Construction Of An Ultrasound Radiomics Model Based On Artificial Neural Network To Predict Large-Volume Lymph Node Metastasis In CN0 Papillary Thyroid Carcinoma

Part Ⅰ Risk Factors Of Large-volume Lymph Node Metastasis in Thyroid Papillary Carcinoma Objective:Accurate determination of high risk of recurrence and poor prognosis is important for the treatment and management of papillary thyroid carcinoma(PTC),and can effectively avoid overtreatment.Lymph node metastasis(LNM),especially large-volume Lymph Node Metastasis(Lv-LNM),is an important factor in the recurrence of PTC,and it is still a great challenge to accurately assess the lymph node status of PTC patients.The purpose of this study is to analyze the independent risk factors that predict the development of LNM and Lv-LNM in PTC patients,especially in clinically N-negative(c N0)patients,and to guide the clinical development of appropriate treatment strategies.Methods:This study included 527 patients who underwent first thyroid surgery and had pathologically confirmed PTC at the First Affiliated Hospital of Nanchang University from January 2018 to November 2022 were retrospectively analyzed.All patients underwent thyroidectomy and cervical lymph node dissection,and pathological results were obtained.Ultrasound examination of the thyroid and evaluation of the cervical lymph nodes were performed within 2 weeks before surgery.Ultrasound characteristics of the patients’ thyroid tumors were collected and ACR TI-RADS scores were given,and clinical information and pathological characteristics of the patients were collected,including age,sex,number of positive lymph nodes,and combined Hashimoto’s thyroiditis.This information was analyzed using univariate and multifactorial analyses so as to identify risk factors predictive of large-volume lymph node metastasis.Results:1.Male(OR=2.53,95% CI=1.36-4.72,P=0.004),age<40 years old(OR=2.52,95% CI=1.43-4.44,P=0.001),nodule diameter>1.0cm(OR=3.16,95% CI=1.74-5.73,P<0.001),abnormal LN found before operation(OR=8.10,95% CI=4.37-15.00,P<0.001),ACR TI-RADS score>8.5(OR=4.71,95% CI=2.00-11.07,P<0.001)are independent risk factors for predicting large-volume lymph node metastasis in PTC patients.2.Male(OR=2.35,95% CI=1.14-4.8,P=0.021),age<40 years old(OR=2.76,95%CI=1.37-5.56,P=0.005),nodule diameter>1.0cm(OR=2.22,95% CI 1.11-4.37,P=0.023)are independent risk factors for predicting large-volume lymph node metastasis in c N0 PTC patients.Conclusions:Patients with PTC can be stratified according to independent risk factors predicting large-volume lymph node metastasis,allowing for better tailoring of initial treatment recommendations.For PTC patients with preoperative imaging findings of metastatic lymph nodes,an aggressive treatment approach,such as surgical treatment with cervical lymph node dissection,is recommended.For men,age <40 years and tumor diameter >1.0 cm,surgical treatment is recommended to be more favorable regardless of whether metastatic lymph nodes are found preoperatively.In contrast,for female PTC patients aged ≥40 years with smaller tumors,active surveillance may be more reasonable if the preoperative clinical lymph nodes are negative.Part Ⅱ Ultrasound Radiomic Combined With Clinical Risk Factors For Predicting Large-volume Lymph Node Metastasis In c N0 Papillary Thyroid CarcinomaObjective:Large-volume lymph node metastasis(Lv-LNM)is considered to be an important risk factor for increased recurrence of PTC and may consequently affect patient survival,but all currently commonly used clinical methods for preoperative diagnosis of PTC lymph node metastasis have many shortcomings.There is evidence that some two-dimensional ultrasound features of PTC tumors correlate with lymph node metastasis,and radiomic is a promising method for data and image analysis that has been successfully applied in various fields of medical research.There have been studies using radiomic to construct PTC lymph node metastasis prediction models,however,their predictive efficacy is still unsatisfactory.In this study,we propose to construct a model using Artificial Neural Network(ANN)and radiomic to predict Lv-LNM in patients with clinically N-negative(c N0)PTC,and to validate the model in clinical cases to evaluate the model efficacy and clinical application value.Methods:A total of 306 patients with PTC who underwent first total thyroidectomy and cervical lymph node dissection at the First Affiliated Hospital of Nanchang University between January 2020 and April 2021 were included in this study,and all patients were pathologically confirmed and underwent preoperative ultrasound examination and stored ultrasound images.All patients were randomly divided into a training cohort(183 patients)and a validation cohort(123 patients)in the ratio of 6:4.Their clinicopathological characteristics and ultrasound information were collected,and their independent risk factors were assessed using univariate and multifactorial logistic regression methods,and clinical models were established based on them.In the radiomic analysis,the preoperative ultrasound images of thyroid nodules of each patient were manually outlined with the Region Of Interests(ROI)of the lesions,and the shape,intensity and texture radiomic features of the original and wavelet-transformed images were extracted.After the best radiomic features are filtered by downscaling and dimensionality,a radiomic model for predicting Lv-LNM is constructed in the training group using ANN and compared with the model constructed using traditional machine learning algorithms.In addition,an additional combined model was built in combination with clinically independent risk factors.The calibration,discrimination and clinical value of the two models were assessed in the validation group and compared with the clinical-only model.The predictive performance of the two models was also evaluated in the Papillary Thyroid Microcarcinoma(PTMC)and Conventional Papillary Thyroid Cancer(CPTC)subgroups.Results:1.The clinical value of the ultrasound radiomic model constructed using the ANN algorithm is superior to that of the model constructed using traditional machine learning algorithms.2.Both the radiomic model and the combined model had good fit and calibration ability.In the validation group,the radiomic model achieved 67% prediction accuracy with an AUC of 0.851(P<0.05),which was significantly higher than the clinical model with an AUC of 0.702(P<0.05).The combined model performed better in predictive efficacy with an accuracy of 79% and an AUC of 0.881(P<0.05).Both the radiomic model and the combined model had good calibration and clinical utility by calibration curve and decision curve analysis.In addition,both models still had excellent and stable predictive efficacy in the PTMC and CPTC subgroups.Conclusions:Based on ANN,We constructed a radiomic model,and a combined model with radiomic features combined with clinical risk factors,both of which can predict preoperatively large-volume lymph node metastasis in PTC c N0 patients with good predictive efficacy and clinical value,with the combined model performing better.ANN,together with radiomic,can be used as a powerful tool to help clinicians predict Lv-LNM in PTC.