Cardiovascular diseases (CVDs) remain a leading cause of global mortality, necessitating accurate and explainable diagnostic systems. In this paper, we propose a novel Autoencoder-Augmented Neural Network (AANN) framework for multi-class classification of cardiovascular conditions from 2D ECG images. The model leverages an unsupervised autoencoder to extract compact and discriminative latent features, followed by a lightweight classifier for disease prediction. We evaluate the framework on the publicly available National Heart Foundation 2023 ECG dataset, achieving a validation accuracy of 99.65%. To enhance transparency and interpretability, we employ Gradient-weighted Class Activation Mapping (Grad-CAM) and Testing with Concept Activation Vectors (TCAV). These explainable AI techniques allow for both pixel-level and concept-level insights into the model’s decision-making process. The experimental results demonstrate not only high diagnostic performance but also strong interpretability, making the proposed method suitable for deployment in clinical decision support systems.