Multi-Label Artifact Classification Algorithm Based on Label Association Modeling

High-quality pathological sections are essential for both manual and computer-aided diagnosis. However, existing deep learning-based artifact detection methods, which typically rely on image patches, face significant challenges, including excessive computational resource consumption and the lack of a comprehensive artifact detection and analysis process. In histopathological images, artifacts such as uneven staining and scratches cover most of the tissue area, and a single image usually contains multiple types of artifacts. To address these issues, this paper proposes a multi-label artifact classification algorithm called Label Association-based Classification Model (LACM), which is suitable for low-magnification whole-slide pathological images. The algorithm integrates features from the last three layers of ResNet50 through the Tag Association-based Feature Fusion Module (TAF_Module), a feature fusion module designed to enhance the model’s feature resolution capability. This paper utilizes Self-attention mechanisms and introduces Graph Convolutional Networks (GCNs) to establish a model of label association, thereby further enhancing the model’s ability to handle complex label structures. Compared to other algorithms, the proposed method demonstrates superior effectiveness in classifying pathological image artifacts. The multi-label artifact classification algorithm proposed in this paper accurately identifies various types of artifacts in digital pathological sections, providing an effective technical solution for assessing the quality of pathological images.