Research on a Multi-Dimensional Feature Extraction Method for Medical Images Based on the MSCNN-GCNN Framework

Abstract Medical image analysis plays a crucial role in clinical diagnosis, yet traditional methods are prone to subjective bias and suffer from labor-intensive feature extraction, limiting their accuracy and robustness. Although deep learning has demonstrated potential in medical image analysis, its application remains challenging due to the complexity of medical data, inconsistent image quality, and inherent model limitations. To address these issues, this study proposes a hybrid framework (MSCNN-GCNN) integrating Multi-Scale Convolutional Neural Network (MSCNN) with Graph Convolutional Network (GCNN) to enhance feature representation. The MSCNN extracts multi-resolution features through parallel subnetworks, while the GCNN models spatial relationships via graph structures, ultimately fusing multi-scale and structural features for abnormality detection. Experiments on the Musculoskeletal Radiographs (MURA) dataset demonstrate that the proposed method significantly outperforms DenseNet169 and CapsNet approaches across key metrics: F1-score (91%), Kappa coefficient (83%), accuracy (85%), and balanced accuracy (86%). This research provides novel insights for deep learning applications in medical imaging, particularly in small-target detection and structural feature modeling.