TLDR: Cardi-GPT is an advanced expert system designed to improve electrocardiogram (ECG) interpretation and clinical communication. It uses a deep learning model, specifically a 16-residual-block convolutional neural network, to analyze 12-lead ECG data and classify 24 cardiac conditions with a weighted accuracy of 0.6194. A unique fuzzification layer translates complex numerical predictions into easy-to-understand linguistic categories. The system also features an integrated chatbot interface, powered by a fine-tuned Large Language Model, which allows healthcare providers to interact with diagnostic insights and receive explanations. Evaluated on a diverse dataset from six hospitals across four countries, Cardi-GPT demonstrated strong performance and achieved an overall chatbot response quality score of 73%, aiming to make ECG diagnosis more accurate and accessible.
Interpreting electrocardiograms (ECGs) is a critical part of diagnosing heart conditions, but it often requires specialized expertise and precise communication. A new system called Cardi-GPT aims to make this process more efficient and understandable by combining advanced artificial intelligence with a user-friendly chatbot.
Developed by Koustav Mallick, Neel Singh, and Mohammedreza Hajiarbabi from Purdue University Fort Wayne, Cardi-GPT is designed to streamline how ECG data is analyzed and how those findings are communicated among healthcare providers. The core of the system is a powerful deep learning model, specifically a convolutional neural network (CNN) with 16 residual blocks. This CNN is trained to process comprehensive 12-lead ECG data, which provides a detailed view of the heart’s electrical activity. The model achieved a weighted accuracy of 0.6194 in classifying 24 different cardiac conditions, demonstrating its capability to identify various heart abnormalities.
One of Cardi-GPT’s innovative features is its “fuzzification layer.” This layer takes the complex numerical outputs from the deep learning model and translates them into clinically meaningful, easy-to-understand linguistic categories. Instead of just numbers, clinicians receive interpretations like “severe,” “high,” “medium,” “low,” or “negligible” for specific conditions. This makes the diagnostic insights much more intuitive and accessible, bridging the gap between intricate data and actionable clinical understanding.
Beyond interpretation, Cardi-GPT also includes an integrated chatbot interface. This chatbot allows healthcare professionals to interact with the system, query diagnostic insights, and receive explanations in natural language. It’s powered by a fine-tuned Large Language Model (LLM), specifically Google 1.5 Flash, enhanced with Retrieval Augmented Generation (RAG) to ensure its responses are grounded in medical and diagnostic knowledge. This interactive component fosters a more collaborative diagnostic process, allowing clinicians to gain transparency into the AI’s decisions and even provide feedback for continuous improvement.
The system was rigorously evaluated using a diverse dataset collected from six hospitals across four countries, ensuring its performance is robust across varied patient populations. In addition to the predictive model’s accuracy, the chatbot’s response quality was assessed using a comprehensive framework that measured coverage, grounding, and coherence. Cardi-GPT’s chatbot achieved an impressive overall response quality score of 73%, indicating its ability to provide comprehensive, relevant, and logically structured answers.
Cardi-GPT represents a significant step forward in cardiovascular healthcare. By automating and simplifying ECG interpretation, it promises to improve diagnostic accuracy, enhance clinical workflows, and ultimately lead to better patient outcomes, especially in areas where access to expert cardiologists might be limited. The developers envision future enhancements, including better grounding for chatbot responses, adaptive personalization based on patient-specific factors like medical history, and seamless integration with existing clinical tools such as electronic health records.
Also Read:
- EcoScale-Net: A Breakthrough in Lightweight ECG Classification for Cardiac Health
- HiMAE: A Compact AI Framework for Smartwatches That Deciphers Wearable Health Data
For more detailed information, you can refer to the full research paper: Cardi-GPT: An Expert ECG-Record Processing Chatbot.
