22nd EANN 2021, 25 - 27 June 2021, Greece

Deep learning topology-preserving EEG-based images for autism detection in infants

Cosmin Stamate, George Magoulas, Michael Thomas


  Developing digital biomarkers that would enable reliable detection of autism–ASD early in life is challenging because of the variability in the presentation of the autistic disorder and the need for simple measurements that could be implemented routinely during checkups. Electroencephalography, widely known as EEG, is an electrophysiological monitoring method that has been explored as a potential clinical tool for monitoring atypical brain function. EEG measurements were collected from 101 infants, beginning at 12 to 15 months of age and continuing until 36 months of age. In contrast to previous work in the literature that analysed EEG signals, our approach considers EEG–as– an–image using an appropriate signal transformation that preserves the spatial location of the EEG signals to create RGB images. It employs Residual neural networks and transfer learning to detect atypical brain function. Prediction of the clinical diagnostic outcome of ASD or not ASD at 36 months was accurate from as early as 12 months of age. This shows that using end-to-end deep learning is a viable way of extracting useful digital biomarkers from EEG measurements for predicting autism in infants.  

*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.