Avaliação de modelos computacionais aplicados na classificação de hemorragias intracranianas pós-traumáticas

Abstract

Traumatic brain injuries (TBIs) are a problem for society as they often result in post-traumatic intracranial hemorrhages (ICHs) responsible for high morbidity and mortality. Its outcomes are even worse when they occur in small municipalities, far from large centers, whose hospital infrastructure does not have specialized professional teams. In these places, the recognition of these pathologies is sometimes delayed, or even not carried out, due to the need to transfer patients to other hospitals. Some of these locations have resources for the acquisition of computerized tomographs, essential for the diagnosis of ICHs, but they have insufficient budgets for the maintenance of permanent professional staff. Therefore, the development of a computational tool capable of recognizing HICs based on the automated classification of tomographic images performed at these locations would be very useful. Classifiers based on Artificial Neural Networks (ANN), SVM (Support Vector Machines) and KNN, are already widely used as proposals for technological solutions in several areas. Because of this, these three models were tested as alternatives to the development of a computational tool used for the classification of ICHs. The predictions of the ANN and KNN models proved to be correct 89 and 85 % of the time, respectively, in the training phase. In the test phase, the predictions of the three models were correct 98% of the time. This type of tool would allow emergency services in small towns to perform quick diagnoses and transfer their patients to more complex hospitals when really indicated.