SINPATCO - Sistema Inteligente para Diagnóstico de Patologias da Coluna Vertebral

• Ano de defesa: 2006
• Autor(a) principal: Rocha Neto, Ajalmar Rêgo da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Teleinformática; Telemedicina; Inteligência artificial; Redes neurais (Computação)
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/15977

This dissertation presents the results obtained from a computer-aided medical diagnostic system implemented through statistical and neural pattern classifiers. The Intelligent System for Diagnosis of Pathologies of the Vertebral Column (SINPATCO) has a modular architecture and is composed of three subsystems, namely: graphical interface, classification of pathology, and knowledge extraction. The graphical interface module allows a friendly man-machine interaction with the physician. The pathology classification module is implemented through difierent algorithms, such as linear and quadratic discriminants, Naive Bayes classifier, K Nearest Neighbors (KNN) classifier, Multilayer Perceptron (MLP) network, Self-Organizing Map (SOM) network, ang Generalized Regression network (GRNN). The knowledge extraction module is responsible for rule extraction from trained neural network based classifiers, in order to elucidate the neural-based diagnostic to the orthopedist. In particular, the pathology classification module of the SINPATCO platform uses recently proposed biomechanical attributes to categorize a patient into one out of three classes: normal subjects, subjects with spondilolistesis, and subjects with disk hernia. All the aforementioned classifiers are evaluated with respect their pathology recognition rate, number of false positive cases, number of false negative cases and sensitivity to outliers. The contribution of this work is manifold. Starting from the fact that it is probably the first to use (within the orthopaedic medicine) a recently proposed set of biomechanical measurements for the design of classifiers, this work also evaluates several pattern classifiers in the diagnosis of patologies of the vertebral column, and allows knowledge extraction from the trained classifiers in order to elucidate the obtained diagnostic to the physician. To the best of our knowledge, the combination of these three contributions makes the SINPATCO platform an innovative computer-aid tool for the orthopedist, facilitating the work of these professionals. Despite the fact that the SINPATCO platform can serve as a computer-aided diagnostic tool in the orthopedic medicine, it can also be used by non-expert clinicians, in order to minimize the lack of orthopedists in remote regions, speeding up the treatment and the transferring of patients to more developed centers.