Temporal Convolutional Neural Networks for Diagnosis from Lab Tests

TL;DR

This paper presents a multi-resolution convolutional neural network that leverages temporal lab test data to improve early disease diagnosis, demonstrating significant predictive improvements over traditional methods.

Contribution

Introduces a novel neural network architecture that combines imputed data and observation matrices for early disease detection from irregular lab tests.

Findings

Significantly outperforms baseline methods in disease prediction accuracy.

Effective handling of irregular, sparse temporal lab data.

Validated on a large dataset of 298,000 individuals over 8 years.

Abstract

Early diagnosis of treatable diseases is essential for improving healthcare, and many diseases' onsets are predictable from annual lab tests and their temporal trends. We introduce a multi-resolution convolutional neural network for early detection of multiple diseases from irregularly measured sparse lab values. Our novel architecture takes as input both an imputed version of the data and a binary observation matrix. For imputing the temporal sparse observations, we develop a flexible, fast to train method for differentiable multivariate kernel regression. Our experiments on data from 298K individuals over 8 years, 18 common lab measurements, and 171 diseases show that the temporal signatures learned via convolution are significantly more predictive than baselines commonly used for early disease diagnosis.