| Objective: To explore the diagnostic value of strain elastography and artificial intelligence in the differential diagnosis of thyroid nodules.Method: This study prospectively enrolled 822 patients from 9 hospitals from April 2019 to January 2021.All thyroid nodules had fine needle aspiration cytology(FNAC)results or surgical resection pathological results.The strain elastography imaging and the artificial intelligence based computer-aided diagnosis system(CAD)on the transverse and longitudinal sections was performed on Samsung RS80 A.We also retrospectively collected2082 thyroid ultrasound images with pathological results from 1396 patients in two hospitals from June 2017 to April 2019,and constructed a deep learning model based on the American College of Radiology(ACR)Thyroid Imaging Reporting and Data System(TIRADS)class 4 and class 5 nodules.The area under the curve(AUC),accuracy,sensitivity,specificity,positive predictive value,and negative predictive value werre calculated and analyzed.Result: The optimal cutoff values for distinguishing benign and malignant thyroid nodules on the transverse and longitudinal sections were 1.31 and 1.50,respectively.When ACR TIRADS classification was combined with elasticity score and strain ratio,the AUC increased from 0.648 to 0.829(P<0.001);the sensitivity increased from 84.7% to 87.8%(P<0.05);the specificity increased from 44.9% to 78.1%(P<0.001);the positive predictive value increased from 78.2% to 90.3%(P<0.001);the negative predictive value increased from55.8% to 73.3%(P<0.001).The optimal cutoff values for distinguishing benign and malignant thyroid nodules on the longitudinal and transverse sections using the elasticity contrast index(ECI)were 1.54 and 1.07,respectively,with sensitivity of 51.1% and 64.1%,and specificity of 74.9% and 60.4%,respectively.When ACR TI-RADS classification(TR)was combined with the elasticity contrast index on both sections,the specificity increased from 44.3% to 77.6%(P<0.001),and there was no significant decrease in sensitivity(81.4%vs.79.0%,P=0.290);the positive predictive value increased from 75.6% to 88.2%(P<0.001),and the negative predictive value increased from 53.3% to 63.7%(P<0.001).The artificial intelligence based CAD system got an AUC of 0.759 on the longitudinal section and 0.750 on the transverse section,with no significant statistical difference(P>0.05).The diagnostic accuracy,sensitivity,specificity,positive predictive value,and negative predictive value were 80.8%,89.1%,62.7%,84.0%,and 72.2%,respectively,on the longitudinal section;and 80.2%,88.9%,61.1%,83.4%,and 71.3%,respectively,on the tranverse section.After referring to the results of CAD software,the diagnostic performace of well-experiecend radiologists and radiologists with less experience were improved(P<0.05).In the independent test set,the deep learning(DL)algorithm of best performance got an AUC of 0.904,0.845,0.829 in TR4 nodules,TR5 nodules,TR4 and TR5 nodules,respectively.The sensitivity and specificity of the optimal model was 0.829,0.831 on TR4 nodules,0.846,0.778 on TR5 nodules,0.790,0.779 on TR4 and TR5 nodules,versus the radiologists of 0.686(P= 0.108),0.766(P= 0.101),0.677(P= 0.211),0.750(P= 0.128),and0.680(P= 0.023),0.761(P= 0.530),respectively.Conclusion: Strain elastography,carotid artery pressure based elastography,artificial intelligence based CAD systems,and deep learning models can serve as effective noninvasive auxiliary diagnostic methods,improving the diagnostic performance of twodimensional ultrasound in distinguishing benign and malignant thyroid nodules,and reducing unnecessary biopsy.Part I: Diagnostic value of strain elastography in the differentiatiton of benign and malignant thyroid nodules: a prospective multicenter studyObjectives: To explore the diagnostic performance of the combination of the strain elastography and ACR TI-RADS based on transverse and longitudinal sections in the differentiation of benign and malignant thyroid nodules.Methods: This prospective multicenter study finally enrolled 822 patients from 9 hospitals from April 2019 to January 2021.Nine trained ultrasound physicians with over 5 years of experience in thyroid ultrasound imaging performed two-dimensional ultrasound and strain elastography.Three other experienced radiologistes evaluated the thyroid images based on ACR TI-RADS.The diagnostic performance for thyroid nodule classification was assessed by comparing AUC,sensitivity,specificity,positive predictive value,negative predictive value before and after combining elasticity score and/or strain rate ratio.Results: The optimal cutoff values for distinguishing benign and malignant thyroid nodules on the transverse and longitudinal sections were 1.31 and 1.50,respectively.The AUC of strain ratio and elasticity score in differentiating benign and malignant thyroid nodules is not affected by the chosen sections(P>0.05).When the ACR TI-RADS classification is combined with elasticity score or strain ratio,the AUC,specificity,positive predictive value,and negative predictive value are all higher than the ACR TI-RADS classification alone(P<0.05).When ACR TI-RADS classification was combined with elasticity score and strain ratio,the AUC increased from 0.648 to 0.829(P<0.001);the sensitivity increased from 84.7% to 87.8%(P<0.05);the specificity increased from 44.9% to 78.1%(P<0.001);the positive predictive value increased from 78.2% to 90.3%(P<0.001);the negative predictive value increased from 55.8% to 73.3%(P<0.001).When ACR TI-RADS classification was combined with elasticity score and strain ratio,the biopsy rate of thyroid nodules reduced(P<0.05)and the malignancy rate of the biopsy nodules(P<0.05)increased.Conclusions: Strain elastography can be used as a useful and non-invasive additional tool to improve the diagnostic performance of ultrasound in differentiating benign and malignant thyroid nodules and reduce unnecessary biopsy rates.Part II: Diagnostic value of carotid artery pressure based elastography in in the differentiatiton of benign and malignant thyroid nodules: a prospective multicenter studyObjectives: To explore the diagnostic performance of carotid artery pressure based elastography combined with ACR TI-RADS classification based on transverse and longitudinal senctions in the differentiation of benign and malignant thyroid nodules.Methods: This multicenter study prospectively enrolled 839 patients from 9 hospitals from April 2019 to January 2021.The conventional ultrasound and carotid artery pressure based strain elastography were performed by 9 well-trained radiologists with over 5 years of experience in thyroid ultrasound imaging.Three other experienced radiologists evaluated ACR TI-RADS classification based on the two-dimensional ultrasound images.The diagnostic performance in thyroid nodules classification was assessed by comparing the AUC,the diagnostic sensitivity,specificity,positive predictive value,negative predictive value before and after the combination of elasticity contrast index.Results: The optimal cutoff values of elasticity contrast index for distinguishing benign and malignant thyroid nodules on the longitudinal and transverse sections were 1.54 and 1.07,respectively,with the sensitivity of 51.1% and 64.1%,and the specificity of 74.9% and 60.4%,respectively,while the AUC is not affected by the thyroid nodule section(P>0.05).When ACR TI-RADS classification was combined with the elasticity contrast index on both sections,the specificity increased from 44.3% to 77.6%(P<0.001),and there was no significant decrease in sensitivity(81.4% vs.79.0%,P=0.290);the positive predictive value increased from 75.6% to 88.2%(P<0.001),and the negative predictive value increased from 53.3% to 63.7%(P<0.001).Conclusions: Carotid artery pressure based elastography is an effective non-invasive auxiliary diagnostic method,and its diagnostic efficacy was not influenced by the chosen selections.It can improve the diagnostic performance of ultrasound in distinguishing benign and malignant thyroid nodules and reduce unnecessary biopsy.Part III: Evaluation of computer-aided diagnosis software in the differential diagnosis of benign and malignant thyroid nodules: a multicenter prospective studyObjectives: To evaluate the diagnostic performance of ultrasound images based computeraided diagnosis(CAD)software in distinguishing benign and malignant thyroid nodules on transverse and longitudinal sections.Methods: This prospective multicenter study included patients with solid thyroid nodules between April 2019 and January 2021.The graysacle ultrasound images and CAD software results for each thyroid nodule were collected.The diagnostic performance of CAD software in distinguishing benign and malignant thyroid nodules was obtained by calculating AUC,diagnostic accuracy,sensitivity,specificity,positive predictive value,negative predictive value.The results of CAD software were provided as reference to radiologists with experience of different years to explore its effects to assist doctors.Results: In total,783 thyroid nodules from 758 patients were included in this study.The AUC of CAD software is 0.759 on the longitudinal section,and 0.750 on the transverse section(P>0.05).The diagnostic accuracy,sensitivity,specificity,positive predictive value,and negative predictive value were 80.8%,89.1%,62.7%,84.0%,and 72.2%,respectively,on the longitudinal section;and 80.2%,88.9%,61.1%,83.4%,and 71.3%,respectively,on the tranverse section.The combination of parallel and tandem pattern can not improve AUC(0.748 vs.0.763,P>0.05).After referring to the results of CAD software,the diagnostic performace of well-experiecend radiologists and radiologists with less experience were improved(P<0.05).Conclusions: Ultrasound images based CAD software can serve as an effective auxiliary diagnostic tool for distinguishing benign and malignant thyroid nodules,and there is no significant difference between the sections chosen.After referring to the results of CAD software,the diagnostic performance of radiologists with different years of experience can be improved.Part IV: Deep learning based on ACR TI-RADS can improve the differential diagnosis of thyroid nodulesObjective: The purpose of this study was to improve the differentiation between malignant and benign thyroid nodules using deep learning(DL)in ACR TI-RADS category 4 and 5.Design and Methods: From June 2,2017 to April 23,2019,a total of 2082 thyroid ultrasound images from 1396 consecutive patients with confirmed pathology were retrospectively collected,of which 1289 nodules were category 4(TR4)and 793 nodules were category 5(TR5).Ninety percent of the B-mode ultrasound images were applied for training and validation,and the residual 10% and an independent external dataset for testing purpose by three different deep learning algorithms.Results: In the independent test set,the DL algorithm of best performance got an AUC of 0.904,0.845,0.829 in TR4,TR5,and TR4&5,respectively.The sensitivity and specificity of the optimal model was 0.829,0.831 on TR4,0.846,0.778 on TR5,0.790,0.779 on TR4&5,versus the radiologists of 0.686(P=0.108),0.766(P=0.101),0.677(P=0.211),0.750(P=0.128),and 0.680(P=0.023),0.761(P=0.530),respectively.Conclusions: The study demonstrated that DL based on ACR TI-RADS could improve the differentiation of malignant from benign thyroid nodules and have significant potential for clinical application on TR4 and TR5 nodules. |
PDF Full Text Request