Purpose <p>Cardiovascular disease (CVD) remains a leading cause of global mortality that emphasizes the need for accurate and automated diagnostic models. This research aims to design a novel Cardio-EPNet for classifying cardiac MRI images. The key research question focuses on whether the proposed dual-path network integrating edge and pixel awareness can enhance classification accuracy and segmentation precision.</p> Methods <p>The proposed Cardio-EPNet employs an Adaptive Unsharp Mask Guided Filter (AUMGF) for removing noisy artifacts from the cardiac MRI images. The proposed Cardio-EPNet integrates an edge-aware pathway and feature-aware pathway for efficient classification of CVD classes. The edge-aware pathway uses Multi-attention-based U-Net (MAU-net) to precisely segment the left ventricular (LV) boundaries. The feature-aware pathway utilizing a Pixel-Level ResNeSt to capture spatial and textural features from both noise-free and segmented images. The dual feature representations are fused and processed in the Multi-layered perceptron to categorize the MRI images into three output stages: Normal controls (NC), coronary artery disease (CAD), and peripheral artery disease (PAD).</p> Results <p>Experimental results demonstrates that the proposed Cardio-EPNet attains an overall accuracy 99.04% for the classification of CVD cases. The Dice and Jaccard indices confirm superior segmentation consistency. Moreover, the proposed Cardio-EPNet increases an accuracy by 0.54%, 1.06%, 0.82%, and 1.06% better than ML-Based Predictive Models, O-SBGC-LSTM, CNN, and MobileNet v2-based DNN respectively.</p> Conclusion <p>The proposed Cardio-EPNet effectively integrates edge and pixel-level feature learning for robust CVD detection. Its dual-path design offers a dependable framework for early and automated CVD detection that greatly improves diagnostic accuracy.</p>