Dielectric Property Measurements For The Rapid Differentiation Of Thoracic Lymph Nodes Using XGBoost In Patients With Non-small Cell Lung Cancer

BackgroundsLung cancer is the second most common malignant tumor with the highest mortality rate in the world,among which Non-Small Cell Lung Cancer(NSCLC)patients account for more than 80%of patients with lung cancer.The National Comprehensive Cancer Network(NCCN)indicates that surgery is the preferred local treatment for patients with operable NSCLC.Surgical methods for patients with lung cancer mainly include lobectomy,segmentectomy and wedge resection.One of the decisive factors in surgical strategies is whether the thoracic lymph nodes are malignant or not.Therefore,for those NSCLC patients planning to undergo segmentectomy,thoracic lymph nodes should be routinely sent for rapid intraoperative frozen section(FS).However,FS is time-consuming and its accuracy is 80%-90%.In recent years,due to the differences of dielectric property between malignant tissues and normal tissues in the human body,dielectric property has attracted a lot of attention from researchers and may be used to distinguish malignant lymph nodes from benign ones.Dielectric property,including permittivity and conductivity,is inherent physical properties of biological tissue and related to water content,protein content and the concentration of various ions in human tissues.Transforming from normal tissues to malignant tissues,its water content,protein content and ion concentration will also be changing accordingly.Until now,a number of studies have shown that the dielectric property of the malignant tissues was significantly greater than the normal tissues for patients with lung cancer,breast cancer,liver cancer and colon cancer.However,there is currently no consistent measurement standard of dielectric property,which may lead to measurement errors and be hard to measure repeatedly.In addition,the dielectric property has the characteristic of high throughput.Therefore,we introduced the XGBoost(the Extreme Gradient Boosting)algorithm into the study as a classifier.XGBoost is widely used and its advantages are as follows.First,regularization terms,being suitable for tree models,is added in XGBoost algorithm and could control the complexity of the model to avoid overfitting.Second,the first derivative and the second derivative are used to make the loss more accurate.Third,it could handle with the condition that the training data is sparse values,which could greatly improve the computational efficiency of the algorithm.Fourth,it could adopt column sampling to reduce the amount of computation,etc.As a result,the method,combing the dielectric property and XGBoost algorithm,was introduced to the study and would be evaluated whether it could be used for the rapid intraoperative diagnosis of thoracic lymph nodes in patients with NSCLC.ObjectiveIn this study,the classifier model of the dielectric property and XGBoost was established to verify its diagnostic efficacy to differentiate malignant lymph nodes from benign ones for patients with NSCLC during surgery,which might provide a reference for thoracic surgeons to make surgical decisions.MethodsThis study was undertaken at Nanfang Hospital,Southern Medical University(Guagnzhou,China).The dielectric properties of thoracic lymph nodes were measured with open-ended coaxial probe method using the frequency of 1MHz4000MHz for patients with NSCLC during surgery.Then XGBoost was used to establish the diagnostic model basing on the data of dielectric properties.At last,we calculated the receiver operator characteristic curve and area under the curve(AUC)analysis and used the SMOTE(the Synthetic Minority Oversampling Technique)algorithm to make the number of malignant and benign lymph nodes balanced,which could verify the diagnostic efficacy of the method.ResultsA total of 204 lymph nodes,with 164 benign lymph nodes and 40 malignant nodes,from 67 NSCLC patients were included in this study.In the frequency of 1MHz-4000Mhz,the permittivity and conductivity about the malignant lymph nodes and benign lymph nodes have significantly different(P<0.001).Before using the XGBoost,the AUC values of permittivity and conductivity respectively are 0.850(95%confidence interval,CI:0.786 to 0.915,P<0.001)and 0.887(95%CI 0.828 to 0.946,P<0.001).The AUC value of the combined permittivity and conductivity was 0.893(95%CI 0.834 to 0.951,P<0.001).Using the XGBoost model,the AUC value was 0.968(95%CI 0.918 to 1.000,P<0.001),and the accuracy was 87.80%.after balancing the difference in the number of benign and malignant lymph nodes using SMOTE algorithm,the value of AUC was 0.954(95%CI 0.883 to 1.000,P<0.001),the accuracy was 92.68%.ConclusionThis study demonstrates that it is feasible to use the method of dielectric properties and XGBoost to rapidly differentiate malignant lymph nodes from benign lymph nodes during surgery,which might provide relatively effective information for thoracic surgeons to make decisions during surgery.