Study On The Diagnostic Value Of Ultrasound Radiomics In Cervical Lymph Node Tuberculosis

Objective:This paper intends to investigate the imaging histological features extracted based on two-dimensional ultrasound gray-scale images of enlarged lymph nodes.Constructing a diagnostic model of cervical lymph node tuberculosis using the extracted features,and to evaluate the diagnostic efficacy of the model.Methods:A total of 99 patients with enlarged lymph nodes who presented to our hospital from January 2019 to February 2023 were retrospectively analyzed.All included patients were got ultrasound-guided lymph node aspiration biopsy to obtain pathological results in our hospital,and the original images in DICOM format of conventional ultrasound examinations were retained,and clinical data were well preserved.All patients were randomly divided into a training set(n=69)and a test set(n=30)according to a 7:3 ratio.The included patients were divided into tuberculosis and non-tuberculosis groups based on pathological findings.The largest cross-sectional ultrasound grayscale images of the target lymph nodes were selected and imported into the professional imaging histology software ITK-SNAP,and the region of interest(ROI)of the boundaries of the target lymph nodes were manually outlined,and the open-source software pyradiomics was applied to extract the imaging histological features from the outlined ROI.The Pearson correlation coefficient(PCC)and LASSO algorithms were used for feature dimensionality reduction,seven classifiers(SVM,KNN,DT,RF,XGBoost,Ada Boost,Light GBM)were used to construct the model,and the ten-fold crossover method was performed to draw the receiver operating characteristic(ROC)curve of the model.The area under the receiver operating characteristic curve(AUC),specificity,and sensitivity were used to evaluate the diagnostic efficacy of the models in the training and test sets,and the best diagnostic models were screened.The ROC curves of the screened best model were plotted against those of two senior ultrasonographers to compare the diagnostic efficacy of the three,which was finally validated using the Delong test.Results:(1)A total of 99 patients with enlarged lymph nodes were included in this study,including 20 cases in the tuberculosis group and 79 cases in the non-tuberculosis group,with no statistically significant differences in gender and age distribution between the tuberculosis and non-tuberculosis groups(P > 0.05).(2)The image histological features were extracted from the ultrasound gray-scale images of the largest cross-section of the target lymph nodes,and the PCC and LASSO algorithms were used to reduce the dimensionality,and ten features with non-zero weight coefficients were finally screened out.(3)The final screened features were used to construct an imaging histology model,and a ten-fold cross-validation was performed to screen the best model among the seven classifiers,which performed well in terms of diagnostic efficacy.The AUC,specificity,sensitivity and accuracy of the training cohort were 0.972,96.9%,87.5%and 95.0%,respectively,and the AUC,specificity,sensitivity and accuracy of the validation cohort were 0.858,86.7%,75.0%,and 84.2%,respectively,for the validation cohort.(4)The diagnostic efficacy of the best screened ultrasound imaging histology model was higher than that of the two senior sonographers(sonographer A AUC:0.687,sonographer B AUC: 0.578).Conclusions:1.The ten ultrasound imaging histological features screened in this study for the diagnosis of cervical lymph node tuberculosis with a non-zero weighting factor were Minimum,Cluster Shade,Large Dependence High Gray Level Emphasis,Short Run Low Gray Level Emphasis.Gray Level Non Uniformity,Small Area Emphasis,Zone Variance,Coarseness,Contrast,and Elongation.2.The ultrasound imaging histology model Ada Boost screened and established in this study has good diagnostic efficacy for cervical lymph node tuberculosis(training set AUC: 0.972,test set AUC: 0.858)and outperforms two senior ultrasonographers(ultrasonographer A AUC: 0.687,ultrasonographer B AUC: 0.587),which is a non-invasive,objective,reliable and diagnostic method with high accuracy and high clinical application value.