Topology based Kernels with Application to Inference Problems in Alzheimer’s disease

The subject IDs are provided below.

Topology-Based Features

We used used kernel density estimation to generate concentration map (PDF) from a scatter plot on a square grid of 50x50 pixels for each cortical surface. PDFs are the final descriptors that are used to construct Topology-based kernels.

Our study include 356 subjects (160 AD, 196 controls). PDF.mat contain descriptor that are used in experiments. PDF.mat is a cell array that contain vectorized descriptors for the two hemispheres, left and right respectively. MAT file also include true class labels, AD and controls respectively.

Software Details

FreeSurfer

We used FreeSurfer to extract cortical thickness from T1 weighted MR images. FreeSurfer v1.133.2.57 was installed for reconstruction. For installation details, please see FreeSurfer. AD brain scans had severe atrophy due to which “auto-Talairach” step failed. Therefore during automated cortical reconstruction process using recon-all directive, we disabled Talairach transformation via -notal-check for all subjects.

Following are the MATLAB routines to generate the minmax pairs (scatter plot) on the cortical surface of individual subject.

Heat kernel smoothing

For more details on Heat kernel smoothing, please visit Heat kernel smoothing

To obtain FreeSurfer spherical atlas read_surf.m

SPHARMsmooth2_DP.m

SPHARM2square.m

Scatter plot

We used following Plex routines to generate minmax pairs for AD classification. Please install the latest version. This is needed to run the complete implementation of Topology based Kernels.

Please report bugs and errors to pachauri@cs.wisc.edu