Cardi-GPT v2: An Adaptive Expert ECG-Record Processing Chatbot

Interpreting and communicating electrocardiogram (ECG) findings are crucial yet challenging tasks in cardiovascular diagnosis, traditionally requiring significant expertise and precise clinical communication. This paper presents Cardi-GPT v2(hereafter referred to as Cardi-GPT2), an enhanced version of our previously proposed expert system [1], by introducing a transformer-based model extension and an adaptive feedback mechanism to improve both ECG interpretation and clinician interaction. By augmenting an 8 -layer CNN backbone with a transformer encoder, Cardi-GPT2 achieves a weighted accuracy of 0.7508 across 24 cardiac conditions in processing 12-lead ECG data. A novel fuzzification layer translates predictions into intuitive linguistic terms, while a conversational chatbot facilitates interactive explanations and allows real-time expert feedback. The system dynamically updates its knowledge base, fuzzification logic, and model weights through continual learning, employing Elastic Weight Consolidation (EWC) to preserve prior knowledge during feedback-driven fine-tuning. Evaluated on a diverse dataset spanning six hospitals across four countries, Cardi-GPT2 demonstrates superior diagnostic performance and a chatbot response quality score of $\mathbf{7 8. 4 \%}$. By bridging the gap between intricate ECG data interpretation and actionable clinical insights, Cardi-GPT2 represents a transformative innovation in cardiovascular healthcare, promising to improve diagnostic accuracy, clinical workflows, and patient outcomes across diverse medical settings.