Background: The thyroid is one of the largest and most important glands in the endocrine system. It controls cellular metabolism by releasing hormones directly into the bloodstream. However, thyroid hormone disorder may lead to depression, fatigue, memory loss, and much more. Therefore, in order to assess thyroid disorder, gamma scintigraphy is extensively used to diagnose the malfunction of the thyroid. The current study proposed a simple computer-aided diagnosis technique to detect any nodular abnormalities in the thyroid.
Methods: The proposed technique involves suppressing noise in thyroid scans, segmentation of thyroid glands from the Technetium-99m (99mTc) scan, and the use of an support vector machine classifier to classify thyroid scan into normal and abnormal based on the extracted features. The local binary pattern technique was employed to extract features of normal and nodular thyroid scans.
Results: Performance of classifiers was optimized in such a way that areas under the curve of receiver operative curve was maximized.
Conclusion: An easy-to-use graphical user interface was developed for the validation of the proposed technique on the NORIN thyroid database which gave results with an accuracy of 92%.
Key words: CAD, AUC, ROC, MATLAB in nuclear medicine, thyroid scintigraphy, binary masking, image feature extraction
|
