Abstract : Cardiovascular diseases have long been a significant medical concern, and early and accurate identification is crucial for effective rehabilitation and treatment. Predicting heart disease promptly enables informed decision-making, reducing patient risks. We utilized the MIT-BIH Database to facilitate this process, containing 48 half-hour excerpts of two-channel ambulatory ECG recordings digitized at 360 samples per second per channel. Before further analysis, the data underwent preprocessing steps, including Wavelet Transformation for Baseline Correction and normalization. Additionally, we detected a PQRST waveform consisting of 290 samples, which was extracted and then converted to 360 Hz after peak detection. Next, we performed a Hybrid Time-Frequency analysis using Short Time Fourier Transformation (STFT) and Wigner-Ville distribution (WVD). This transformation process turned the 1D ECG recordings into a 2D time-frequency spectrum, providing a more comprehensive signal representation. We employed a ResNet50 classifier with a 2D architecture to classify the ECG signals. Three distinct cases were considered: normal (non-ectopic cardiac beat), arrhythmia (ventricular ectopic beat), and abnormal (unknown beat). Remarkably, the ResNet50 classifier achieved an impressive accuracy of 97.96% in accurately identifying the different ECG signal categories.