Cardi-GPT: An Expert ECG-Record Processing Chatbot
Koustav Mallick, Neel Singh, Mohammedreza Hajiarbabi · 2025
Read Cardi-GPT: An Expert ECG-Record Processing Chatbot by Koustav Mallick, Neel Singh, Mohammedreza Hajiarbabi (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-24737.txt.
Introduction
Koustav Mallick, Neel Singh, and Mohammedreza Hajiarbabi built Cardi-GPT, a chat tool that helps read heart trace data. Reading a twelve lead heart trace takes skill, and clear talk about the finds matters for care. Their system pairs a deep net with a chat front end for joint use.
The net uses sixteen blocks to scan trace data for twenty four heart states. A soft layer then turns raw scores into word grades that match how doctors speak. Tests used files from six sites in four lands, and the tool beat base models. A chat view lets staff ask about the finds in plain words.
Catalog record
| CREATOR | Koustav Mallick, Neel Singh, Mohammedreza Hajiarbabi |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-24737
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Cardi-GPT: An Expert ECG-Record Processing Chatbot Koustav Mallick, Neel Singh, Mohammedreza Hajiarbabi Abstract Interpreting and communicating electrocardiogram (ECG) findings are crucial yet challenging tasks in cardiovascular diagnosis, traditionally requiring significant expertise and precise clinical communication. This paper introduces Cardi-GPT, an advanced expert system designed to streamline ECG interpretation and enhance clinical communication through deep learning and natural language interaction. Cardi-GPT employs a 16-residual-block convolutional neural network (CNN) to process 12-lead ECG data, achieving a weighted accuracy of 0.6194 across 24 cardiac conditions. A novel fuzzification layer converts complex numerical outputs into clinically meaningful linguistic categories, while an integrated chatbot interface facilitates intuitive exploration of diagnostic insights and seamless communication between healthcare providers. The system was evaluated on a diverse dataset spanning six hospitals across four countries, demonstrating superior performance compared to baseline models. Additionally, Cardi-GPT achieved an impressive overall response quality score of 73%, assessed using a comprehensive evaluation framework that measures coverage, grounding, and coherence. By bridging the gap between intricate ECG data interpretation and actionable clinical insights, Cardi-GPT represents a transformative innovation in cardiovascular healthcare, promising to improve diagnostic accuracy, clinical workflows, and patient outcomes across diverse medical settings. I Introduction Electrocardiography (ECG) is an indispensable tool in diagnosing and monitoring cardiovascular diseases, with the 12-lead ECG configuration offering a comprehensive view of cardiac electrical activity. Its widespread use and significant clinical value make accurate and timely interpretation crucial for effective patient care [ 1 ] . Manual ECG interpretation, while reliable, is restricted by expert availability, time-intensive, error-prone, and challenged by growing patient data complexity. [ 2 ] . These challenges are further exacerbated in remote areas where access to doctors is limited, but paramedical staff are available. In such settings, an intelligent expert system, incorporating knowledge from medical professionals, can enable paramedics to assess disease severity and make informed decisions, including referring patients to the nearest medical center where doctors are available. Automated systems that integrate predictive modeling and expert knowledge have demonstrated significant potential to address these challenges. Predictive models, often leveraging machine learning algorithms, analyze vast datasets to identify patterns and enhance diagnostic accuracy [ 3 ] , while expert systems emulate human expertise through rule-based logic, providing consistent and explainable solutions. This dual approach holds potential for transforming diagnostic workflows, especially in resource-constrained environments. II Background II-A Electrocardiogram An electrocardiogram (ECG/EKG) is a diagnostic tool that records the heart’s electrical activity over time through electrodes placed on the body. The standard 12-lead configuration provides a comprehensive view of the heart’s electrical activity from multiple angles, making it invaluable for identifying cardiac abnormalities such as arrhythmias, myocardial infarction, and other cardiovascular conditions [ 4 ] . ECG data are typically represented as time series signals, with each lead capturing the activity of a specific part of the heart. Interpretation of these data requires specialized knowledge to recognize subtle patterns and deviations across multiple leads. Fig. 1 : ECG Signal. II-B Expert Systems Expert systems are computer-based tools designed to replicate the decision-making capabilities of human experts by combining data-driven algorithms with domain-specific knowledge. Traditional expert systems in ECG analysis are often rule-based, relying on predefined if-then logic to identify specific conditions [ 5 ] .These systems are limited by the complexity of encoding all possible scenarios into rule sets. II-C Neural Networks Neural networks, particularly deep learning models, have emerged as powerful tools for time series analysis, making them well-suited for interpreting ECG data. Convolutional Neural Networks (CNNs) and their variants have demonstrated high accuracy in single-lead ECG classification tasks [ 6 ] . Recent research emphasizes the potential of neural networks as expert systems, especially when trained on expert-labeled datasets. By combining neural networks with explainable AI techniques, these systems can provide accurate predictions and insights into the decision-making process [ 7 ] . III Related Works III-A Rule-Based Expert Systems Early attempts to automate ECG interpretation relied heavily on rule-based expert systems. These systems encoded diagnostic rules provided by domain experts to classify ECG signals into specific cardiac conditions [ 5 ] . For instance, simple rules might identify arrhythmia based on predefined thresholds for heart rate variability or waveform amplitude. Although effective in well-defined cases, traditional systems lacked adaptability, particularly when handling edge cases or noisy datasets. III-B Deep Learning in ECG Classification Deep learning has revolutionized ECG interpretation by enabling models to learn directly from data. More recently, deep learning approaches have become the standard, with Convolutional Neural Networks (CNNs) achieving state-of-the-art results in single-lead ECG classification tasks [ 8 ] . The transition from single-lead to 12-lead ECG analysis introduces additional complexity due to the spatial relationships between leads. To tackle these complexities, recent studies have focused on advanced architectures like residual networks and attention mechanisms. CNNs have shown great promise due to its ability to automatic feature extraction and robustness to spatial invariance. III-C Previous Works Several Works have shown good results. [ 7 ] have achieved an accuracy of overall accuracy of 87.33% over MIT-BIH Arrhythmia database. Zhang et. al. [ 9 ] have achieved an F1 Score of 0.813 over CPSC-2018 dataset [ 10 ] . Zhao et. al. [ 11 ] have achieved a 5-fold cross-validation score of 0.684 over a compiled dataset by PhysioNet [ 12 ] . On a same dataset, Bjorn [ 13 ] ran another model with a combinaiton of CNN and Rule-based system to achieve 0.54. These results set the benchmark for our experiment. IV Methodology Our proposed system, Cardi-GPT, integrates predictive modeling, fuzzification, and chatbot-based solution delivery to streamline the interpretation of 12-lead ECG data. IV-A Dataset We used the PhysioNet/Computing in Cardiology dataset [ 12 ] , consisting of 43,101 12-lead ECG recordings labeled with 27 diagnoses, collected from six hospitals across four countries spanning three continents. This demographic diversity addresses a key limitation of previous works. Three diagnosis pairs, having similar clinical implications but differing regional names, were merged, reducing the total classes to 24. Each recording is annotated with one of these 24 conditions. The dataset is imbalanced, with some conditions significantly underrepresented, and was de-noised for consistency. The recordings vary in frequency (257 Hz – 1000 Hz) and duration (6s – 60s), with a small subset (n=74) extending to 30 minutes. Each ECG entry includes patient age and gender. The dataset was stratified into training (72%), validation (18%), and test (10%) sets to ensure proportional class representation. The training set was used for model learning, the validation set for hyperparameter tuning and performance tracking, and the test set for final evaluation to assess generalization. IV-B Predictive Model The predictive model serves as the backbone of the system, analyzing 12-lead ECG data to classify cardiac conditions. IV-B 1 Preprocessing We preprocessed ECG data for standardization and reliable predictive analysis. First, All ECGs were resampled to the minimum frequency of 257 Hz. Then, signals were zero-padded to a uniform length of 4,096 samples, i.e., just under 16 seconds to accommodate the fixed input dimensions required by the convolutional neural network, preserving signal integrity. Likewise, we used a window with the length 4096 to randomly clip longer duration signals. Amplitude normalization was applied to mitigate variations caused by patient-specific factors, electrode placement, or device calibration, ensuring the model