Research Article
Computational and Comparative Study on Multiple Kernel Learning Approaches for the Classification Problem of Alzheimer’s Disease
@INPROCEEDINGS{10.1007/978-3-319-46909-6_30, author={Ahlam Mallak and Jeonghwan Gwak and Jong-In Song and Sang-Woong Lee}, title={Computational and Comparative Study on Multiple Kernel Learning Approaches for the Classification Problem of Alzheimer’s Disease}, proceedings={Nature of Computation and Communication. Second International Conference, ICTCC 2016, Rach Gia, Vietnam, March 17-18, 2016, Revised Selected Papers}, proceedings_a={ICTCC}, year={2017}, month={1}, keywords={Alzheimer’s disease Support vector machines Multiple kernel learning Generalized multiple kernel learning}, doi={10.1007/978-3-319-46909-6_30} }- Ahlam Mallak
Jeonghwan Gwak
Jong-In Song
Sang-Woong Lee
Year: 2017
Computational and Comparative Study on Multiple Kernel Learning Approaches for the Classification Problem of Alzheimer’s Disease
ICTCC
Springer
DOI: 10.1007/978-3-319-46909-6_30
Abstract
Several classification methods have been proposed for assisting computer-aided diagnosis of Alzheimer’s disease (AD). Among them, classification methods including (i) support vector machines (SVM), and (ii) generalized multiple kernel learning (GMKL) are getting increasing attention in recent studies. Nevertheless, there is little research on the comparison among these methods to find a better classification framework and further analysis of brain imaging features in the study of AD. To deal with this issue, we carry out exhaustive comparative study in this work to evaluate efficiency of these different classification methods. For the experiments, we used FreeSurfer mean cortical thickness dataset downloaded from the ADNI database (adni.loni.usc.edu) baseline data. The classification accuracy (in classifying the three classes CN, LMCI, AD) of comparative methods has been evaluated using 3-fold cross validation. From the comparative study, we could observe that GMKL is the most promising framework if the sufficient training data can be provided.
