Effective segmentation and classification of thyroid histopathology images

详细信息    查看全文

• 作者：J. Angel Arul Jothi ^{jothi@cs.annauniv.edu" class="auth_mail" title="E-mail the corresponding author} ; V. Mary Anita Rajam ; ^{anitav@annauniv.edu" class="auth_mail" title="E-mail the corresponding author}
• 关键词：Computer Aided Diagnosis (CAD) ; Multilevel thresholding ; Segmentation ; Feature extraction ; Feature reduction ; Rule generation ; Classification ; Particle Swarm Optimization (PSO) ; Variable Precision Rough Sets (VPRS) ; Histopathology image
• 刊名：Applied Soft Computing
• 出版年：2016
• 出版时间：September 2016
• 年：2016
• 卷：46
• 期：Complete
• 页码：652-664
• 全文大小：2701 K

文摘

This paper proposes a Computer Aided Diagnosis (CAD) system that semi-automatically segments and classifies H&E-stained thyroid histopathology images into two classes: Normal Thyroid (NT) or Papillary Thyroid Carcinoma (PTC) based on nuclear texture features. Our system segments the given histopathology image into different binary images using Particle Swarm Optimization (PSO)-based Otsu's multilevel thresholding. From the segmented binary images, a binary image containing the nuclei is chosen manually. Nuclei are extracted from the manually selected binary image by imposing an area constraint and a roundness constraint. The intensity variations of pixels within the nuclei are quantified by extracting texture features. Variable Precision Rough Sets (VPRS)-based β-reduct is used to identify redundant features and generate rules. The rules are then stored in a rule base. A novel closest-matching-rule (CMR) algorithm is proposed to classify a new test sample as PTC or NT using the rules in the rule base. We verified experimentally that the proposed CAD system provides promising results and it is supposed to assist pathologists in their decisions.