A weakly supervised multi-instance learning based on graph neural network for breast cancer pathology image classification

Breast cancer is a malignant tumor that occurs in the breast, posing a serious threat to women’s health. Histopathological examination is an important means of breast cancer diagnosis, and the analysis of histopathological images is an important application of artificial intelligence in histopathological examination. Despite the remarkable achievements of convolutional neural networks (CNNs), there are still challenges in distinguishing between multiple subtypes and overcoming imbalanced datasets. Therefore, to overcome these challenges, this paper proposes a weakly supervised multi-instance learning model GMIL based on graph neural networks (GNNs) for breast cancer pathology image classification. The model consists of three modules: the feature extraction module for feature extraction, the GNN module for aggregating neighbor information, and the gated attention module for feature fusion. Multiple experiments on the BreakHis dataset show that the proposed model outperforms the latest algorithms in terms of performance.