Ultrasound Radiomics Analysis To Predict Central Lymph Node Metastasis In Patients With Papillary Thyroid Microcarcinoma

Objective:To develop a Gradient Boosting Decision Tree(GBDT)model for preoperative prediction of central compartment lymph node metastases(CLNM)in papillary thyroid microcarcinoma(PTMC)based on ultrasound(US)radiomics features and clinical risk factors that may predict CLNM,and to further evaluate its performance.Methods:A total of 274 patients with PTMC treated in the Department of thyroid surgery of China-Japan Union Hospital of Jilin University from January 2020 to July 2022 were analyzed retrospectively.Under the principle of random allocation,patients were divided into training cohorts and verification cohorts according to the ratio of8:2.The US images of PTMC lesions were imported into the open-source software(ITK-SNAP,3.8.0,www.itksnap.org),the region of interest(ROI)of the lesions was manually drawn in the ultrasound images.Using the open source software Pyradiomics to extract the radiomics features.The radiomics features with missing values were removed,and the abnormal data in the remaining features were filled with median.The raw data is then preprocessed by the maximum and minimum normalization method.Weka software is used to select CLNM-related radiomics features.Clinical risk factors associated with CLNM were statistically screened.The GBDT model was established by scikit-learn Python library(version 1.0.2)based on radiomics characteristics and clinical risk factors,and the diagnostic efficacy of cervical lymph node status reported by the US was compared.The mode FRONTIER software is used to constrain the GBDT model of the training and verification queues to avoid over-fitting of the model.Draw the receiver operating characteristic curve(ROC)and calculate the area under curve(AUC)for the evaluation of the diagnostic performance of the model.SHapley Additive Ex Planations(SHAP)was used to implement global and local visual analysis of GBDT models,and the SHAP algorithm is implemented by the SHAP Python framework(version 0.41.0).Evaluate the judgment performance of the model more accurately by plotting the Detection error tradeoff(DET)curve,and use decision curves to measure net benefits at different threshold probabilities to assess the clinical utility of the model.Results:In the training cohort,1034 features were extracted from each raw ultrasound image.52 invalid features were removed and 982 valid features were retained.After screening by Weka software,seven radiomics features were significantly correlated with the status of central lymph nodes.After the student’s t-test and chi-square test,gender(31.5% vs.12.0%,P<0.001),age(41.88±9.66 vs.46.94±8.12,P)were among the clinical factors <0.001)was significantly correlated with CLNM,and the GBDT model was constructed by combining radiomics characteristics,sex,age,and body mass index(BMI).The AUC values of the GBDT model in the training and validation cohorts were 0.946(95% CI,0.920-0.972)and 0.845(95% CI,0.714-0.976),respectively,while the AUC values for lymph node status reported by the US alone were 0.583(95% CI,0.508-0.659)and 0.582(95% CI,0.430-0.736),respectively.The Delong test showed a significant difference(0.946 vs.0.583,P<0.001)and the validation cohort showed similar results(0.845 vs.0.582,P<0.001).The DET curve shows that the curves of both cohorts are concentrated in the third quadrant,indicating that their false rejection rate and false acceptance rate are both at a low level,and the discriminant performance of the GBDT model is better.SHAP plots suggest that gender,radiomics characteristics,age 46-55,and age groups 26-35 are the key features of the model.Decision curve analysis When the threshold probability is between 0.1 and 1.0,the GBDT model is more beneficial to predict CLNM than all treated or untreated patients.Conclusion:1.GBDT model based on US radiomics and clinical risk factors can predict CLNM in PTMC patients and guide clinical decisions on patients’ treatment.2.GBDT models based on US radiomics and clinical risk factors outperform traditional ultrasound in predicting CLNM in PTMC.3.According to the analysis of clinical features,young male PTMC patients are more likely to develop CLNM.