| Background Thyroid nodule is a common disease with a detection rate of up to 65%in the general population.Nowadays,the incidence of thyroid cancer is increasing year by year and has become one of the most common malignant tumors.Accurate diagnosis and treatment are particularly important for patients.One of the major clinical problems is thyroid nodule screening,that is,the use of imaging or laboratory methods to diagnose malignant or suspected malignant thyroid nodules.Ultrasonography is a simple and easy thyroid nodule examination method,which can be sensitive to the detection of thyroid nodule,and can accurately evaluate the location,size and number of thyroid nodule.It is considered as the preferred method for differentiating benign from malignant thyroid lesions.However,at present,the diagnostic level of sonographers is uneven,and the diagnostic level of sonographers in large hospitals is high,but there are many patients and the workload is huge.In some remote areas where the medical level is relatively low,the diagnostic accuracy of ultrasound doctors can not be guaranteed.To solve these problems,researcher began to use computer simulation of human brain structure to calculate,designed an artificial neural network,and applied it to medical imaging,greatly reducing the pressure of medical workers.Based on the intelligent ultrasonic diagnosis of thyroid nodules,the products that have been approved to market include "An Ke Zhen"robot from Taiwan,Demetics ultrasonic intelligent thyroid diagnosis system developed by Deshangyunxing Medical Technology Co.,Ltd,S-Detect of Samsung,etc.Demetics was newly developed by Deshangyunxing Medical Technology Co.,Ltd.This system not only adds modern mathematical theory such as "rotation invariance" into the system but also independently developed the deep learning framework De-Light for ultrasound images based on big data analysis of thyroid nodules,but the diagnostic accuracy of AI has reached the level of which seniority sonographers,whether AI will really improve sonographers’ accuracy to the diagnosis of benign and malignant thyroid nodule,what cause the reason of misdiagnosis of AI products,these problems in the scientific community has not a clear answer.This study intends to make further discussion from these aspects.Purpose This paper aims to compare the accuracy difference between Demetics ultrasonic-assisted diagnosis system and physicians of different seniority in the diagnosis of benign and malignant thyroid nodule,to explore whether Demetic is helpful to sonographers of different seniority in the diagnosis of benign and malignant thyroid nodule,and to analyze the influencing factors of missed diagnosis and misdiagnosis caused by Demetic.Materials and methods This retrospective study included 284 thyroid nodules from 186 patients who underwent thyroid surgery and pathological examination in our hospital.Two-dimensional grayscale ultrasound images of thyroid nodules were collected by a doctor who was not involved in reading the ultrasound images.The patient information was removed from the thyroid nodule images,and each thyroid nodule was numbered.The two-dimensional grayscale ultrasound images of the thyroid nodules were imported into Demetics for image processing and analysis,and the risk coefficient and diagnostic results were recorded.Four sonographer were involved,including two junior sonographers(1,2)with three years of experience in ultrasound diagnosis and two senior sonographers(A,B)with more than 10 years of experience in ultrasound diagnosis;the sonographers were blinded to the pathologic results of the thyroid nodules.The sonographers used the American College of Radiology(ACR)Thyroid Imaging Reporting and Data System(TI-RADS)to score the main characteristics of the thyroid nodules(composition,echo,shape,margin and echogenic foci).The classification of TI-RADS and the judgment of benign and malignant results were obtained respectively,and the pathological results were still unknown after evaluation.The thyroid nodules were reclassified by TI-RADS and adjusted to be benign or malignant by two different combined diagnostic methods.In combined method 1,four sonographers forcibly upgraded or downgraded the original TI-RADS classification by one level when the Demetics classified the thyroid nodules as malignant or benign.The new benign and malignant classification would be determined based on the cut-off value,which was calculated by the new TI-RADS classification,if the cut-off value is less than,it is considered benign;if the cut-off value is greater than or equal to,it is considered malignant.In combined method 2,four sonographers flexibly adjusted the TI-RADS and benign or malignant classification of the thyroid nodules after they learned the Demetics results.The McNemar test was used to compare the sensitivity and specificity,and the Z test was used to compare the area under the curve(AUC).In all analyses,a P value less than 0.05 was considered a statistically significant difference.Results1.General characteristicsA total of 284 thyroid nodules from 184 patients were included in this study;the nodules had a maximum diameter of 0.3～7.1 cm and a mean diameter of 2.6cm.There were 195 benign nodules(68.7%),including 100 cases of nodular goiters,40 cases of thyroid adenomas,52 cases of nodular goiters with adenoma,and 3 cases of thyroiditis with adenoma.There were 89 malignant nodules(31.3%),including 85 cases of papillary carcinoma and 4 cases of follicular carcinoma.2.Comparison between junior sonographers and senior sonographersThe AUC of the two senior sonographers was greater than that of the two junior sonographers.The sensitivity of the two senior sonographers was greater than that of junior sonographer 1(PA1=0.031,PB1=0.049).The specificity of the two senior sonographers was greater than that of junior sonographer 2(PA2=0.001,PB2=0.004).3.Comparison between Demetics and sonographers with different seniority levelsThe sensitivity of Demetics was higher than that of junior sonographer 1(PD1=0.029),but there was no significant difference between Demetics and the two senior sonographers as well as the junior sonographer 2.The specificity of Demetics was lower than that of the two senior sonographers(PDA=0.008,PDB=0.017).Demetics had a higher AUC than the two junior sonographers(PD1=0.042,PD2=0.038),but no significant difference was noted between Demetics and the two senior sonographers.4.Comparison between sonographers before and after applying Demetics methodsAfter using combined method 2,only the AUC of junior sonographer 2 improved(0.802 vs 0.858,P=0.045).After using combined method 1,the sensitivity(61.80%vs 76.40%,P=0.035)and AUC(0.811 vs 0.870,P=0.031)of junior sonographer 1 improved,and the specificity(88.72%vs 96.92%,P<0.001)and AUC(0.802 vs 0.878,P=0.026)of junior sonographer 2 improved.However,the sensitivity,specificity and AUC of the two senior sonographers did not significantly improve.Additionally,no significant differences in AUC were noted between the two senior sonographers without applying Demetics and the two junior sonographers after applying the two combined methods.5.Comparison between the two combined methods for junior sonographersAfter using combined method 1,the sensitivity(P=0.041)of junior sonographer 1 and the specificity(P=0.003)of junior sonographer 2 were both higher than the corresponding values after using combined method 2,whereas there was no significant difference in AUC was noted between the junior sonographers after applying the two combined methods.6.Cases of missed diagnosis and misdiagnosis of the Demeticsandthe sonographers.Among the 284 thyroid nodules that were included in this retrospective analysis,24 nodules were misdiagnosed by both the Demetics and the sonographers.Three malignant thyroid nodules were misdiagnosed only by the Demetics because Demetics erroneously identified the very low echogenic sign as anechoic.In addition,4 malignant thyroid nodules were misdiagnosed exclusively by Demetics because Demetics failed to recognize dotted strong echogenicity in solid-cystic nodules.In addition,4 nodules were not detected because the echo of the 3 nodules was similar to normal thyroid,and 1 nodule was too small.Conclusion Demetics for diagnosis of benign and malignant thyroid nodules with high sensitivity and accuracy,and low qualification through different joint diagnosis method after doctors may improve the diagnostic accuracy of benign and malignant thyroid nodule,shortening the period of the cultivation of the low qualification of ultrasound doctors.The factors influencing Demetics diagnosis including the echo characteristics of thyroid nodules,echo contrast with the surrounding normal tissues and the size of the nodule. |
PDF Full Text Request