Deep Learning-Based Hybrid CNN Architecture for Automated Medical Image Segmentation

Abstract

Medical image segmentation remains one of the most challenging and critical tasks in clinical diagnosis and treatment planning. Accurate delineation of anatomical structures and pathological regions from medical images such as MRI, CT scans, and histopathology slides requires robust computational methods capable of handling complex visual patterns, noise, and class imbalance. In this paper, we present a novel Hybrid Convolutional Neural Network (HCNN) architecture that integrates dense connectivity, attention mechanisms, and multi-scale feature fusion for precise segmentation of medical images. The proposed model builds upon the foundational encoder–decoder paradigm of U-Net while incorporating residual dense blocks at each encoding stage and dual attention gates at the skip connections to selectively emphasize diagnostically relevant features. A multi-scale feature pyramid module aggregates contextual information from different levels of abstraction, enabling the network to capture both fine-grained local details and broad semantic context simultaneously. We evaluate the proposed architecture on four publicly available benchmark datasets: ISIC 2018 (skin lesion segmentation), BraTS 2019 (brain tumor segmentation), DRIVE (retinal vessel segmentation), and ChestX-ray14 (chest pathology localization). Extensive experiments demonstrate that our HCNN outperforms state-of-the-art baselines including U-Net, ResU-Net, Attention U-Net, and Dense U-Net across all datasets, achieving an overall accuracy of 94.8%, a Dice coefficient of 0.9482, and an Intersection over Union (IoU) score of 0.9387 on the ISIC 2018 dataset. The model also exhibits strong generalization capability with minimal overfitting, attributed to our custom hybrid data augmentation strategy and label-smoothing regularization. These results confirm the clinical viability of the proposed framework and its potential for real-world deployment in computer-aided diagnosis systems.