Cancer diagnosis in smart healthcare: Optimization of the MamCancerX model’s multiple instance learning framework

With the development of smart healthcare, the application of artificial intelligence in cancer diagnosis has become increasingly widespread. Whole slide images (WSIs) play an important role in cancer diagnosis, especially within the multiple instance learning (MIL) framework, where large-scale medical image data processing can significantly improve diagnostic accuracy. However, traditional convolutional neural networks (CNNs) and Transformer models still have limitations in handling WSIs, particularly in local feature extraction and global context modeling, leading to high computational complexity and memory consumption. To address these issues, this paper proposes MamCancerX, an innovative model that combines the Fusion Mamba module, Agent Attention module, and ResNet50 feature extractor. MamCancerX optimizes the fusion of local features and global information through cross-layer token fusion and self-attention mechanisms, enhancing performance and efficiency in cancer classification tasks. Specifically, the Fusion Mamba module improves global perception, while the Agent Attention module enhances the model’s focus on key regions through self-attention. Experimental results show that MamCancerX excels on the Camelyon16 and BRACS datasets, outperforming existing methods in key metrics such as accuracy, AUC, and F1 score, while also demonstrating significant advantages in memory consumption and computational efficiency.