Establishment Aod Validation Of The Intelligent Assessment Model For Multimodal Ultrasound Imaging Of Thyroid Nodules

Part ?:Ultrasound Real-Time Tissue Elastography Improves theDiagnostic Performance of Malignancy Risk Stratification in ThyroidNodulesObjectivesTo explore the correlation between the ultrasound elasticity score(ES)of real-time tissue elastography(RTE)and malignancy risk stratification(MRS),evaluate the added value of RTE to MRS in differentiating malignant nodules from benign nodules.Materials and methodsA total of 1,498 patients(885 women and 613 men;mean age of 43.5±12.4 years)with 1,525 confirmed thyroid nodules(D = maximum diameter,D?2.5 cm)confirmed by fine-needle aspiration(FNA)and/or surgery were included.The nodules were divided into four groups based on their sizes(D?0.5 cm,0.5<D?1.0 cm,1.0<D?2.0 cm,and 2.0<D?2.5 cm).We assigned an ES and MRS to each nodule.The correlation between the RTE and the MRS was analyzed using the Spearman rank correlation.The diagnostic performances of RTE,MRS and combined RTE and MRS were compared using(ROC)analysis.ResultsRTE and MRS showed a strong correlation in the intervals of 0.5<D?1.0 cm,1.0<D?2.0 cm,and 2.0<D?2.5 cm(r = 0.768,0.711,and 0.743,respectively).The diagnostic performance of the combination of RTE and MRS in each size interval was always better than RTE or MRS alone(for all,P<0.001).ConclusionsOverall,a strong correlation was found between RTE and MRS.The combination of RTE and MRS outperformed RTE or MRS alone.Therefore,RTE is an adjunctive diagnostic tool to MRS for in differentiating benign and malignant thyroid nodules.Part ?:Diagnostic value of multimodal ultrasound imaging indifferentiating benign and malignant TI-RADS category 4 nodulesBackgroundDifferential diagnosis of benign and malignant Thyroid Imaging Reporting and Data System category 4(TI-RADS-4)nodules can be difficult using conventional ultrasound(US).This study aimed to evaluate whether multimodal ultrasound imaging can improve differentiation and characterization of benign and malignant TI-RADS 4 nodules.MethodsMultimodal ultrasound imaging,including US,superb microvascular imaging(SMI),and real-time elastography(RTE),were performed on 196 TI-RADS-4 nodules(78,benign;118,malignant)in 170 consecutive patients.The sensitivity,specificity,accuracy,false negative rate(FNR),and false positive rate(FPR)of each single method and that of multimodal US imaging were determined by comparison with surgical pathology results.ResultsThe sensitivity,specificity,accuracy,FNR,and FPR for US were 65.25%,69.23%,66.84%,34.75%,30.77%,respectively;for SMI were 77.97%,93.59%,84.18%,22.03%,6.41%,respectively;RTE,80.51%,84.62%,82.14%,19.49%,15.38%;and for multimodal US imaging were 94.08%,87.18%,91.33%,6.93%,12.82%,respectively.The areas under the received operating characteristic curve for US,SMI,RTE,and multimodal US imaging in evaluating benign and malignant TI-RADS 4 nodules were 67.2%,84.40%,86.60%,and 95.50%,respectively.ConclusionsThe initial clinical results suggest that multimodal US imaging improves the diagnostic accuracy of TI-RADS 4 nodules and provides additional information for differentiating malignant and benign nodules.Part ?:Machine Learning-Assisted System for Thyroid NoduleDiagnosisBackgroundUltrasound(US)examination is helpful in differential diagnosis of thyroid nodules(malignant vs.benign),but its accuracy relies heavily on examiner experience.Therefore,we aimed to develop a less subjective diagnostic model aided by machine learning.MethodsInstitutional review board approval was obtained,and informed consent was waived in this retrospective study.A total of 2,064 thyroid nodules(2,032 patients;695 men and 1,337 women;45.25 ± 13.49 years)met all the following inclusion criteria:(1)hemi-or total thyroidectomy,(2)maximum nodule diameter 2.5 cm,(3)examination by conventional US and real-time elastography(RTE)within 1 month before surgery,and(4)no previous thyroid surgery or percutaneous thermotherapy.Models were developed using 60%of randomly selected samples based on 9 commonly used algorithms,and validated using the remaining 40%cases.All models function with a validation dataset that has a pre-test probability of malignancy of 10%.The models were refined with machine learning that consisted of 1,000 repetitions of derivatization and validation,and compared to diagnosis by an experienced radiologist.Sensitivity,specificity,accuracy and area under the curve(AUC)were calculated.ResultsRandom forest algorithm led to the best diagnostic model,which performed better than radiologist diagnosis based on conventional US only[AUC 0.924(95%confidence interval[CI]:0.895-0.953)vs.0.834(95%CI:0.815-0.853)]and based on both conventional US and RTE[AUC:0.938(95%CI:0.914-0.961)vs.0.843(95%CI:0.829-0.857)].ConclusionsMachine learning algorithms based on US examinations,particularly the random forest classifier,may diagnose malignant thyroid nodules better than radiologist diagnosis.Part ?:Development and validation of a web-based malignancyrisk-stratification system of thyroid nodules for precision treatment PurposeTo develop a practical and simplified prediction model for categorizing the malignancy risk of thyroid nodules based on clinical,biochemical,and ultrasonographic characteristics,would be clinically important and could provide guidance to clinical decision making.MethodsA total of 2818 consecutive patients(female,1890;mean age,46.0±12.9 years)with 2850 thyroid nodules were retrospectively evaluated at the authors' institution between April 2011 and October 2016.Two thyroid radiologists independently reviewed the clinical,biochemical,and ultrasonographic data of each nodule.We then used a randomly selected sample of 80%of the nodules to perform a multivariate logistic regression analysis.Cut points were determined to create a risk-stratification scoring system.Within the system,patients were classified as having low,moderate and high probability of malignancy according to their scores.After these determinations,we validated the models to the remaining 20%of the patients.The area under the receiver operating characteristic curve(AUC)was used to evaluate the discrimination ability of the system.ResultsNine variables were selected as predictors for malignancy in the risk-stratification system.Patients were considered at low risk if the score was<4.0,moderate risk if the score was 4.0 to 6.0,and high risk if the score was?6.0.Malignancy risk was segmented from 0%to 100.0%and was positively associated with an increase in risk scores.The AUC of the development and validation sets were as high as 0.936(95%CI:0.923-0.949)in the development set and 0.934(95%CI:0.908-0.960),respectively.ConclusionThe study shows that a simple and reliable web-based risk-stratification system could be practically used in stratifying the risk of malignancy in thyroid nodules.