An Improved Prostate Grading Method Based on Multi-Instance Learning

In recent years, with the development of deep learning technology, the medical field has begun to widely apply artificial intelligence technology for automated diagnosis and prediction of diseases. As one of the common male cancers, early diagnosis of prostate cancer is important for treatment. Traditional prostate cancer grading methods rely on manually analyzing pathology images, which is both time-consuming and susceptible to human factors. In this study, we propose an automatic prostate cancer grading method that combines multi-instance learning (MIL) and graph neural network (GCN) based on deep learning methods. We used the Karolinska dataset, which contains a large number of prostate cancer microscopy images and corresponding grading labels. In this study, ResNet18, pre-trained with ImageNet, was first utilized as the backbone network for feature extraction, and then a multi-instance learning approach was introduced to deal with labeling inconsistency, and the modeling of spatial relationships between local regions was further enhanced by graph neural networks. The experimental results show that the method combining MIL and GCN achieves significant improvements in accuracy, sensitivity and F1 score, and the performance of the model is significantly improved compared with the traditional method.