Desarrollo de algoritmos para el análisis y procesamiento de imágenes de resonancia magnética para diferenciar los tejidos presentes en el hígado graso

Abstract

Advancement and development of different imaging has facilitated the diagnosis of diseases. MR imaging has gained prominence in the observation and diagnosis of multiple conditions that occur in organs such as the liver because of its high resolution. Due to the amount of image data increases and the analysis becomes more difficult, it is necessary to implement methods that improve diagnostic accuracy and decrease analysis time. In this project, we use an approach based on Machine Learning to differentiate liver fatty and normal zones in MR. This problem is solved within the category of supervised learning, using multi-layer perceptron-type neural networks with radiomic characteristics as input data. The regions of interest are segmented using 3D-Slicer, the characteristics are extracted using PyRadiomics and the neural networks are built from the Scikit-Learn library. Our best results guarantee shorter analysis with an accuracy of about 70%. However, precision is about 60% which causes that the number of false positives and negatives must be analyzed carefully with respect to the number of true positives and negatives obtained.