基于人工智能的胃癌脉管癌栓病理辅助诊断模型研究OA

Objective To construct an artifical intelligence(AI)diagnostic model based on a deep learning algo-rithm and mount it on a photomicrographic camera system,and to assist pathologists in real time identification of vascu-lar tumor thrombus in gastric cancer under microscope,thereby improving diagnostic efficiency and accuracy.Meth-ods A total of 282 patients suffering gastric adenocarcinoma with(n=141)and without(n=141)vascular tumor thrombus diagnosed in the Affiliated Hospital of North Sichuan Medical College from January 2018 to April 2024 were collected.The corresponding paraffin blocks(a total of 803)were re-sectioned for HE staining and immunohistochemi-cal staining,and then HE sections were scanned to form digital sections.In combination with the results of immunohis-tochemical staining,6 234 single-field images with and 6 295 single-field images without vascular tumor thrombus fea-ture areas were captured in HE sections at a ratio of 200x,and were randomly divided into a training set and a test set according to the ratio of 8∶2.A model was built based on the convolutional neural network algorithm,and a diagnosis model was obtained after the training set images(a total of 12 529)were input into the network for deep learning,and then the test set images(a total of 2 506)were input into the model for testing and evaluation.The diagnosis model was connected and integrated with the photomicrographic camera system to be the final aided diagnosis model.Pathologists diagnosed the whole HE section test set in two states(model-aided and model-free),and their diagnostic accuracy and time consumption were compared.Results The accuracy,sensitivity,specificity,Kappa coefficient,and AUC value of the AI diagnostic model in the single-field image test set were 93.70%,95.03%,92.37%,0.873,and 0.974,re-spectively.The average time taken by the AI model to recognize single-field images was 0.062 seconds,which was sig-nificantly shorter than that of pathologists.In the whole HE section test set,the difference in the diagnostic accuracy of vascular tumor thrombus between pathologists using the AI model and those using traditional methods was not statis-tically significant,and pathologists using the AI model made shorter diagnoses than those using traditional methods[(72.46±16.25)seconds vs(91.18±17.05)seconds,t=7.946,P<0.05].Conclusion The AI-based patho-logically aided diagnosis model of vascular tumor thrombus in gastric cancer has high accuracy,and is helpful for pa-thologists to improve the diagnostic efficiency.