HybridNet: Advancing MRI image quality using dense attention and deep learning

Magnetic Resonance Imaging (MRI) is widely used for brain disease diagnosis due to its superior tissue contrast, but low-field MRI scanners often generate low-resolution images that hinder accurate interpretation. Extending scan time or upgrading to high-field systems increases cost and patient discomfort, making them impractical solutions. Deep learning-based super-resolution has emerged as a promising alternative; however, conventional CNN and GAN-based models frequently oversmooth details, introduce artifacts, or rely on synthetic downsampling, thereby limiting their clinical reliability. To address these challenges, we propose HybridNet, a novel CNN-based framework that integrates even–odd pixel decomposition with Dense Attention Blocks (DABs) to capture both global structural consistency and fine anatomical details. The novelty of HybridNet lies in its dual-focus strategy pixel decomposition to preserve structural integrity and reduce aliasing, and dense attention integration to highlight diagnostically significant regions, resulting in sharper and more reliable MRI reconstructions. The objectives of this study includes — design a clinically reliable MRI super-resolution framework that minimizes artifacts and oversmoothing, to ensure generalization across different scanners and acquisition settings, and to provide perceptually faithful reconstructions that aid radiological interpretation without increasing scan time or hardware cost. Proposed HybridNet model is trained on real paired MRI datasets from different scanners, The result demonstrates strong robustness and diagnostic relevance in both Quantitative and qualitative aspects. Quantitative evaluations reveal an improvement of 6–15 dB in PSNR and 2%–4% in SSIM compared to state-of-the-art methods such as SRCNN, FSRCNN, VDSR, and EDSR. Ablation studies further confirm the role of pixel decomposition, attention integration, and feature fusion in enhancing performance. Radiologists’ subjective assessments also validate superior perceptual quality, achieving a Mean Opinion Score of 87.3 and a Perceptual Evaluation score of 73.35. Overall, HybridNet offers a cost-effective and clinically practical solution for enhancing MRI image quality.