The Continuous Wavelet Transform (CWT) and a modified VGG19 deep learning model are used in this manuscript to create a novel method for ECG categorization. To extract time-frequency information and create scalograms that graphically depict the signal’s energy distribution, the suggested procedure starts with the recording of ECG signals. The model is trained using a dataset of 8,092 scalogram images that are split equally between normal and pathological categories. Training takes up 70% of the dataset, validation takes up 10%, and testing takes up 20%. By utilizing transfer learning from VGG19 in conjunction with a bespoke prediction layer, the classifier attains remarkable training and validation accuracies of 97.97% and 97.32%, respectively. Metrics like the confusion matrix, classification report, ROC curve, and other statistical studies are used to further assess the performance and show how resilient the model is. Furthermore, the model’s generalizability is confirmed by tenfold cross-validation findings, which show advantages above state-of-the-art techniques. This method enhances practical applications in ECG analysis and interpretation by demonstrating the possibility of efficient automated cardiac diagnosis.