Abstract

Motivation

The EchoNarrator project aims to bridge the gap between automated EF prediction and clinician interpretability. By providing natural language explanations for echocardiography insights, EchoNarrator enhances trust and transparency, supporting clinicians in making informed decisions.

Summary

EchoNarrator combines a Graph Convolutional Network (GCN) for EF prediction with a Large Language Model (LLM) to generate human-readable explanations. This integration produces accurate EF estimates alongside clear, text-based explanations for better clinical understanding of model predcitions.

Results

Our model achieves robust EF predictions with high accuracy and clinically relevant natural language explanations. Evaluations demonstrate improved interpretability and alignment with clinical expectations, highlighting EchoNarrator’s potential for real-world applications.

Article

The paper is accepted for publication in the 2024 International Conference on Medical Image Computing and Computer Assisted Intervention.

If you find the work interesting, please cite it:

Citation:

S. Thomas, Q. Cao, Daria K., Anna N., and G. Ben-Yosef, “Generating natural text explanations for ejection fraction predictions”, International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI). Springer, Cham, 2024.

Data

The paper relied on the EchoNet dataset which is available from Stanford here. We would like to thank the authors (Ouyang et. al, 2020) for making these resources available.

Code

The code is available on GitHub

Authors

Authors: Sarina Thomas¹, Qing Cao ², Daria Kulikova², Anna Novikova² and Guy Ben-Yosef^4,*

1: University of Oslo, Oslo, NO

2: GE Healthcare Ultrasound, Wuxi, China

3: GE Healthcare, Kharxiv, Ukraine

4: GE HealthCare Technology and Innovation Center, Niskayuna, New York, USA

*: Corresponding author: guy.ben-yosef@gehealthcare.com