Leveraging Convolutional Block Attention Module in Transfer Learning for Scalable Multi-class Classification of Knee Osteoarthritis

Deep learning has greatly benefited from the Convolutional Block Attention Module (CBAM), which offers a useful method for integrating attention into Convolutional Neural Networks (CNNs) and modifying models for tasks requiring more focused attention to pertinent details. This study uses X-ray images to determine the impact of incorporating CBAM into well-known transfer learning models on Knee Osteoarthritis (KOA) classification. Four prominent transfer learning models such as EfficientNetb7, DenseNet169, DenseNet201, and ResNet101 are compared with and without incorporating a convolutional block attention module. The models are evaluated using classification precision, recall, f1-score, and accuracy metrics. According to the results, the CBAM-embedded DenseNet201 outperforms the others in the KOA's multi-class classification. The scalability of the proposed model is assessed using an expanded number of class instances in the dataset of knee osteoarthritis. Furthermore, the proposed model outperforms the state-of-the-art models. The method presented is likely significant for creating a lightweight CNN attention mechanism to improve classification performance in medical diagnosis, particularly for X-ray imaging.