The gastric cancer is one of the most common malignant tumors of alimentary canal. According to statistics, the incidence of gastric cancer in China ranks third among malignant tumors, and the fatality rate ranks second. Because early gastric cancer has no specific symptoms and signs, a few people will have nausea, vomiting or upper gastrointestinal symptoms similar to ulcer disease, but it is difficult to attract enough attention. Most patients are in the advanced stage at the time of discovery. Therefore, identifying high-risk groups and improving the early diagnosis rate of gastric cancer are the key to improve the survival rate of patients.

In recent years, with the improvement of deep learning, machine learning algorithms, hardware levels and databases, artificial intelligence technology has ushered in the third wave of development, which can provide powerful assistance for medical diagnosis in the medical field. This research team constructed a smart image recognition and diagnosis generation system for gastric cancer. The system is divided into the following modules: (1) Extract stomach image from uploaded CT image sequence; (2) Lesions were detected on stomach images; (3) A medical diagnosis report is generated by distinguishing between gastric images and lesions detected.

Technical solution:

The system (1) focuses on the stomach body to establish the relative position relationship between the stomach body and the lesion, and adds two channels in the input model to store the position information, so as to complete the connection with the original image to realize the position coding of the image. (2) The multi-scale feature fusion method based on self-attention mechanism is used to mine advanced semantic information of small lesions from complex stomach images for medical diagnosis report generation.

Technical index:

(1) Lesion detection rate reached 0.875; (2) The accuracy of staging reached 70.81%; (3) Fully combined with the prior knowledge of experts, the location, depth and stage of the lesions were accurately located, and the description accuracy rate of automatic diagnosis was 74.52%.

Type of cooperative enterprise:

Smart medical company