Nuclei Segmentation of Microscopy Images of Thyroid Nodules via Active Contours and K-means Clustering

Purpose: In this work the problem of automatic segmentation of nuclei in cytological images of thyroid nodules is investigated. Materials and Methods: 50 Fine Needle Aspiration images of thyroid nodules were digitized (768x576x8 bit) using a light microscopy imaging system comprising a Zeiss Axiostar plus microscope connected to a Leica DC 300 F color video camera. An automatic segmentation algorithm was designed combining the k-means clustering algorithm and active contours. The k-means algorithm is used to classify all pixels in a given image into nuclei or surrounding tissue, in order to provide an initial gross estimation of nuclei regions. Starting from the boundaries of these regions, an active contour snake fired and propagated till converging to nuclear boundaries. Results: On average, 94% of all nuclei were correctly delineated according to a histopathologist evaluation. Conclusions: The algorithm might be of value for computer-assisted microscopy systems, since accurate nuclei segmentation enables the accurate quantification of DNA content that may potentially allow the prediction of the disease course.