Antonis D. Savva was born in Larnaca, Cyprus in 1989. He attended the School of Electrical and Computer Engineering of the National Technical University of Athens from 2009 to 2014. He graduated in 2014 with a specialization in Electronics and Systems. From 2014 to present he continues his studies as a PhD candidate at the School of Electrical and Computer Engineering of the National Technical University of Athens. His PhD thesis concerns the processing of resting-state fMRI data, using advanced signal processing techniques. Currently he focuses on the quantification of dynamic properties of fMRI recordings of the human brain, by developing and applying novel methodologies of dynamic functional connectivity. More specifically, his current research aims at extracting discriminant features, by using dynamic functional connectivity, from autistic subjects towards the development of a medical decision support system which will be able to classify people with autism spectrum disorders and controls. His research interests also include digital processing of 2D/3D/4D imaging data (CT, MRI, fMRI), GPU-based algorithms optimization and bio-informatics. Moreover he is an associate of the Medical Image & Signal Processing Lab, Department of Biomedical Engineering – Technological Educational Institute of Athens. He has programming experience in Matlab as well as in C/C++.
Deriving Resting State fMRI Biomarkers for Classification of Autism Spectrum Disorder
Antonis D. Savva, Aikaterini S. Karampasi, George K. Matsopoulos
School of Electrical and Computer Engineering, National Technical University of Athens
Correspondence concerning this article should be addressed to Antonis D. Savva, 9, Iroon Polytechniou str, Zografos, 15780, Athens, Greece, Contact: email@example.com
Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by social deficits concerning communication and social interaction, repetitive patterns of behavior and limited interests. The purpose of the current study is to classify autistic people using resting state fMRI data. Towards this direction, data were retrieved from the Autism Brain Imaging Data Exchange (ABIDE) initiative, based on the Harvard-Oxford parcellation scheme. The extracted timeseries were used in order to calculate plethora of functional features which were set as input to a medical Decision Support System (DSS).
This study focused on Default Mode Network, which has been previously reported to be related with ASD and is utilized to construct an extensive feature vector which quantifies within-network functional connections. These interactions were evaluated based on static and dynamic functional connectivity analysis, information based metrics for instance entropy, skewness, kurtosis and Kullback-Leibler divergence. In addition, Haralick texture features are introduced for calculating texture measures of the network’s responses. Finally, after comprehensive trials, head motion parameters, age, sex and information regarding the acquisition protocol, were found to improve the overall performance of the DSS. Internal parameters of the DSS were chosen based on a Bayesian optimization framework, which aimed to maximize the Area Under Curve (AUC). The DSS comprised of a common Support Vector Machine classifier featuring several kernels such as linear, polynomial and radial basis function (RBF) kernel.
In this study several combinations were made in order to construct a suitable feature vector, in terms of higher classification performance. The best results were obtained using the combination of static and dynamic functional connectivity, head motion parameters, handedness, sex and information concerning the hardware setup. Using the RBF kernel the above combination resulted in accuracy 66.47%, sensitivity 73.63% and specificity 58.55%.
Through the current study it is shown that it is feasible to achieve high classification performance despite the plenty acquisition parameters and different demographics or other information present in the data, by setting them as features to the DSS. Moreover, it is concluded that the quantification of functional interactions could be utilized for classifying ASD.
Abraham, A., Milham, M. P., Di Martino, A., Craddock, R. C., Samaras, D., Thirion, B., & Varoquaux, G. (2017). Deriving reproducible biomarkers from multi-site resting-state data: An autism-based example. NeuroImage, 147, 736-745. doi:10.1016/j.neuroimage.2016.10.045
Funakoshi, Y., Harada, M., Otsuka, H., Mori, K., Ito, H., & Iwanaga, T. (2016). Default mode network abnormalities in children with autism spectrum disorder detected by resting-state functional magnetic resonance imaging. Journal of Medical Investigation, 63(3-4), 204-208. doi:10.2152/jmi.63.204
Itahashi, T., Yamada, T., Watanabe, H., Nakamura, M., Ohta, H., Kanai, C, Iwanami, A., Kato, N., Hashimoto, R.-I. (2015). Alterations of local spontaneous brain activity and connectivity in adults with high-functioning autism spectrum disorder. Molecular Autism, 6(1) doi:10.1186/s13229-015-0026-z
Hernandez, L. M., Rudie, J. D., Green, S. A., Bookheimer, S., & Dapretto, M. (2015). Neural signatures of autism spectrum disorders: Insights into brain network dynamics. Neuropsychopharmacology, 40(1), 171-189. doi:10.1038/npp.2014.172
Aarthi Padmanabhan, Charles J. Lynch, Marie Schaer, Vinod Menon, The Default Mode Network in Autism, Biological Psychiatry: Cognitive Neuroscience and Neuroimaging, Available online 22 April 2017, ISSN 2451-9022, https://doi.org/10.1016/j.bpsc.2017.04.004.
Jamal, W., Das, S., Oprescu, I. -.A, Maharatna, K., Apicella, F., & Sicca, F. (2014). Classification of autism spectrum disorder using supervised learning of brain connectivity measures extracted from synchrostates. Journal of Neural Engineering, 11(4) doi:10.1088/1741-2560/11/4/046019