel deep learning approach combining a multi-task variational autoencoder and HazardWalk to model hazard factors directly in unstructured data spaces, enhancing interpretability in survival analysis.

Contribution

It proposes a new method for modeling hazard factors in unstructured data using a multi-task VAE and HazardWalk, improving interpretability of deep survival models.

Findings

Effective on simulated data and CT imaging data of liver metastases.

Provides hazard factor insights in unstructured data spaces.

Enhances interpretability of deep survival analysis models.

Abstract

The application of deep learning in survival analysis (SA) allows utilizing unstructured and high-dimensional data types uncommon in traditional survival methods. This allows to advance methods in fields such as digital health, predictive maintenance, and churn analysis, but often yields less interpretable and intuitively understandable models due to the black-box character of deep learning-based approaches. We close this gap by proposing 1) a multi-task variational autoencoder (VAE) with survival objective, yielding survival-oriented embeddings, and 2) a novel method HazardWalk that allows to model hazard factors in the original data space. HazardWalk transforms the latent distribution of our autoencoder into areas of maximized/minimized hazard and then uses the decoder to project changes to the original domain. Our procedure is evaluated on a simulated dataset as well as on a dataset of CT imaging data of patients with liver metastases.