17 November 2017 Gaussian mixture models for detection of autism spectrum disorders (ASD) in magnetic resonance imaging
Author Affiliations +
Proceedings Volume 10572, 13th International Conference on Medical Information Processing and Analysis; 105720F (2017) https://doi.org/10.1117/12.2285902
Event: 13th International Symposium on Medical Information Processing and Analysis, 2017, San Andres Island, Colombia
Abstract
Autism Spectrum Disorder (ASD) is a complex neurological condition characterized by a triad of signs: stereotyped behaviors, verbal and non-verbal communication problems. The scientific community has been interested on quantifying anatomical brain alterations of this disorder. Several studies have focused on measuring brain cortical and sub-cortical volumes. This article presents a fully automatic method which finds out differences among patients diagnosed with autism and control patients. After the usual pre-processing, a template (MNI152) is registered to an evaluated brain which becomes then a set of regions. Each of these regions is the represented by the normalized histogram of intensities which is approximated by mixture of Gaussian (GMM). The gray and white matter are separated to calculate the mean and standard deviation of each Gaussian. These features are then used to train, region per region, a binary SVM classifier. The method was evaluated in an adult population aged from 18 to 35 years, from the public database Autism Brain Imaging Data Exchange (ABIDE). Highest discrimination values were found for the Right Middle Temporal Gyrus, with an Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) the curve of 0.72.
© (2017) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Javier Almeida, Nelson Velasco, Charlens Alvarez, Eduardo Romero, "Gaussian mixture models for detection of autism spectrum disorders (ASD) in magnetic resonance imaging", Proc. SPIE 10572, 13th International Conference on Medical Information Processing and Analysis, 105720F (17 November 2017); doi: 10.1117/12.2285902; https://doi.org/10.1117/12.2285902
PROCEEDINGS
9 PAGES


SHARE
Back to Top