Application Of CT Image Analysis And Prediction Of Benign And Malignant Ovarian Tumor With Tumor Markers Based On Deep Learning

objective: To explore the application value of artificial intelligence technology in CT image analysis of ovarian tumor and in the diagnosis of benign and malignant tumor with tumor markers.Methods: The data of a total of 138 patients with benign,borderline and malignant ovarian serous and mucinous tumors who underwent surgical treatment for ovarian tumors in Sichuan Provincial People?s Hospital from July2018 to November 2021 were retrospectively collected.Those patients also underwent preoperative enhanced CT examination in the pelvic or abdominal cavity and were confirmed by postoperative histopathological examination.The study selects images with complete ovarian tumor lesion region and clear edge from the enhanced CT images of the patient?s pelvic and abdominal cavity.The patients were randomly divided into a training set and a test set in a ratio of 3:1.Among them,a total of 3,261 images were included in the training set of 103 cases.A total of 960 images were collected from 35 patients in the test set.Then3,216 CT original medical images in the training set were labeled with the lesion regions.Through data enhancement,these images were input into the artificial intelligence image analysis model for training.The study adopts artificial intelligence image analysis model,which is an improved U-NET model and Res Net model based on deep learning.After the training,960 CT original medical images in the test set were used to test and diagnose the benign and malignant ovarian tumors.Finally,the 960 images in the test set were analyzed by the artificial intelligence image analysis model and then input into the artificial intelligence fusion model together with the corresponding tumor marker indexes to obtain the judgment of benign and malignant ovarian tumors in the final test set.In addition,the 960 original images in the test set were numbered,patient information was hidden and the order was randomly disrupted,and the chief physician,attending physician and resident physician were used to judge the CT images one by one in combination with or without tumor marker indicators.This study adopts SPSS21.0 statistical software for data analysis and processing.The study counts and collates the judgment results of CT images in artificial intelligent image analysis models and different physicians with or without tumor markers respectively and adopts the diagnostic criteria of postoperative histopathological examination results.The study also considers the computation accuracy,sensitivity,specificity,positive predictive value,negative predictive value,and Youden index.The study adopts Mc Nemar test(paired Chi-square test)to compare the parameters between groups.P<0.05 was considered to demonstrate statistically significant differences.Results: Statistics and artificial intelligence model in joint and not joint tumor markers and the different doctors,each joint and not joint tumor markers for CT images combined with or without combined with the index of tumor markers,resident and the attending physician diagnosis of CT images of 960 respectively,specific degree of accuracy and sensitivity of no statistical difference(p > 0.05).AI fusion model was superior to AI model in accuracy,specificity and sensitivity,with statistical differences(P <0.05).Both the AI model and the AI fusion model,each parameter index value was greater than the diagnosis results of residents and attending physicians(P <0.05),and the accuracy,sensitivity and specificity of the AI fusion model were 91.9%,91.9%and 91.8%,respectively,and each index was greater than 90%.The sensitivity of AI fusion model was better when chief physician combined tumor markers(P<0.05),but the specificity of AI fusion model was better than that of chief physician(P <0.05).When the chief physician did not combine tumor markers,the specificity of THE AI fusion model and the chief physician for CT image diagnosis was similar(P >0.05),and there was no statistical difference between the two,but the chief physician was superior to the AI fusion model in sensitivity and accuracy(P <0.05).Conclusions:1.CT image analysis based on deep learning combined with tumor markers has high accuracy,sensitivity and specificity in judging benign and malignant ovarian tumors,and each index is more than 90%,showing a certain clinical value.2.There is still a gap between the artificial intelligence technology and the diagnostic level of doctors compared to the judgment of CT images by senior doctors,which needs to be further improved.