Fully automated computer-aided volume estimation system for thyroid planar scintigraphy

Patient-specific dosimetry calculations are often performed for radioiodine therapy in patients with Graves′ hyperthyroidism. The radioiodine doses are typically calculated to deliver the desired amount of radiation based on gland size and radioactive iodine uptake. Thus the estimation of thyroid gland volume is of great importance for radioiodine therapy. In clinical practice, thyroid volume determinations are usually performed with ultrasonography (US) or with planar scintigraphy (PS). In traditional planar scintigraphic studies, the thyroid boundary is estimated using a fixed threshold value if the shape of the thyroid is well-defined or a manually drawn region of interest (ROI) if the thyroid shape is irregular. The thyroid volume is then calculated based on the area thus determined. Delineating the thyroid area on a planar scintigram is not easy when applying a fixed threshold value. Moreover, hand-drawn ROIs are time consuming, tedious, and highly operator-dependent. In this study, for a PS image, a fully automated thyroid volume estimation system mainly consisting of four steps, i.e. preprocessing, image contrast enhancement, image segmentation, and automated ROI finding, was proposed to obtain the maximum height and area of each thyroid lobe, and thus calculate the thyroid volume using either Himanka-Larsson′s formula or Allen-Goodwin′s formula. A set of 40 Graves′s disease patients regarded as training set were used to determine empirically some parameters operated in the system. A set of 30 Graves′s disease patients being independent of the training set, regarded as test set for thyroid volume measurements were used for comparisons and performance analyses. In this study, the US was adopted as a standard reference. The statistical analyses were performed with bias, precision, and relative differences. The results of thyroid volume estimation from the proposed approach correlated well with those from US, and the statistical performance analyses showed good agreement between them. In comparison, our automated approach with Allen-Goodwin′s formula had not only good correlation with US (R2=0.99) but also the best bias (0.8), precision (±2.32 ml), and low relative differences (2.2±6.1%). It is expected that this automated computer-assisted approach can help physicians in the determination of patient-specific administered activities for treatment of thyroid disease.