Lifelong Learning based Disease Diagnosis on Clinical Notes

TL;DR

This paper introduces a lifelong learning approach for disease diagnosis from clinical notes, addressing catastrophic forgetting and enabling models to adapt to new diseases without losing previous knowledge, validated on a new benchmark.

Contribution

It proposes a novel lifelong learning framework with attention and memory mechanisms for disease diagnosis, and establishes the Jarvis-40 benchmark for clinical notes.

Findings

Achieves state-of-the-art performance on Jarvis-40

Effectively mitigates catastrophic forgetting in disease diagnosis

Demonstrates adaptability to sequential disease tasks

Abstract

Current deep learning based disease diagnosis systems usually fall short in catastrophic forgetting, i.e., directly fine-tuning the disease diagnosis model on new tasks usually leads to abrupt decay of performance on previous tasks. What is worse, the trained diagnosis system would be fixed once deployed but collecting training data that covers enough diseases is infeasible, which inspires us to develop a lifelong learning diagnosis system. In this work, we propose to adopt attention to combine medical entities and context, embedding episodic memory and consolidation to retain knowledge, such that the learned model is capable of adapting to sequential disease-diagnosis tasks. Moreover, we establish a new benchmark, named Jarvis-40, which contains clinical notes collected from various hospitals. Our experiments show that the proposed method can achieve state-of-the-art performance on the proposed benchmark.