Localization and Detection of Cardiovascular Diseases using Artificial Intelligence

Abstract

Cardiovascular diseases (CVDs) remain a leading global cause of death, with early diagnosis being critical to improving outcomes. Traditional electrocardiogram (ECG) analysis, especially in under-resourced settings like interior Sindh, Pakistan, often suffers from inaccuracies due to reliance on outdated, manual interpretation and limited-lead data. This research presents a smart diagnostic framework that leverages a hybrid Faster Region-based Convolutional Neural Networks (Faster-RCNNs) and Region Proposal Networks (RPN) for high-fidelity feature extraction and disease localization from 12-lead ECGs. Unlike existing models focused on 1- or 6-lead inputs, the proposed system processes full 12-lead data to segment waveform components (P, QRS, T waves), quantify intervals (PR, QT, RR), and detect abnormalities such as ST deviations and Corrected QT interval (QTc) prolongation. A key innovation of this work is the integration of Explainable AI (XAI) through a clinical rule-based inference engine—referred to as the Condition Combiner—which interprets extracted features using cardiology-derived expressions. This XAI layer emulates expert reasoning, providing transparent and clinically interpretable diagnoses. The system is capable of detecting and localizing over 55 cardiac conditions, including myocardial infarctions, arrhythmias, conduction blocks, and electrolyte imbalances. To the best of our knowledge, our model offers maximum possible heart related issues which amounts to 55 different cardiac diseases. The proposed system has the capability to localize heart disease as well, with decent accuracy. The model achieved a 91.2% Exact Match Accuracy and F1 scores 93% across all conditions, demonstrating strong generalization and clinical reliability. Designed with scalability in mind, the system offers an explainable, automated solution suitable for both hospital-grade deployments and low-infrastructure healthcare settings, advancing early detection of cardiac diseases with trust and interpretability.