Morphometric data fusion for early detection of alzheimer’s disease
Author
Type
Trabajo de grado - Maestría
Document language
EspañolPublication Date
2015-06Metadata
Show full item recordSummary
Abstract. We present a morphometry method which uses brain models generated using Nonnegative Matrix Factorization (NMF) characterized by signatures calculated from perceptual features such as intensities, edges and orientations, of some regions obtained by comparing the models. Two different measures are used to calculate volume-models distances in the regions of interest. The discerning power of these distances is tested by using them as features for a Support Vector Machine classifier. This work shows the usefulness of both measures as metrics in medical image applications when they are used in binary classification tasks. Our methodology was tested with two experimental groups extracted from a public brain MR dataset (OASIS), the classification between healthy subjects and patients with mild AD reveals an equal error rate (EER) measure which is better than previous approaches tested on the same dataset (0.1 in the former and 0.2 in the latter). When detecting very mild AD, our results (near to 75% of sensitivity and specificity) are comparable to the results with those approaches.Summary
Presentamos un m´etodo de morfometr´ı que usa modelos de cerebro que se generan usando factorizaci´on de matrices no-negativas (NMF por su nombre en ingl´es) y se caracterizan por firmas calculadas de rasgos perceptules como las intensidades, bordes y orientaciones de algunas regiones del cerebro obtenidas de la comparaci´on entre modelos. Dos medidas, la divergencia de Kullback-Leibler y la “Earth Mover’s Distance”, son usadas para calcular la distancia entre vol´umenes y modelos en las regiones de inter´es. Probamos el poder discriminante de estas distancias us´andolas para construir los vectores de caracter´ısticas para una m´aquina de soporte vectorial. Este trabajo muestra la utilidad de ambas medidas en tareas de clasificaci´on binaria. Nuestra metodolog´ıa fue probada con dos grupos experimentales extra´ıdos de la base de datos OASIS, la clasificaci´on entre sujetos sanos y pacientes con Alzheimer leve revela un EER que mejora los resultados obtenidos por trabajos publicados previamente con los mismos grupos experimentales. Cuando se trata de detectar Alzheimer muy leve, los resultados (cercanos a 75% de sensibilidad y especificidad) son comparables con los resultados obtenidos en dichas publicaciones.Keywords
Collections
This work is licensed under a Creative Commons Reconocimiento-NoComercial 4.0.This document has been deposited by the author (s) under the following certificate of deposit