Establishment And Application Of Artificial Intelligence-assisted Diagnosis Platform For Gastric Cancer T Staging Based On Pathological Large Section

Object: To study the value of artificial intelligence-assisted diagnosis platform for gastric cancer T staging based on pathological large sections.At present,the overall incidence of gastric cancer worldwide is increasing,and the requirements for the efficiency of pathological diagnosis and treatment of gastric cancer are increasing.In recent years,artificial intelligence technology has become increasingly mature and can be used to assist clinicians in diagnosis and treatment,and the use of pathological large sections is more accurate and contains more pathological information than conventional pathological small sections splicing observation.In this study,we established a deep learning model of large digital images of gastric cancer pathology to assist pathologists in determining the T stage of gastric cancer.Methods: The pathological specimens of 106 patients with gastric cancer who underwent gastric cancer resection from the Affiliated Hospital of Qingdao University from January to December 2019 and met the inclusion criteria were retrospectively collected to make pathological wax blocks,and a total of 1000 HE-stained whole slide scan images were obtained after sectioning,staining and scanning,which were grouped by random list method,and randomly divided into learning group and verification group according to the ratio of 7:3,including 700 in the learning group and 300 in the verification group,and the image data of the learning group were used to train and establish artificial intelligence models.Validate the set of image data to test the performance of the model.The accuracy,sensitivity and precision of the AI model and the pathologist were compared,the average diagnosis time and the AUC(area under the curve)and ACC(accuracy)of different stages were counted,and the diagnostic results of the AI and pathologists were compared to analyze their learning level.Results: The area under the curve(AUC)of the subject’s operating characteristic curve(ROC curve)in the T1-T4 stage of the AI model was 0.81,0.88,0.92 and 0.77,respectively,and there was no significant difference between the accuracy,sensitivity and precision of the AI model and that of pathologists(P>0.05).The average diagnostic time of the AI model was 0.41±0.12 seconds,which was shorter than the 0.86±0.26 seconds(P<0.05)for pathologists.The AUC and ACC of T1 and T4 stages were lower than those of T2 and T3(P<0.05).Conclusion: The artificial intelligence-assisted diagnosis platform for gastric cancer T staging based on pathological large sections established in this study has high accuracy of automatic identification of T stage,short recognition time,and high performance,which can assist clinicians in more efficient diagnosis and treatment,and can better avoid the differences in diagnosis of different pathologists,which is worthy of promotion and use.