Multimodel Data-driven For Intelligent Decision-making In Gastric Cancer Screening

Gastroscopy is an important method for finding suspicious stomach lesions,screening for gastric cancer and providing early diagnoses.The diagnostic ability of gastric cancer varies greatly in different grades of medical institutions.The diagnosis level of gastric cancer in primary medical institutions is relatively low,and the sensitivity of early diagnosis of gastric cancer is low in the world for a long time.An assistive system for gastroscopy report analysis can be helpful to improve the success rate of gastric cancer detection.Therefore,the research on clinical intelligent decision support for gastric cancer screening methods has great theoretical and practical value.This paper focuses on the problems of non-equalization of unstructured gastroscope report text processing,easy over-fitting,misdiagnosis and missed diagnosis of gastric cancer in the existing clinical decision support for gastric cancer diagnosis,a homogeneous ensemble-learning-based clinical decision support method for gastric cancer screening(Endo-GCS).Considering that the Endo-GCS method relies on the construction of manual features,the inability to process multimodal medical data,and the lack of interpretability of diagnostic results,a multimodal semantic fusion based clinical intelligent decision support method for gastric cancer screening(ID-GCS)is proposed.The specific research contents and innovations of this paper are as follows:(1)A homogeneous ensemble-learning-based clinical decision support method for gastric cancer diagnosis(Endo-GCS)is proposed.A progressive local weighting algorithm(PLWA)and an optimal threshold estimation algorithm(OTEA)is developed to enhance the specificity and sensitivity of the Endo-GCS method for gastric cancer screening.Finally,experiments were carried out to verify the improvement of the efficiency and sensitivity of the Endo-GCS method for gastric cancer screening.(2)A multimodal semantic fusion based clinical decision support method for gastric cancer screening(ID-GCS)is proposed.The gastroscope image semantic extraction network GCSNet and the gastroscope report semantic extraction network TextGCS is constructed,and the multimodal semantic fusion method is designed.Finally,contrastive experiments were performed using real gastroscopy reports and gastroscopic image data to verify the gastric cancer screening performance of ID-GCS and to perform semantic visualization to highlight its interpretability.