The Value Of Ultrasonic Histogram And Deep Learning In The Differential Diagnosis Of Medullary Thyroid Carcinoma And Thyroid Adenoma

Part Ⅰ:Value of Gray-scale Histogram Analysis Based on Two-dimensional Ultrasound Images in Differential Diagnosis of Medullary Thyroid Carcinoma and Thyroid AdenomaObjective To investigate the feasibility and value of histogram analysis based on twodimensional gray-scale ultrasonography in the differential diagnosis of medullary thyroid carcinoma(MTC)and thyroid follicular adenoma(FTA).Methods The preoperative ultrasound images of 86 newly diagnosed MTC patients and 100 FTA patients treated in the Cancer Hospital of Chinese Academy of Medical Sciences from January 2015 to October 2021 were collected.The boundaries of the lesions were manually sketched by two ultrasound doctors with more than 5 years experience in ultrasonic diagnosis of thyroid diseases one by one,and the gray histograms were performed based on the regions of interest(ROIs)delineated manually by two radiologists,thereafter,mean,variance,skewness,kurtosis,percentiles(1st,10th,50th,90th,99th)were generated.The histogram parameters between the MTC group and the FTA group were compared,and the independent predictors were screened by multivariate logistic regression analysis.ROC analysis was used to compare the individual diagnostic efficacy and joint diagnostic efficacy of independent predictors.Results Multivariate regression analysis showed that mean,skewness,kurtosis and 50th percentile were independent factors.The skewness and kurtosis in the MTC group were significantly higher than thoses in the FTA group,and the mean and 50th percentile were significantly lower than thoses in the FTA group.The area under the individual ROC curve of mean,skewness,kurtosis and 50th percentile is 0.654-0.778.The area under the combined ROC curve is 0.826.Conclusion The gray histogram analysis based on two-dimensional sonogram has certain application value in the differential diagnosis of MTC and FTA,in which the joint diagnosis value of mean,skewness,kurtosis and 50th percentile is the highest.Part Ⅱ:Deep learning based on ultrasound to differentiate pathologically proven atypical and typical medullary thyroid carcinoma from follicular thyroid adenomaObjectiveTo investigate the feasibility and value of deep learning based on grayscale ultrasonography in the differentiation of pathologically proven atypical and typical medullary thyroid carcinoma(MTC)from follicular thyroid adenoma(FTA).MethodsThe preoperative 770 ultrasound images consisted of 354 MTCs(66%were typical MTCs with a high suspicion sonographic pattern,34%were atypical MTCs with a suspicion pattern of intermediate or less)and 416 FT As.All images were delineated manually by a senior sonographer to achieve the regions of interest.All images are divided into training set and test set according to the proportion of 9:1.Two deep neural networks of ResNet-34 and ResNet-18 were performed on the training set(n=690).The test data set(n=80)was subsequently evaluated by the two models and two sonographers,their diagnostic performances and misdiagnosis lesions were compared and analyzed.The misdiagnosed cases of two deep learning models and two sonographers were analyzed.The calculation time of two deep learning network models was recorded,and the diagnostic accuracy of different data sources in the same deep learning model was compared.ResultsThe ResNet-34 model shows higher diagnostic ability than the senior sonographer with an area under the receiver operating curve of 0.992(95%CI:0.840-0.970)versus 0.838(95%CI:0.739-0.911)and the junior sonographer 0.992(95%CI:0.840-0.970)versus 0.754(95%CI:0.645-0.843).Moreover,12 of 16 atypical MTCs were successfully identified by the ResNet34,which is significantly better than the senior and junior sonographer,suggesting that these patients could benefit from timely serological examination and surgical strategy at an earlier stage.It takes 1.09 seconds for ResNet-34 model to calculate a ROI,and only 0.69 seconds for ResNet-18 model.In the test set,the diagnostic accuracy of ResNet-34 for GE images is 93.5%,and the diagnostic accuracy of Philips images is 72.2%.ConclusionsDeep learning to differentiate MTC from FTA on grayscale ultrasound may be a useful diagnostic support tool,especially in atypical MTC and FTA.Moreover,the computing time of deep learning is short,which will help to incorporate it into real-time ultrasound diagnosis.