Up-Gastrointestinal AI


Intelligent diagnosis of upper gastrointestinal cancer

Upper gastrointestinal cancer is one of the common cancer with high incidence in China. According to statistics, about 50% of the world’s upper gastrointestinal cancers occur in China, more than 85% of which are in the advanced stage at the time of diagnosis. Gastroscopy and endoscopic biopsy are the gold standard for the diagnosis of upper gastrointestinal cancer. However, patients with early gastrointestinal cancer often lack typical endoscopic characterization, who are most likely to be missed during endoscopic inspection. Moreover, there is a big different in the experience and operating techniques of endoscopists in different hospitals, which has a great impact on the accuracy of endoscopic lesion diagnosis.

Therefore, we intend to develop an intelligent diagnosis system for upper gastrointestinal endoscopy, to greatly improve the recognition rate and accuracy of early lesions. We will collect endoscopic image data from more than 150,000 cases of upper gastrointestinal cancer patients, and 150,000 cases of normal people. Combined with their clinical information, we will apply artificial intelligent to deeply study and analyse medical images and clinical data, and simulate doctor's thinking and diagnostic reasoning patterns. Ultimately achieve the goal of intelligent recognition and analysis of endoscopic tumor lesions, as well as assist doctors in locating lesions and making a diagnosis. On this basis, a biopsy real-time navigation system will be further developed, which will automatically capture images during endoscopic process. With cloud AI analysis and real-time feedback, this system can prompt endoscopist for biopsy of suspected lesions, which help doctors to identify early lesions more accurately, and improve early detection rate and diagnostic accuracy.

Currently, we have completed the recognition of endoscopic image data and deep learning modeling of 30,000 patients with upper gastrointestinal cancer. Through randomization, data of these 30,000 cases were divided into training set, verification set and independent test set. In the independent test set, the model accuracy rate reached 98%, which initially showed good performance and great potential of the gastrointestinal artificial intelligent diagnosis system. After that, we will fully integrate, mine, and share expert knowledge and big data information through the integration of big data, artificial intelligence and Internet technologies, to continuously optimize the gastrointestinal artificial intelligent diagnosis system. We expect to form an international leading tumor artificial intelligent diagnosis program, and ultimately form artificial intelligence applications and services, which will provide open access to society.