HEART: Heart Expert Assistant with ReTrieval-augmented
Track: Traditional track
Keywords: Clinical Foundation Models, Cardiovascular diseases prediction, Large language model, Retrieval-Augmented model
Abstract: As the incidence of cardiovascular diseases continues to rise, people are increasingly emphasizing the prevention and treatment of cardiovascular diseases. However, in economically disadvantaged areas, the scarcity of medical resources and lack of clinical experience make the early detection of cardiovascular diseases particularly challenging. For this challenge, the HEART (HEART Expert Assistant with Retrieval-augmented) model was proposed, which leverages the powerful logical reasoning capabilities of Large Language Models (LLMs) to assess whether patients have heart disease. Specifically, HEART operates on a dual-component structure, consisting of a Diagnostic Module and a Case Retrieval Module. For the Diagnostic Module, the LLM is pre-trained on a cardiac ultrasound assessment dataset to master the relevant evaluation techniques. As for the Case Retrieval Module, a text encoder transforms input cases into hidden features, which are then used to retrieve auxiliary cases. The input case and auxiliary case are merged through a Case Fusion Layer to obtain the fused case features, which are combined with prompts for inference. We have tested our model on a congenital disease dataset and achieved encouraging results. The proposed HEART model has shown tremendous potential in becoming the foundational model for predicting cardiovascular diseases.
Presentation And Attendance Policy: I have read and agree with the symposium's policy on behalf of myself and my co-authors.
Ethics Board Approval: No, our research does not involve datasets that need IRB approval or its equivalent.
Data And Code Availability: Yes, we will make data and code available upon acceptance.
Primary Area: Clinical foundation models
Student First Author: Yes, the primary author of the manuscript is a student.
Submission Number: 2
Loading