Interpreting clinical latent representations using autoencoders and probabilistic models

作者：

Highlights：

• Unsupervised methodology for interpreting clinical latent representations by using Gaussian Mixture Models and Autoencoders.

• Discovering clinical patterns through latent representations, hierarchical clustering and the Kullback Leibler Divergence.

• Pattern visualization and identification of the most significant clinical features associated with clinical conditions.

• Our approach was validated on groups of healthy and chronic patients, characterized both through diagnoses and drugs codes.

摘要

•Unsupervised methodology for interpreting clinical latent representations by using Gaussian Mixture Models and Autoencoders.•Discovering clinical patterns through latent representations, hierarchical clustering and the Kullback Leibler Divergence.•Pattern visualization and identification of the most significant clinical features associated with clinical conditions.•Our approach was validated on groups of healthy and chronic patients, characterized both through diagnoses and drugs codes.

论文关键词：Autoencoder,Learning latent representations,Gaussian mixture model,Clustering,Chronic diseases,Electronic health records

论文评审过程：Received 5 April 2021, Revised 29 October 2021, Accepted 1 November 2021, Available online 9 November 2021, Version of Record 16 November 2021.

论文官网地址：https://doi.org/10.1016/j.artmed.2021.102211