Detection Of Preoperative CT Radiomics In Diagnosis Of Esophageal Cancer

PurposeThe establishment of an esophageal cancer detection model based on preoperative chest CT plain scan and imageology will help radiologists screen out early asymptomatic esophageal cancer as soon as possible,so as to improve the diagnosis rate of early esophageal cancer.Material and MethodsWe retrospectively collected 686 patients with esophageal cancer(average age 61.2 years,range about 22-90 years)and 668 patients with non-esophageal cancer(average age 61.6 years,range about 23-95 years)from January 2017 to October 2019 in the Third Bethune Hospital of Jilin University and named them as data sets.Divide the data set into training group and test group in the proportion of 4:1,and use the in-depth learning model provided by Shanghai Lianying Intelligent Medical Technology to automatically segment tumors.In the u AI Research Portal of Lianying scientific research platform(http://urp.unitedimaging.com)feature extraction and filtering of the image.In the experiment,the absolute maximum normalization method is used to preprocess the data after feature selection,the principal component analysis method is used to reduce the dimension of the data,the support vector machine method is used to establish the prediction model,and the performance of the model is evaluated by the operating characteristic curve of the subject.In order to evaluate the clinical efficacy of the model,we will evaluate the performance of the data set based on the preoperative CT radiomics esophageal cancer detection model,the depth learning threshold method esophageal cancer detection model that has been applied to our hospital,three radiologists independently reading the data set and three radiologists assisted by the preoperative CT radiomics esophageal cancer detection model.ResultsThe constructed imaging histology model had an AUC value of0.923(95% CI:0.908-0.94),specificity of 0.869,sensitivity of 0.823 and accuracy of 0.846 in the training group,and an AUC value of 0.899(95%CI:0.864-0.936),specificity of 0.849,sensitivity of 0.789 and accuracy of0.818 in the test group.In the assessment of the clinical efficacy of the esophageal cancer detection model with preoperative CT imaging,the sensitivity,specificity and accuracy of the model were 81.8%,87.1% and84.4%,while the sensitivity,specificity and accuracy of the esophageal cancer detection model with the deep learning threshold method that has been applied to our hospital were 89.8%,46.6% and 68.5%.The average sensitivity,specificity and accuracy of the diagnostic results of the three radiologists who independently reviewed the films were 63.9%,80.3%and 72%,respectively.With the aid of this preoperative CT imaging model for the detection of esophageal cancer,the sensitivity,specificity,and accuracy of the three radiologists’ review results increased to 83.1%,93.4%,and 88.2%,respectively.ConclusionThe high sensitivity and specificity of the preoperative CT radiomics-based esophageal cancer detection model can effectively improve the detection rate of early esophageal cancer and reduce missed diagnoses for radiologists in preoperative chest CT plain scan.