The electrocardiogram (ECG) is a significant non-invasive diagnostic modality utilized for the interpretation and detection of a wide range of cardiac conditions. This study presents a novel Deep Learning (DL) methodology for the automated detection of Congestive Heart Failure (CHF) and Arrhythmia (ARR), achieving a elevated level of exactness while minimizing computing demands. This research paper presents a novel electrocardiogram (ECG) diagnostic algorithm that integrates the Convolutional Neural Network (CNN) by the Constant-Q Non-Stationary Gabor Transform (CQ-NSGT), marking the first instance of such an approach. The present study examines the CQ-NSGT technique as a means of converting the one-dimensional electrocardiogram (ECG) signal into a two-dimensional representation in the time-frequency domain. This transformed representation will subsequently be executed into a pre-trained convolutional neural network (CNN) representation known as AlexNet. The AlexNet architecture is engaged to remove features, which are subsequently utilized as discriminative features in a Multi-Layer Perceptron (MLP) method. This strategy aims to classify three distinct cases: Congestive Heart Failure (CHF), Atrial Fibrillation (ARR), and Normal Sinus Rhythm (NSR). The efficacy of the CQ-NSGT algorithm is demonstrated through a comparative analysis of the presentation of the projected Convolutional Neural Network (CNN) by CQ-NSGT and the CNN by Continuous Wavelet Transform (CWT). The efficacy of the proposed methodology is evaluated using authentic electrocardiogram (ECG) data. The empirical findings demonstrate the higher presentation of the projected approach in comparison to other established methodologies, as evidenced by its accuracy rate of 98.82%, sensitivity rate of 98.87%, specificity rate of 99.21%, and precision rate of 99.20%. This study provides evidence supporting the efficacy of the recommended approach in civilizing the accuracy of ECG diagnoses.