Medico

Journal of Microscopy, Vol. 223, Pt 2 August 2006, pp. 120– 132 Received 20 July 2005; accepted 10 April 2006

Automatic identification of Mycobacterium tuberculosis by Gaussian mixture models
Blackwell Publishing Ltd

M . G . F O R E RO *, G . C R I S T Ó B A L † & M . D E S C O ‡
*School of Biosciences, University of Birmingham, U.K. †Instituto de Optica (CSIC), Serrano 121, 28006 Madrid,Spain ‡Hospital General Gregorio Marañon, Madrid, Spain

Key words. Bayesian classification, clustering, colour, feature extraction, fluorescence microscopy, Gaussian mixture models, tuberculosis.

Summary Tuberculosis and other kinds of mycobacteriosis are serious illnesses for which early diagnosis is critical for disease control. Sputum sample analysis is a common manual technique employedfor bacillus detection but current sample-analysis techniques are time-consuming, very tedious, subject to poor specificity and require highly trained personnel. Image-processing and pattern-recognition techniques are appropriate tools for improving the manual screening of samples. Here we present a new technique for sputum image analysis that combines invariant shape features and chromatic channelthresholding. Some feature descriptors were extracted from an edited bacillus data set to characterize their shape. They were statistically represented by using a Gaussian mixture model representation and a minimal error Bayesian classification procedure was employed for the last identification stage. This technique constitutes a step towards automating the process and providing a high specificity.Introduction Mycobacterium tuberculosis bacilli are the aetiology of pulmonary tuberculosis (TB) disease. These organisms can also infect other organs or tissues such as brain, kidneys, bone and skin. TB is the main cause of death produced from an infectious illness. According to the World Health Organization, one-third of the world’s population (1722 million people) are carriers of these bacteria,originating 10 million cases of active TB worldwide and approximately 3 million deaths per year (Ginsberg, 1998). Some studies conducted in Spain have shown that cigarette smoking is a risk factor for pulmonary TB in young people (Alcaide et al., 1996). An epidemiological study of TB in the U.S.A. presents a Markovian model nature showing an increase in
Correspondence to: Dr Gabriel Cristóbal.Tel: 34915616800; fax:

34915645557; e-mail: gabriel@optica.csic.es

the previous decade and then a decrease (Debanne et al., 2000). Two reasons for the increasing incidence of this disease are human immunodeficiency virus infection (in fact TB is one of the main causes of death in human immunodeficiency virusinfected people) and multidrug resistance. The manual identification of TB bacilli isroutinely performed in sputum smears dyed with fluoroscope auramine, using a fluorescence microscope. The non-invasive characteristics of the sputum procedure are important because repeated examinations are needed for an early detection of the disease (Crncevik-Urek et al., 2002). However, this technique lacks sensitivity and consequently clinicians must wait as long as 2 months for culture resultsbecause these bacilli take from 5 to 20 h to duplicate themselves. Manual screening for the bacillus identification is a labour-intensive task with a high falsenegative rate (Veropoulos et al., 1999). Automatic screening has several advantages, such as a substantial reduction in clinician’s workload, improved test sensitivity and a better diagnostic accuracy by increasing the number of images that canbe analysed by the computer. However, the segmentation of particular species of bacteria entails a complex process. The shape of the bacillus is not enough to be used as a discriminant feature because other particles and species of bacteria share the same morphology. Therefore, besides the shape of the bacillus, it is also necessary to consider its colour to improve discrimination accuracy....