Decoding Resting-state EEG to Predict Visual Field Defect with Convolutional Neural Network in Stroke
- Resource Type
- Conference
- Authors
- Xu, Jiahua.; Wu, Zheng.; Nurnberger, Andreas; Sabel, Bernhard A.
- Source
- 2021 10th International IEEE/EMBS Conference on Neural Engineering (NER) Neural Engineering (NER), 2021 10th International IEEE/EMBS Conference on. :807-810 May, 2021
- Subject
- Bioengineering
Signal Processing and Analysis
Visualization
Three-dimensional displays
Neural engineering
Stroke (medical condition)
Predictive models
Brain modeling
Electroencephalography
- Language
- ISSN
- 1948-3554
Stroke is one of the leading factors of human being's death and disability. One-third of stroke patients may suffer partial visual field loss in both eyes. The relationship between brain oscillation after a unilateral occipital stroke and visual field defect is worth investigating. Decoding resting-state Electroencephalography (EEG) to predict patients' visual field distribution could be an essential reference for a better understanding of the compensation of visual functions after a stroke. The result could be beneficial for clinical diagnostics and treatment. This paper proposed a frequency spectrum-based 2D convolutional neural network(CNN) and brain connectivity-based 3DCNN model to predict the visual field defect. The results show that the frequency spectrum-based 2DCNN achieved a higher accuracy on visual field location than the connectivity-based 3DCNN model. Simultaneously, the percentage seems to be not predictable for both domains, and therefore we also explored the patterns of electrophysiological data for feature visualization and interpretation.