Zero-Shot ECG Diagnosis with Large Language Models and Retrieval-Augmented Generation

Han Yu, Peikun Guo, Akane Sano
Proceedings of the 3rd Machine Learning for Health Symposium, PMLR 225:650-663, 2023.

Abstract

Recently, Large Language Models (LLMs) have become essential players in the deep learning domain. While their capabilities are evident across various textual tasks, this study aims to bridge the gap and explore the potential of leveraging LLMs in diagnosing cardiac diseases and sleep apnea from Electrocardiography (ECG). Earlier work touched on converting ECG signals into text for LLMs, but a comprehensive LLM-based approach for dealing with more complicated symptoms remains relatively unexplored. To investigate the ECG diagnosis with an LLM-based approach, our research introduces a zero-shot retrieval-augmented diagnosis technique. We have built databases filled with specific domain knowledge for cardiac symptom and sleep apnea diagnosis, which encourages the LLMs from merely relying on the inherent LLM knowledge to a more holistic pipeline from carefully crafting prompts and infusing expert knowledge to guide LLMs. We evaluate the proposed approach on two datasets for diagnosing arrhythmia and sleep apnea, respectively. The evaluation results indicate that our zero-shot approach not only surpasses previous few-shot LLM-based methods but is also competitive with supervised learning techniques fully trained on extensive datasets.

Cite this Paper

BibTeX
@InProceedings{pmlr-v225-yu23b, title = {Zero-Shot ECG Diagnosis with Large Language Models and Retrieval-Augmented Generation}, author = {Yu, Han and Guo, Peikun and Sano, Akane}, booktitle = {Proceedings of the 3rd Machine Learning for Health Symposium}, pages = {650--663}, year = {2023}, editor = {Hegselmann, Stefan and Parziale, Antonio and Shanmugam, Divya and Tang, Shengpu and Asiedu, Mercy Nyamewaa and Chang, Serina and Hartvigsen, Tom and Singh, Harvineet}, volume = {225}, series = {Proceedings of Machine Learning Research}, month = {10 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v225/yu23b/yu23b.pdf}, url = {https://proceedings.mlr.press/v225/yu23b.html}, abstract = {Recently, Large Language Models (LLMs) have become essential players in the deep learning domain. While their capabilities are evident across various textual tasks, this study aims to bridge the gap and explore the potential of leveraging LLMs in diagnosing cardiac diseases and sleep apnea from Electrocardiography (ECG). Earlier work touched on converting ECG signals into text for LLMs, but a comprehensive LLM-based approach for dealing with more complicated symptoms remains relatively unexplored. To investigate the ECG diagnosis with an LLM-based approach, our research introduces a zero-shot retrieval-augmented diagnosis technique. We have built databases filled with specific domain knowledge for cardiac symptom and sleep apnea diagnosis, which encourages the LLMs from merely relying on the inherent LLM knowledge to a more holistic pipeline from carefully crafting prompts and infusing expert knowledge to guide LLMs. We evaluate the proposed approach on two datasets for diagnosing arrhythmia and sleep apnea, respectively. The evaluation results indicate that our zero-shot approach not only surpasses previous few-shot LLM-based methods but is also competitive with supervised learning techniques fully trained on extensive datasets.} }
Endnote
%0 Conference Paper %T Zero-Shot ECG Diagnosis with Large Language Models and Retrieval-Augmented Generation %A Han Yu %A Peikun Guo %A Akane Sano %B Proceedings of the 3rd Machine Learning for Health Symposium %C Proceedings of Machine Learning Research %D 2023 %E Stefan Hegselmann %E Antonio Parziale %E Divya Shanmugam %E Shengpu Tang %E Mercy Nyamewaa Asiedu %E Serina Chang %E Tom Hartvigsen %E Harvineet Singh %F pmlr-v225-yu23b %I PMLR %P 650--663 %U https://proceedings.mlr.press/v225/yu23b.html %V 225 %X Recently, Large Language Models (LLMs) have become essential players in the deep learning domain. While their capabilities are evident across various textual tasks, this study aims to bridge the gap and explore the potential of leveraging LLMs in diagnosing cardiac diseases and sleep apnea from Electrocardiography (ECG). Earlier work touched on converting ECG signals into text for LLMs, but a comprehensive LLM-based approach for dealing with more complicated symptoms remains relatively unexplored. To investigate the ECG diagnosis with an LLM-based approach, our research introduces a zero-shot retrieval-augmented diagnosis technique. We have built databases filled with specific domain knowledge for cardiac symptom and sleep apnea diagnosis, which encourages the LLMs from merely relying on the inherent LLM knowledge to a more holistic pipeline from carefully crafting prompts and infusing expert knowledge to guide LLMs. We evaluate the proposed approach on two datasets for diagnosing arrhythmia and sleep apnea, respectively. The evaluation results indicate that our zero-shot approach not only surpasses previous few-shot LLM-based methods but is also competitive with supervised learning techniques fully trained on extensive datasets.
APA
Yu, H., Guo, P. & Sano, A.. (2023). Zero-Shot ECG Diagnosis with Large Language Models and Retrieval-Augmented Generation. Proceedings of the 3rd Machine Learning for Health Symposium, in Proceedings of Machine Learning Research 225:650-663 Available from https://proceedings.mlr.press/v225/yu23b.html.

Related Material