Skip to main content

A Superpixel Pipeline for Neurodiagnosis

A Superpixel Pipeline for Neurodiagnosis

Teo Gelles, Andrew Gilchrist­Scott Ameet Soni and Sriraam Natarajan

We developed a pipeline to diagnose neurological disorders in humans using multiple supervised machine learning approaches. Neurological disorders severely affect quality of life, with Alzheimer's disease currently ranked the third most likely cause of death in the United States (Bahrampour, 2014). Current diagnosis techniques are subjective and most diseases can only be identified after the onset of symptoms; the only time the brain structure comes into play is at autopsy (Mayo Clinic Staff, 2013). While attempts have been made to diagnose based on magnetic resonance images (MRI), these depend on a technique called atlasing which warp images for the purpose of labeling, often obscuring large structural abnormalities (Magnano et al, 2014). We constructed an alternative pipeline where our images, taken from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database, were segmented into the three brain tissue types by a technique developed by Magnano et al. (2014) using a conditional random field (CRF). These images were then further segmented into superpixels, or conglomerations of voxels in similar areas and of similar colors (Achanta et al, 2010). Once these segments were created, we generated features based on their properties including size, tissue segmentation, location, etc. These features were then analyzed with multiple different techniques, including but not limited to support vector machines, spectral clustering, and adaboost (Natarajan et al, 2012). Research is still ongoing. Current results have three­way prediction accuracy ranging from 35 to 43 percent depending on diagnosis, which are too poor for diagnosis but does imply that our features have sufficient information on our diagnosis to beat chance.

Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi, Pascal Fua, and 
Sabine Süsstrunk, "SLIC Superpixels", EPFL Technical Report no. 149300, June 2010. 

Bahrampour, Tara. "New Study Ranks Alzheimer's as Third Leading Cause of Death, 
after Heart Disease and Cancer." Washington Post. The Washington Post, 3 Mar. 2014. 
Web. 13 Sept. 2014. 

Chris S. Magnano, Ameet Soni, Sriraam Natarajan, Gautam Kunapuli, "A Graphical 
Model Approach to ATLAS­free Mining of MRI Images", SDM 2014 

Mayo Clinic Staff. "Alzheimer's Disease." Diagnosing Alzheimer's: How Alzheimer's 
Is Diagnosed. Mayo Clinic, 01 Aug. 2013. Web. 13 Sept. 2014. 

Sriraam Natarajan, Tushar Khot, Kristian Kersting, Bernd Gutmann and Jude Shavlik. 
Gradient­based Boosting for Statistical Relational Learning: The Relational 
Dependency Network Case, Invited contribution to special issue of Machine 
Learning Journal (MLJ), Volume 86, Number 1, 25­56, 2012.