The Applicative Study Of Using Multimodality Ultrasond Imaging To Diagnose Thyroid Nodules

Part I Mathematical model of Malignant and benign thyroid nodulesObjectives: The incidence of thyroid nodules increased year by year, but the technology to identify benign and malignance through ultrasound images has little progress. If we choose by surgery, then the harm to the patient’s body can not be ignored. Because of thyroid nodules in 95% of benign and malignance of only 5%, to improve the identification technology of thyroid nodule ultrasound image is imperative. With the rapid development of Biomedical image processing technology, and continuous improvement of high-speed, large-capacity, multi-function hardware and the theory of numerical compute, it is possible to simulate thyroid nodules by the technologic and method. And to study the characteristics of benign and malignant with thyroid nodules, then the results will be processed and analyzed. Numerical simulation as a means of combining research and prediction can provide the detail of numerical parameters by using different ways for medical research.Methods:First we have read each gray scale value of the 2-D ultrasound image of thyroid nodule, which includes 415 benign nodules and 131 malignant nodules. To find the salient features,which can distinguish between benign and malignant nodules, we made rank-sum test on the gray scale values of these nodules. Then using these salient features creates multiple classifiers. Finally, we have found the best classification model by comparing the accuracy and coverage of different classifiers; therefore, our work will provide a scientific basis for the diagnosis of thyroid nodules.Results:This research is cross-disciplinary study including medicine, mathematics, and computer science. It involves innovative algorithms thinking and different classification models. By using the k-mer string algorithm, we can not only find a comprehensive and detailed grayscale image features, but also improve the accuracy of classification. To improve the speed of classification, we use rank-sum test to find significant features between benign and malignance. Salient features can be used to build different classifiers, and then we can get the optimal classifier LIBSVM by comparing the accuracy and coverage of different classifiers. This classifier for benign and malignance has high degree of accuracy and coverage. We can determine the benign and malignance of thyroid nodules in ultrasound image more quickly and accurately by using LIBSVM.Conclusions:The accuracy and coverage of LIBSVM-classifier has been further improved by increasing the sample and making equalization process of the sample size of benign and malignant thyroid nodules. LIBSVM-classifier not only provides a fast and accurate technical methods and ideas of diagnosing benign and malignant thyroid nodules for doctors and researchers, but also has a significant role in promoting the clinical diagnosis of thyroid nodules. Part II Contrast-enhanced ultrasound and real-time elastography in differential diagnosis of malignant and benign thyroid nodulesObjectives: The diagnostic values of contrast-enhanced ultrasound(CEUS) or real-time elastography(RTE) alone, as well as a combination of CEUS and RTE in distinguishing benign from malignant thyroid nodules were investigated.Methods:Between Sept 2012 and Mar 2015, a total of 97 consecutive patients(50 male and 47 female; mean age, 45.9 ± 12.8; age range, 17-76 years) with thyroid nodules referred for surgical treatment were examined by CEUS and RTE. The final diagnosis was obtained from histological findings. Image analysis of CEUS and RTE were performed. Taking postoperative pathological results as golden standard, receiver operating characteristic(ROC) curve was draw. The sensitivity, specificity, positive predictive value(PPV), negative predictive value(NPV) and accuracy of CEUS alone, RTE alone and CEUS+RTE combination were calculated, respectively.Results:Pathological examination showed 66 papillary carcinomas and 43 benign lesions including 21 adenoma and 22 nodular goiters. The sensitivity, specificity, PPV, NPV and accuracy of CEUS were 81.82%, 90.70%, 93.10%, 90.70% and 85.32%, respectively. In case of RTE, the sensitivity, specificity 88.37%, PPV, NPV and accuracy were 80.30%, 88.37%, 91.38%, 88.37% and 83.49%, respectively. CEUS+RTE combination had a sensitivity of 95.45%, specificity 95.35%, PPV96.92%, NPV95.35% and accuracy 95.41%, respectively.CEUS+RTE combination showed a significantly higher sensitivity and specificity compared to either CEUS or RTE alone(all P < 0.05). The area under the ROC curve(AUC) for CEUS, RTE and CEUS+RTE combination were 0.883, 0.863 and 0.959, respectively. The AUC of RTE alone was significantly lower than that of CEUS+RTEcombination.Conclusions: Our results show that CEUS+RTE combination significantly increases the diagnostic performance for differential diagnosis of malignant and benign thyroid nodules compared to either CEUS or RTE alone. Part III Association study of the ultrasonographic characteristics in thyroid carcinoma and cervical lymph node metastasisObjectives: To determine ultrasonographic characteristics in thyroid carcinoma(TC), and to explore the diagnostic efficacy of these ultrasonographic characteristics for predicting cervical lymph node metastasis(LNM)Methods:From Jan 2010 to Jan 2015, a total of 186 TC patients were recruited from the Second and the Third Hospital of Hebei Medical University. We divided them into two groups including metastatic group(129, nodules; n = 86) and nonmetastatic group as the control(117 nodules; n = 100).The univariate analysis and multivariate analysis were used to evaluate the relationship of ultrasonographic characteristics and cervical LNM. The Doppler ultrasound was employed to estimate peak systolic velocity(PSV), pulsatility index(PI) and resistive index(RI). ROC curve was drawn to evaluate the efficacy of ultrasonographic characteristics for predicting cervical LNM.Results : The sensitivity, specificity, positive predictive value, and negative predictive value of ultrasonographic diagnosis was 81.40%(105/129), 92.32%(108/117), 92.11%(105/114) and 81.82%(108/132), respectively. The cervical LNM in TC was frequently occurred in cervical level VI(37.98%), and mainly located in either middle pole(46.51%) or lower pole(41.09%).The PSV and RI values in metastatic group were significantly higher than those of nonmetastatic group(both P < 0.001).Multivariate analysis showed that nodular diameter, contact of capsular invasion, microcalcification and flow grade were risk factors for TC patients with cervical LNM(all P < 0.05). Furthermore, ROC curve analysis demonstrated excellent accuracy of nodular diameter, capsular invasion, microcalcification and flow grade for predicting cervical LNM.Conclusions:Our study concluded that ultrasonographic characteristics in TC may predict cervical LNM including maximum nodular diameter, capsular invasion, microcalcification and flow grade.