Elastic SCAD SVM Cluster
For The Selection of Informative Functional Connectivity
in Autism Spectrum Disorder Classification

By: Yap Sin Yee (A16CS0226) - syyap4 [at] graduate.utm.my

Supervisor: Dr. Chan Weng Howe

Universiti Teknologi Malaysia, 2020

đź”— https://fyp.yapsinyee.com

PSM 1

(February - June 2019)

DOCUMENTATION


PSM 2

(February - August 2020)

DEVELOPMENT

Final

  • Changed the algorithm to ES-SVM Cluster.

  • Find the corresponding brain regions of the selected FC.

  • Plot connectomes on brain structure.

Performance Comparison
(Accuracy, Sensitivity, Specificity)

  • SVM

  • Elastic SCAD SVM

  • Elastic SCAD SVM Cluster

Version 1

  • Progress until first SVM & Elastic SCAD SVM classification, 70/30 train-test split

  • Produced confusion matrices

Version 2

  • Tried Leave-one-out cross validation

Version 3

  • Changed to 5-folds cross validation due to long computational time for LOO-CV

  • Added the part of count the frequency of selected features

  • Used the selected features to create a new model

Version 4

  • Added seed number to shuffle data rows

  • Repeat 5 sets of experiment, with 5-folds CV in each experiment

DOCUMENTATION

RESULTS HIGHLIGHT

  • The proposed Elastic SCAD SVM cluster has better performance in terms of accuracy, sensitivity and specificity compared to single SVM and single penalized SVM.
  • The proposed method has selected a set of significant FC features that had been proposed as regions of interest for the ASD. There are other recent papers which have similar findings towards the selected brain regions. The proposed method provide further support for a link between motor and social and communicative abilities in ASD.

Acknowledgement

Special thanks to:

  • Prof John Suckling from Brain Mapping Unit, University of Cambridge for providing research facilities, technical supports as well as assistance on neuroscience based validation of the results.

  • Dr. Hang See Pheng from Universiti Teknologi Malaysia for providing financial support & short course learning opportunity at "Big Data Analysis of Neuroimaging (fMRI-EEG) & Genetics"

External Links