Comparative MR image analysis for thyroid nodule detection and quantification

Nodular disease of the thyroid gland is widespread that affects more women than men. The term thyroid nodule refers to any abnormal growth that forms a lump in the thyroid gland which is located low in the front of the neck, below the Adam's apple. It is shaped like a butterfly and wraps around the windpipe or trachea. In this paper, we have proposed two segmentation methods for thyroid nodule detection by using magnetic resonance imaging (MRI). These methods are region based active contour (RBAC) and single seeded region growing (SSRG) methods. The pre-processed MR images are segmented and obtained cross sectional areas of nodules are compared with those which are manually identified nodule areas by trained readers. Detection success of both methods is calculated by comparing their segmentation results with the manually obtained one. For this purpose, Correlation Coefficient (CC) and Zijdenbos Similarity Index (ZSI) have been used to have accuracy measures of two methods. Although both methods have acceptable ZSI values, RBAC method has yielded (average 0.938) higher ZSI values than SSRG method (average 0.906). Similar results are obtained by calculating average cross-correlation values (RBAC method: 0.919 and SSRG: 0.911). Also two error analysis methods, namely Area Error Rate (AER) and Overlap Error (OE) rate employed. RBAC method has yielded (average error rate 9.3%) lower AER values than SSRG method (average error rate 18%). Similar results are obtained by calculating average error rate values (RBAC method: 8.9% and SSRG: 16.3%) by using OE. According to the analysis results it can be concluded that RBAC method gives better segmentation accuracy rates in MRI thyroid nodule detection. High accuracy rates are achieved with the algorithms implies that the proposed study can be suitable and have high capability to be used as an image extraction technique that would assist radiologists and otorhinolaryngologists in the thyroid nodule screening by providing accurate value of size. (C) 2013 Elsevier Ltd. All rights reserved.