Author :

Saeid Akbar Jalali

Address :

Department of Computer Engineering,Iran University of Science & Technology, iran

Abstract :

High-resolution histopathology images are critical for accurate medical diagnosis, but often suffer from visual artifacts introduced during digitization, compression, or staining variations. These artifacts can obscure delicate tissue structures, reducing diagnostic reliability and model performance in automated analysis. Conventional artifact removal methods rely heavily on supervised learning and struggle with generalization, particularly when annotated data is limited or artifact patterns vary widely. To address these challenges, it proposes a Contrastive MemoryAugmented Denoising Network (CMADN) that integrates contrastive learning with a memory module. The contrastive learning component trains the model to differentiate between clean and artifacted image patches. The memory module stores feature representations of clean patches to guide artifact suppression during inference. This framework is applied as a preprocessing step in AI-based histopathology pipelines to enhance image clarity before diagnostic classification. Experimental results demonstrate that CMADN significantly reduces artifacts while preserving cellular structures, outperforming existing denoising approaches in both visual quality and downstream diagnostic accuracy.

Keywords :

Contrastive learning, memory network, histopathology, visual artifacts, image denoising, medical image preprocessing.

DOI :