Explainable Deep Learning for Disease Activity Prediction in Chronic Inflammatory Joint Diseases

ICML 2023 Workshop IMLH Submission57 Authors

Keywords: patient journey similarity, case-based explanations, disease activity prediction, attention mechanism

TL;DR: We propose an explainable deep-learning model to transform rheumatic disease patient journeys into comparable representations, predicting future disease activity and offering case-specific explanations.

Abstract:

Analysing complex diseases such as chronic inflammatory joint diseases, where many factors influence the disease evolution, is a challenging task.
We propose an explainable attention-based neural network model trained on data from patients with different arthritis subtypes for predicting future disease activity scores. The network transforms longitudinal patient journeys into comparable representations allowing for additional case-based explanations via computed patient journey similarities.
We show how these similarities allow us to rank different patient characteristics in terms of impact on disease progression and discuss how case-based explanations can enhance the transparency of deep learning solutions.

Submission Number: 57