White Matter Connectome Correlates of Auditory Over-Responsivity: Edge Density Imaging and Machine-Learning Classifiers

Frontiers in Integrative Neuroscience

Annie Brandes-Aitken¹  Pratik Mukherjee²  Molly Gerdes³  Elysa J. Marco⁴  Teresa Tavassoli⁵  Eva M. Palacios⁶  Seyedmehdi Payabvash⁶  Julia P. Owen⁸  Maxwell B. Wang⁹ 

[1] Department of Applied Psychology, New York University, New York, NY, United States;Department of Bioengineering and Therapeutic Sciences, University of California, San Francisco, San Francisco, CA, United States;Department of Neurology, University of California, San Francisco, San Francisco, CA, United States;Department of Pediatric Neurology, Cortica Healthcare, San Rafael, CA, United States;Department of Psychology and Clinical Sciences, University of Reading, Reading, United Kingdom;Department of Radiology and Biomedical Imaging, University of California, San Francisco, San Francisco, CA, United States;Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, CT, United States;Department of Radiology, University of Washington, Seattle, WA, United States;University of Pittsburg School of Medicine, Pittsburgh, PA, United States;

关键词: machine-learning;    edge density imaging;    diffusion tensor imaging;    sensory over-responsivity;    auditory over-responsivity;    neurodevelopmental disorders;   

DOI  :  10.3389/fnint.2019.00010

来源: DOAJ

【 摘 要 】

Sensory over-responsivity (SOR) commonly involves auditory and/or tactile domains, and can affect children with or without additional neurodevelopmental challenges. In this study, we examined white matter microstructural and connectome correlates of auditory over-responsivity (AOR), analyzing prospectively collected data from 39 boys, aged 8–12 years. In addition to conventional diffusion tensor imaging (DTI) maps – including fractional anisotropy (FA), mean diffusivity (MD), radial diffusivity (RD), and axial diffusivity (AD); we used DTI and high-resolution T1 scans to develop connectome Edge Density (ED) maps. The tract-based spatial statistics was used for voxel-wise comparison of diffusion and ED maps. Then, stepwise penalized logistic regression was applied to identify independent variable (s) predicting AOR, as potential imaging biomarker (s) for AOR. Finally, we compared different combinations of machine learning algorithms (i.e., naïve Bayes, random forest, and support vector machine (SVM) and tract-based DTI/connectome metrics for classification of children with AOR. In direct sensory phenotype assessment, 15 (out of 39) boys exhibited AOR (with or without neurodevelopmental concerns). Voxel-wise analysis demonstrates extensive impairment of white matter microstructural integrity in children with AOR on DTI maps – evidenced by lower FA and higher MD and RD; moreover, there was lower connectome ED in anterior-superior corona radiata, genu and body of corpus callosum. In stepwise logistic regression, the average FA of left superior longitudinal fasciculus (SLF) was the single independent variable distinguishing children with AOR (p = 0.007). Subsequently, the left SLF average FA yielded an area under the curve of 0.756 in receiver operating characteristic analysis for prediction of AOR (p = 0.008) as a region-of-interest (ROI)-based imaging biomarker. In comparative study of different combinations of machine-learning models and DTI/ED metrics, random forest algorithms using ED had higher accuracy for AOR classification. Our results demonstrate extensive white matter microstructural impairment in children with AOR, with specifically lower connectomic ED in anterior-superior tracts and associated commissural pathways. Also, average FA of left SLF can be applied as ROI-based imaging biomarker for prediction of SOR. Finally, machine-learning models can provide accurate and objective image-based classifiers for identification of children with AOR based on white matter tracts connectome ED.

【 授权许可】

Unknown