A Hybrid Self-Supervised Denoising and Attention-Guided Segmentation Framework for Robust Medical Image Analysis

Abstract: Medical image analysis plays a fundamental role in modern clinical diagnosis and treatment planning. However, the presence of acquisition noise, low contrast, and indistinct anatomical boundaries often degrades image quality, thereby reducing the reliability of automated image analysis systems. Conventional denoising techniques are effective in suppressing noise but frequently compromise fine structural details that are essential for accurate medical interpretation. Similarly, segmentation models trained on degraded images often struggle to identify complex anatomical structures and pathological regions with high precision. To address these challenges, this paper proposes a hybrid medical image processing framework that integrates a self-supervised image denoising network with an attention-guided U-Net segmentation architecture. The denoising stage is designed to remove noise while preserving structural information and edge continuity, producing high-quality intermediate representations for subsequent analysis. The restored images are then processed by an attention-enhanced segmentation network that selectively focuses on diagnostically significant regions while suppressing irrelevant background features. This integrated strategy improves the interaction between image restoration and segmentation, leading to enhanced robustness under challenging imaging conditions. The effectiveness of the proposed framework is evaluated using widely accepted quantitative metrics, including Peak Signal- to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), Mean Squared Error (MSE), Dice Similarity Coefficient (DSC), and Intersection over Union (IoU). The proposed methodology provides a scalable and reproducible solution for medical image enhancement and segmentation and offers potential applicability across multiple imaging modalities, including Magnetic Resonance Imaging (MRI), Computed Tomography (CT), ultrasound, and chest X-ray imaging.

Keywords: Medical image processing, image denoising, semantic segmentation, attention-guided U-Net, self- supervised learning, deep learning, structural similarity, Dice similarity coefficient, PSNR, IoU, MRI image analysis, CT image segmentation.