Un algoritmo de detección de bacilos de Koch en imágenes de baciloscopía fluorescente basado en máquina de soporte vectorial para telediagnóstico

• Autor: Dianderas Caut, Erwin Junger
• Formato: tesis de maestría
• Fecha de Publicación: 2018
• Institución: Consejo Nacional de Ciencia Tecnología e Innovación
• Repositorio: CONCYTEC-Institucional
• Lenguaje: español
• OAI Identifier: oai:repositorio.concytec.gob.pe:20.500.12390/1694
• Enlace del recurso: https://hdl.handle.net/20.500.12390/1694
• Nivel de acceso: acceso abierto
• Materia: Tuberculosis; Detección de enfermedades; Máquina de soporte vectorial; https://purl.org/pe-repo/ocde/ford#2.02.05

Tuberculosis (TB) is an infectious and contagious disease, but it is preventable and curable. However, it remains one of the leading causes of death by an infectious agent in the world (surpassed only by the Human Immunodeficiency Virus). Although it is true that TB is curable, the detection of this disease remains a major obstacle, as the World Health Organization (WHO) has recommended that a medical technologist should not test more than 20 samples per day. But, for example, in Peru, according to the Ministry of Health, in 2014, there were about one and a half million people who were candidates to be carriers of the bacterium, would have required 300 technologists to work for approximately 250 days only analyzing sputum samples (the most used method for detecting tuberculosis), a fact that was not given. For the automatic detection of TB, some algorithms have already been developed, which although they have encouraging results, only use digitized images using the direct preparation method. In addition, only one technologist is considered for validation, when at least two is ideal given that the criteria for evaluating the bacillus are very subjective. Finally, they work under very particular conditions (specific types of microscopes) that cannot be replicated in Peru because of a cost issue. It is based on this problem, that it is proposed to develop computational algorithms capable of processing sputum samples for the detection of bacilli (regardless of method of sample preparation and / or environment), in order to grant greater elements of judgment to medical technologists and thus be able to help improve the diagnostic quality of Koch bacilli. In order to identify the Koch bacilli within the images obtained by fluorescent fluoroscopy, a series of algorithms was first developed, which depending on the type of sample preparation (direct, pellet or diluted pellet) allowed to eliminate the background in order to obtain the candidate objects. Then descriptors were implemented (hu moments, geometric and photometric descriptors), which would serve as input to train a vector support machine (SVM) which would allow to discern if the object analyzed is a bacillus or not. In order to validate the algorithms developed, the support of two medical technologists was given. They served as a reference to validate approximately one thousand candidate objects in order to obtain the percentages of sensitivity and specificity. The values obtained for any sample preparation method exceeded the 90% threshold, which allows to affirm that the work developed can be used as a means of helping to make decisions about the presence or absence of bacilli.