Clinical Research On The Recognition And Application Of Artificial Intelligence For Pathological T Staging Of Rectal Cancer Based On Pathological Large Section

Object:At present,the overall incidence of colorectal cancer is increasing worldwide.Moreover,pathological diagnosis is becoming increasingly arduous.Artificial intelligence has demonstrated the ability to fully excavate image features and assist doctors in making decisions.Large pathological sections contain considerable pathological information.which can avoid the error of splicing when using small slices in diagnosis.In this study,we used large panoramic pathological sections to establish a deep learning model to assist pathologists in identifying the cancerous area on whole-slide images of rectal cancer and judging T staging.Methods:We collected 168 cases of primary rectal cancer from the Affiliated Hospital of Qingdao University that had tissue surgically removed from January to September in 2019.After sectioning,staining and scanning,a total of 1400 Hematoxylin-Eosin staining(H&E-stained)whole-slide images were obtained.The random number table method was used to divide all patients into training cohort and validation cohort at the ration of 7:3.The training cohort was used to construct the artificial intelligence model,and the test cohort is used to verify the validity of the model.Finally,we used the accuracy,Dice coefficient,sensitivity,specificity,ROC curve and AUC area to evaluate the performance of the established artificial intelligence platform.Results:In this model,the accuracy of image segmentation and cancer recognition was 95%,Dice coefficient was 0.90,the AUC of the ROC curve of the model was 0.92,the sensitivity was 90%with the specificity was 95%.The total accuracy of automatic T staging recognition was 85%.The automatic recognition accuracy of T1,T2,T3 and T4 staging was 79%,81%,85%and 75%,respectively,and the associated AUC values were 0.86,0,88,0.90 and 0.70 respectively.The sensitivity and specificity were 80%,83%,88%,71%and 85%,86%3 88%,75%respectively.The time required for automatic image recognition is 0.2 seconds.Conclusion:The deep convolutional neural network model we established has high accuracy and strong model performance in the automatic identification of tumor regions and T-stage in postoperative pathological large sections.It can be used as an auxiliary diagnostic tool in clinical work to improve the efficiency of pathological diagnosis to some extent.