Comparativa de Técnicas de Preprocesamiento y Entrenamiento para Detección de Hemorragias Intracraneales
Year: 2020
Type: dissertation
Abstract: In the last two decades, the medical world has been taking a remarkable interest in the use of \nimage processing with Artificial Intelligence for the detection of numerous diseases and injuries \nthat occur in humans. Specifically, the techniques of Machine Learning and Deep Learning \nhave shown their benefits to reduce human error and support more accurate diagnoses than \nthose obtained through traditional methods by trying to find hidden patterns within the images \nto be processed. In this Master's Thesis, different image pre-processing techniques and \ntraining algorithms are applied for the detection and identification of different types of \nFernando Arroyo Rubio Máster Universitario en Inteligencia Artificial \n5 \nSistema Artificial para la Detección de Hemorragias Intracraneales \nIntracreneal Hemorrhage (ICH) through computed tomography (CT). All this is implemented in \nan experimental setting with the aim of knowing which technique and algorithm offers the best \nresult. The results show that the use of convolutional neural networks, pre-processing \ntechniques such as Windowing and the use of data balancing methods allow the elaboration \nof quite precise models with results that surpass even current reference works. The best \nperforming model achieves values > 0.91% in F1 Score and >96% in average accuracy for all \nsubtypes. This model can classify ICH and its subtype with a high-level accuracy and can \nbecome a support system for physicians and radiologists in hospitals.
Author: Fernando Arroyo-Rubio
Language: Spanish
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W3018172720