Purpose: Fine needle aspiration biopsy remains the most effective way for the detection of malignant thyroid nodules. Nevertheless, the requirement for a highly qualified cytologist and the possibility of repeated biopsy to gain adequate results limit its use. Accurate classification of benign and malignant thyroid nodules is needed to reduce patientsâ€™ risk and reduce medical care costs.
Methods: In this abstract, a thyroid nodule classification method in ultrasound images is proposed based on support vector machine (SVM). Pre-processing steps were as follows: 128*128-pixel images containing thyroid nodule was extracted from each of the original image to remove annotations and markers. Artifacts were removed for the re-sized image. Data normalization was performed. After pre-processing, dataset was separated into training set and testing set. An SVM classifier with a Gaussian kernel function was adopted for thyroid nodule classification. A binary label with -1 for benign nodules and 1 for malignant nodules was used according to the needle aspiration biopsy results. Optimal parameters for the SVM classifier were obtained through 5-fold cross-validation on the training set. The classification process consisted of two steps: training and testing. During the training step, SVM found a decision boundary that separated the training set in the input space using their class labels. After the decision function is determined from the training set, it was used to predict the class label of the test data. Holdout cross-validation was used to evaluate the performance of the classifier.
Results: Results showed that the proposed SVM based classifier can accurately diagnose benign and malignant thyroid nodules with an accuracy of 84.62%, sensitivity of 100%, and specificity of 83.3%. The area under the ROC curve (AUC) of the SVM classifier was 0.9286.
Conclusion: The results demonstrate the potential clinical applications of SVM in the differential diagnosis of benign and malignant thyroid nodules.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by National Key Research and Development Program of China (2016YFC0103400), Key Research and Development Program of Shandong Province (2017GGX201010), Jianfeng Q. was supported by the Taishan Scholars Program of Shandong Province (TS201712065).