Year: 2017
Type: dissertation
Abstract: Different studies have demonstrated an augmented prevalence of different midline brain abnormalities in patients with\nboth mood and psychotic disorders. One of these abnormalities is the cavum septum pellucidum (CSP), which occurs\nwhen the septum pellucidum fails to fuse.\nThe study of these abnormalities in Magnetic Resonance Imaging requires a tedious and time-consuming process of\nmanually analyzing all the images in order to detect them. It is also problematic when the same abnormality is analyzed\nmanually by different experts because different criteria can be applied. In this context, in would be useful to develop\nan automatic method for localizing the abnormality and give the measure of its depth.\nIn this project, several methods have been developed by using machine learning and deep learning techniques. To test\nthe proposed approaches, data from of 861 subjects from FIDMAG, have been gathered and processed to use with our\nalgorithms. Among the subjects, 639 are patients with mood or psychotic disorders and 223 are healthy controls. This\nsame dataset was previously used in a study where the authors analyzed the prevalence of this abnormality across the\nsubjects manually, the depth of the abnormality previously reported is finally compared with the results obtained from\nour automatic methods.\nThe first proposed method implemented is an unsupervised algorithm that segments the abnormality over 2D slices\nwith an accuracy of the 89.72% and a sensitivity of an 84%.\nThen, a comparison among different Machine Learning classifiers have been conducted along with different\ndimensionality reduction methods. Between them, K Nearest Neighbors produced the best overall accuracy, with a\n99.1% of accuracy and a sensitivity of a 99% by using HOG descriptors to reduce the dimensionality of the data.\nThe last group of methods are different Deep Learning architectures. Three different implementations have been\ndesigned, and over the best performing architecture, two different input options to take into account contextual\ninformation have been proposed. The best Deep Learning method reached an accuracy of 98,8% and a specificity and\nsensitivity of a 99%.\nFinally, an analysis of the similarity with respect to the results with the manually handcrafted ground-truth have been\nconducted to validate the usability of the automatic method in further analysis of the abnormality.
Author: Aleix Solanes Font
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2596614006