Lesion region recognition of lung cancer pathology based on the Inception-V4 model

Lung cancer diagnosis has invariably been a top related research topic in medical image analysis, which aims to build a classification model to determine whether there is lung cancer by analyzing the lung cancer tissue structure and nuclear morphology in lung images. Benefiting from the powerful feature expression capabilities of convolutional neural networks, the research on lung cancer diagnosis based on deep learning has developed rapidly. In this paper, we mainly apply different convolutional neural networks to recognize the Lesion region of lung cancer pathology and compare the performance of Inception-V3 and Inception-V4 models for predicting malignant tumors on image blocks of 299×299 size. The accuracy and loss function results of Inception-V4 can reach 99.75% and 0.0173 much better than 98.33% and 0.2419 on Inception-V3 models. Finally, we use the trained model to predict the tumor region in Full-size tissue Pathology images and compare it with the expert manually labeled area for discussion. This study shows that the Inception-V4 model contributes to rapid tumor detection and is expected to provide assisted diagnosis and decision support for lung cancer.