Relationship Induced Multi-atlas Learning for Alzheimer’s Disease Diagnosis

Multi-atlas based methods using magnetic resonance imaging (MRI) have been recently proposed for automatic diagnosis of Alzheimer’s disease (AD) and its prodromal stage, i.e., mild cognitive impairment (MCI). However, most existing multi-atlas based methods simply average or concatenate features gen...

Full description

Saved in:
Bibliographic Details
Published inMedical Computer Vision: Algorithms for Big Data pp. 24 - 33
Main Authors Liu, Mingxia, Zhang, Daoqiang, Adeli-Mosabbeb, Ehsan, Shen, Dinggang
Format Book Chapter
LanguageEnglish
Published Cham Springer International Publishing 2016
SeriesLecture Notes in Computer Science
Subjects
Online AccessGet full text
ISBN3319420151
9783319420158
ISSN0302-9743
1611-3349
DOI10.1007/978-3-319-42016-5_3

Cover

More Information
Summary:Multi-atlas based methods using magnetic resonance imaging (MRI) have been recently proposed for automatic diagnosis of Alzheimer’s disease (AD) and its prodromal stage, i.e., mild cognitive impairment (MCI). However, most existing multi-atlas based methods simply average or concatenate features generated from multiple atlases, which ignores the important underlying structure information of multi-atlas data. In this paper, we propose a novel relationship induced multi-atlas learning (RIML) method for AD/MCI classification. Specifically, we first register each brain image onto multiple selected atlases separately, through which multiple sets of feature representations can be extracted. To exploit the structure information of data, we develop a relationship induced sparse feature selection method, by employing two regularization terms to model the relationships among atlases and among subjects. Finally, we learn a classifier based on selected features in each atlas space, followed by an ensemble classification strategy to combine multiple classifiers for making a final decision. Experimental results on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database demonstrate that our method achieves significant performance improvement for AD/MCI classification, compared with several state-of-the-art methods.
ISBN:3319420151
9783319420158
ISSN:0302-9743
1611-3349
DOI:10.1007/978-3-319-42016-5_3