Nuclei Location Enhancement and Cytoplasm Segmentation Based on Statistical Active Contour Model for Nuclear-to-Cytoplasmic (NC) Ratio Analysis in Third Harmonic Generation Microscopy Image

Abstract

For the purpose of establishing a computer-aided diagnosis protocol to provide quantitative and objective bio-information a new cell segmentation algorithm for biomedical image analysis system is proposed and the algorithm could be effectively applied to process plenty of changeable in vivo virtual biopsy images provided by healthy individuals The proposed approach includes cautious identification of the nuclei position by considering both virtual staining intensity and ellipse shape feature examining based on the Hough transform besides combining the knowledge of real cell size to quantify cellular structure to increase the accuracy of nuclear recognition thus achieving nuclei location enhancement For cytoplasm boundary extraction we propose a novel approach called the statistical pressure snake which is an optimal parameter setting snake driven by a pressure force that measuring the local statistics similarity between the snake contour movement and the image data of cytoplasm In this new snake model a convergence control factor and a shifting parameter are introduced to overcome the well-known drawbacks of initialization and parameterization on active contour model Experimental results show that the aforementioned algorithm is not only demonstrated having high accuracy for cell positioning and cytoplasm segmentation but also a systematic and adaptive approach with ability to process various kinds of cellular images Moreover the proposed approach also reveals potential for non-invasive analysis of cell Nuclear-to-Cytoplasmic ratio (NC ratio) which is significant for the differentiation of skin disease with abnormal NC ratio in clinical diagnosis

Date of Award	2015 May 18
Original language	English
Supervisor	Gwo-Giun Lee (Supervisor)