A Deep Learning Model for Stroke Patients’ Motor Function Prediction

Table 2

Parameter of EEG-DenseNet: T = temporal filter, DP = depth, P = point filter. K is the count of motor imagery units.

Structure	Layer	Filters	Size	Output	Activation
1	Input: input layer
	Reshape: first convolutional layer (CL)
	Second CL	T	(1, 64)		Linear activation function
	Normalization			T× M × S
	Depth CL		(C, 1)		Linear activation function
	Batch sizing
	Nonlinear activation layer				ReLu
	Max pooling		(1, 4)	(DP × T) ×1 × S/4
	Dropout layer (one out of four)		Probability or
2	Separable CL		(1, 16)		Linear activation function
	Batch sizing
	Nonlinear activation layer				ReLu
	Max pooling		(1, 6)
	Failure layer		or probability =0.6
	Flattening out
Classifier	Dense classified fully connected				Softmax