Determinación de patologías a partir del procesamiento digital de centellografías

JUANA ARMESTO; AGUSTINA MILDENBERG; CAROLINA MALDONADO

Carlos Paz

Congreso; 7mo Congreso ALFIM & 13avo Congreso SAFIM.; 2016

Sociedad Argentina de Fisica Médica

This work shows the development of an Algorithm based on Images obtained from a Myocardic Perfusion Study of a specific Patient. The Algorithm performed an automatic classification in four categories, related with the presence or not of coronary diseases suffered by the patient. These categories are: Normal, Ischemia, Heart Attack, or a combination of the last two. The development of the project demanded an interdisciplinary team, involving Bio - Images acquisition technicians, Cardiologist Physicians, from the Nuclear Service of Medicine of the Cordoba Hospital. The multidisciplinary environment added a rich point of view for the analysis of the present job. The development of the algorithm, started with a Image that contains graphic representation of the Left Ventricle Short Axle, named, Polar Map. The Polar Map was acquired in relax and stress way. Both maps allowed to build of a third Map called Net Polar Map. The last map register the difference between the two original maps. The Key information was supplied by these Images, defining particular characteristics, for instance: quantity of healthy tissue and the position of cardiac illness. Lately, an Associated Vector of characteristic was built for each patient. Those vectors, allowed the classification using a Multilayer Perceptron Neural Network. The output of the Neural Network classified the type of Pathology suffered by the patient, such as: Normal, Ischemia, Heart Attack or the combination of the last two.Complementary with the Classification, the Algorithm offered additional information, detecting the position of the illness and the Coronary Artery affected. In order to identify the damaged area, a segmentation was performed, according to the Cedar Sinai Method, who divides the heart in 17 segments. Each segment has an reference at Coronary Vascular Level. The proposed algorithm was implemented in MATLAB. The neutral Network, was trained using 75 images. The validation of the net was performed with another 50 images. The database of Scintigrafic Myocardic Perfusion was obtained from a Gamma Chamber at the Cordoba Hospital. The verification of the development to determine Pathologies obtained a Global Precision Index of 86.6 % and a Kappa Index of 0.812. The determination of the Artery affected obtained 91.4% and 0.869. These indexes showed that the algorithm is valid.