Effective segmentation of orphan annie-eye nuclei from papillary thyroid carcinoma histopathology images using a probabilistic model and region-based active contour

The presence of orphan annie-eye nuclei is a significant feature for the diagnosis of papillary thyroid carcinoma (PTC), a cancer of the thyroid gland. Automated detection and segmentation of orphan annie-eye nuclei from histopathology imagery is an intricate procedure due to traditional and specific challenges. The specific challenges are posed by the biological properties of these nuclei. This paper propositions an automated method to detect and segment orphan annie-eye nuclei from papillary thyroid carcinoma histopathology images. Our proposed method (EM/MPM-CV) initially uses a Markov random field-based segmentation technique to detect the orphan annie-eye nuclei seeds from the given images. A region-based active contour model (ACM) is initialized and evolved over the nuclei seeds to identify the final nuclei contours. The EM/MPM-CV method is evaluated on 149 PTC histopathology images for detection and segmentation performance. This technique gives a detection sensitivity of 87% and positive predictive value of 93%. The directed Hausdorff distance (DHD) and the mean absolute distance (MAD) values for the proposed method are found to be 3.79 and 1.55 pixels respectively.