Software para la Clasificación de Fibras Musculares en Imágenes Histológicas

Abstract

Histochemical stains are used for the characterization of the fibers types present in skeletal muscles of organisms. Usually, the identification and classification of these fibers is carried out visually by an expert researcher, which entails a considerable investment of resources and time. Because of that, a computational system was developed wich allows the automation of the classification of the muscle fibers in a shorter time and with high efficiency. For that purpose we used computational tools such as data mining algorithms, artificial intelligence and pattern recognition, implemented in the programming language Java, which guarantees multiplatform support. The algorithms used are K-means, Fuzzy c-means, Kohonen self-organized maps and an algorithm supervised by an expert. Because of its high compatibility with other plataforms, the results obtained by this computational system can be stored in spreadsheets to perform other types of analysis, for example to compute the fractal dimension of a group of fibers of the same type and thus determine whether they are distributed randomly or not. The spreadsheet data is used later to determine distribution patterns and organization of muscle fibers, both under normal and pathological conditions. The results obtained from the different analyzes of the developed computer system indicated a good degree of efficiency, reducing processing time by up to 90 %, although expert assistance improves its effectiveness.