基于无监督-监督融合深度学习模型在慢性萎缩性胃炎病理诊断中的应用研究OA

Objective To introduce an unsupervised-supervised fusion deep learning model to assist pathologists in grading the severity of gastric mucosal atrophy and intestinal metaplasia in gastroscopic biopsy specimens of chronic atrophic gastritis(CAG).Methods A total of 455 gastroscopic biopsy specimens diagnosed as CAG within one year were collected from Jiangsu Provincial Hospital of Chinese Medicine,and all specimens were digitized into whole slide images(WSIs).An unsupervised-supervised fusion deep learning framework was employed.WSIs from 155 biopsy specimens were utilized for model training,and the remaining 300 WSIs were adopted for model testing.Results For the grading of gastric mucosal atrophy,the model achieved an overall accuracy of 96.67%,weighted average sensitiv-ity of 96.67%,weighted average specificity of 98.88%,with a Kappa consistency coefficient of 0.95.For the grading of intestinal metaplasia,the overall accuracy,weighted average sensitivity and weighted average specificity reached 98.33%,98.33%,and 98.89%,respectively,with a Kappa consistency coefficient of 0.98.Conclusion The unsupervised-supervised fusion deep learning model can efficiently and accurately assist pathologists in grading gastric mucosal atrophy and intestinal metaplasia in CAG specimens,and exhibits excellent diagnostic consistency with manual pathological evaluation.