In this chapter, an attempt has been made to automate the analysis of positive and negative Tuberculosis (TB) sputum smear images using multifractal approach. The smear images (N=100) recorded under standard image acquisition protocol are considered. The images are subjected to multifractal analysis and the corresponding spectrum parameters are extracted. Most significant parameters are selected based on the principal component analysis. Further, these parameters are subjected to classification using support vector machine classifier with different kernels. Results demonstrate that the multifractal analysis is capable of discriminating positive and negative TB images. The values of apex, broadness and aperture of the singularity spectrum are higher for TB positive than negative images and are statistically significant. The performance estimators obtained in the classification process show that the polynomial kernel performs better. It appears that this method of texture analysis could be useful for automated analysis of TB using digital sputum smear images.
TopIntroduction
World Health Organization (WHO) has declared Tuberculosis (TB) a global emergency, as it is an airborne infectious disease. TB is a major cause of illness and death in many countries, especially in Asia and Africa (WHO Tuberculois fact, 2010). TB is caused by Mycobacterium tuberculosis which is an Acid Fast Bacillus (AFB). The bacilli most frequently infect the lungs causing pulmonary tuberculosis.
There are several diagnostic techniques which include both invasive and non-invasive procedures. Smear microscopy, culture and chest radiography are the less expensive non-invasive diagnostic procedures (Sotaquira, Rueda, & Narvaez, 2009). Among the non-invasive tests, microscopic examination of stained sputum smears still remains a cornerstone for the diagnosis of pulmonary tuberculosis throughout the world (Steingart, Henry, Ng, Hopewell, Ramsay, Cunningham, Urbanczik, Perkins, Aziz, & Pai, 2006). The diagnostic method of smear microscopy involves manual examination which is labour intensive and may lead to high false negative rate. Further, repeated tests are needed to be performed for early detection of the disease (Crnčević-Urek, Stipić-Marković, Kardum-Skelin, Stipić, Crnek-Kunstelj, & Urek, 2002). Hence there is a need to automate the diagnostic process for improvement in the sensitivity and accuracy of the test. It also entails substantial reduction in the labour workload despite large number of images analyzed (Veropoulos, Campbell, Learmonth, Knight, & Simpson, 1998).
The smear images are obtained by either using a fluorescence microscope for auramine stained smear or a bright field microscope for Ziehl–Neelsen stained sputum smear (Khutlang, Krishnan, Whitelaw, & Douglas, 2010). Mobile phone based microscopy systems enable greater access to high quality health care monitoring system especially for TB patients. This system support the WHO’s Directly Observed Treatment, Short course program, which guide TB eradication efforts (Breslauer, 2009).
Several image analysis techniques have been reported for the automatic identification and classification of sputum smear samples. Existing methods are based on colour and shape analysis of TB images. The shape of the bacillus alone is not a discriminable feature because the bacilli may be faint, occluded, obscured by cells or remnants, or inside macrophages. This imparts a hazy outline to the bacilli, which may cause oversights in recognition (Veropoulos, Campbell, Learmonth, Knight, & Simpson, 1998; Forero, Cristobal, & Desco, 2006). Also colour alone cannot distinguish the stained areas from the fine bacteria because colour spectrum of bacilli is quite wide making it indistinguishable from debris (Tadrous, 2010).
During the past decade, results from numerous published articles have shown the ability of texture analysis algorithms to extract diagnostically meaningful information from medical images (Lopes & Betrouni, 2009). But texture analysis of microscope images faces problem which depends on the level of observation (magnification of the microscope) (Beil, Irinopoulou, Vassy, & Wolf, 1995). This problem of scaling is overcome by the fractal-based texture analysis as fractal is invariant to geometrical changes which includes view point changes and non-rigid deformations of the texture surface (Xu, Ji, & Fermuller, 2009). Fractal is shown to be a tool that describes the fractal geometry and allows capturing that is not available in traditional geometrical representation of shapes (Lopes & Betrouni, 2009).