CPNet: A Hybrid Neural Network for Identification of Carcinoma Pathological Slices

In the medical field, pathological carcinoma images look much more complicated than other medical images. Identifying carcinoma pathology images is a time-consuming and error-prone task for regular doctors and even for some specialists. Nowadays, deep learning has been widely applied in medicine, which could significantly reduce the time cost and improve accuracy. To save time and improve the accuracy of identifying pathological carcinoma slices, we propose a novel ViT-CNN hybrid neural network called CPNet, specially for the classification of different categories of carcinoma pathological slices. CPNet achieves the state-of-the-art performance in PatchCamelyon and our own dataset. We adopt a transfer learning method to identify degrees of malignancy using a few samples. Furthermore, we design and develop a fast medical decision system, where we deploy the CPNet in it. The system could effectively assist doctors in identifying the cancer pathology images with high accuracy and speed. The code of CPNet is in https://github.com/GuanRunwei/CPNet.