Automatic Recognition of Mycobacterium Tuberculosis Based on Active Shape Model

Chao XU  Dongxiang ZHOU  Tao GUAN  Yongping ZHAI  Yunhui LIU  

Publication
IEICE TRANSACTIONS on Information and Systems   Vol.E99-D   No.4   pp.1162-1171
Publication Date: 2016/04/01
Online ISSN: 1745-1361
DOI: 10.1587/transinf.2015EDP7253
Type of Manuscript: PAPER
Category: Pattern Recognition
Keyword: 
object recognition,  tuberculosis,  active shape model,  watershed,  segmentation,  

Full Text: PDF(2MB)>>
Buy this Article




Summary: 
This paper realized the automatic recognition of Mycobacterium tuberculosis in Ziehl-Neelsen stained images by the conventional light microscopy, which can be used in the computer-aided diagnosis of the tuberculosis. We proposed a novel recognition method based on active shape model. First, the candidate bacillus objects are segmented by a method of marker-based watershed transform. Next, a point distribution model of the object shape is proposed to label the landmarks on the object automatically. Then the active shape model is performed after aligning the training set with a weight matrix. The deformation regulation of the object shape is discovered and successfully applied in recognition without using geometric and other commonly used features. During this process, a width consistency constraint is combined with the shape parameter to improve the accuracy of the recognition. Experimental results demonstrate that the proposed method yields high accuracy in the images with different background colors. The recognition accuracy in object level and image level are 92.37% and 97.91% respectively.