A New Multichannel Parallel Network Framework for the Special Structure of Multilead ECG

Table 2

Detailed parameters used for all the layers of proposed model.

Channel	Layer	Layer name	Kernel × unit	Other layer parameters
MLCNN	1	Conv2D	5 × 32	Strides = 1, padding = 2
	2	Batch norm.	—	—
	3	ReLU	—	—
	4	Max pooling 2D	2	Strides = 2
	5	Conv2D	5 × 64	Strides = 1, padding = 2
	6	Batch norm.	—	—
	7	ReLU	—	—
	8	Max pooling 2D	2	Strides = 2
	9	Conv2D	5 × 128	Strides = 1, padding = 2
	10	Batch norm.	—	—
	11	ReLU	—	—
	12	Max pooling 2D	2	Strides = 2
	13	Linear	256	ReLU, dropout
BiLSTM	1	BiLSTM	—	Hidden size = 128
—	1	Channel fusion	—	α
	2	Linear	512	ReLU, dropout
	3	Linear	128	ReLU, dropout
	4	Linear	64	ReLU, dropout