Meeting Banner
Abstract #3194

Information Extraction from Raw DTI Data Using Texture Based Analysis: A Preliminary Study of Classification and Regression

Che-Wei Chang1, 2, Chien-Chang Ho1, Jyh-Horng Chen1, 2

1Electrical Engineering, National Taiwan University, Taipei, Taiwan; 2Interdisciplinary MRI/MRS Lab, National Taiwan University, Taipei, Taiwan

This study presents a texture based analysis, Local Binary Pattern on Three Orthogonal Planes (LBP-TOP), to extract effective features from raw DTI data. Examples of sex classification and age estimation were used to demonstrate the performance of this method. A total 204 subject downloaded from NKI/Rockland Samples were used to evaluate those approaches. Our results show that the best sex classification accuracy is 0.81, and the best age estimation mean average error is 6.32 years. We demonstrated that LBP-TOP is capable of extracting effective information from DTI and could be a good candidate for classifying or evaluating neurological diseases based on raw DTI data.

Keywords

access accuracy approaches atlas atlases automatic best better binary brain brains build candidate capability capable cerebellar classification classifying clinical code combining complex complexity compute computed computing conducted context cortical cross defining descriptor diffusion diffusivity disease diseases distant distribution downloaded effective efficient eigenvectors eight elder electrical encodes engineering error estimation evaluate evaluating examples extract extracting extraction feature features fold fractional good histogram important institute intelligence interdisciplinary introduce introduced isotropic journal library linear local machine males mapping masks materials model models modifying national needed neighborhood neighborhoods neighbors neurological operator orientation orthogonal overlapped oxford pattern percentages performance pixel pixels preliminary preprocessing probabilistic projects public radius ranking registration regression resolution rigid sample samples select selection separate separated simple simplified skull software source space spacial special steps stripped structural structure studies subject subjects table template tensor texture threshold thresholding tool toolbox train transactions transform transformed unbalanced unique unmodified usage useful validation various white window years zero