AI-Enhanced ECG Diagnosis System for Myocardial Infarction with RBBB: Constant-Q Transform and ResNet50 Integration

Myocardial infarction, more commonly referred to as heart attack, is a significant cause of mortality related to cardiovascular diseases. Conduction disorders are commonly associated with this condition, the most prominent being RBBB. RBBB refers to an anomaly in electrical conduction of the heart, which can cause distortions to standard ECG patterns. Both of these can mask and mimic the classical ECG signs of MI and therefore result in misdiagnosis or delayed diagnosis. Moreover, early and accurate diagnosis is very important in saving patients from MI with a good prognosis; hence, it becomes a prime concern for the clinician. In this work, a novel approach to the accurate classification of ECG signals has been proposed with a Q-transform deep learning model of horizontal data concatenation. This study focuses essentially on the differentiation of myocardial infarction from RBBB-associated myocardial infarction. Further, the proposed model uses collective information from various ECG leads, drastically improving its capability to capture intricate cardiac patterns. In addition, the proposed model harnesses the unique electrical signatures of MI and MI with RBBB, which may manifest differently between leads. By merging multi-lead data and spectral-temporal features, the proposed model gains a comprehensive understanding of these conditions and leads thus to a substantial improvement in diagnostic accuracy. However, publicly available digital PTB-XL datasets are also used for the evaluation of the suggested architecture, where the ECGs are categorized into two classes: MI and MI associated with RBBB. In this regard, this system demonstrates exceptional performance, achieving an impressive 97. 82% precision and an exceptionally low 0.0032% training loss after 100 trained epochs. Stringent 10-fold cross-validation reinforces and strengthens these results. This groundbreaking approach simplifies diagnostic complexities by consolidating 12-lead ECG data and using CQT for precise analysis in the time-frequency domain.