Table 8.

A summary of CNN applications in heart classification surveyed in Sect. 4

References	Preprocessing	Architecture	Dataset	Comparison against	Performance metric
[40]	ECG signals filtered using bandpass filter at 0.1-100 Hz and digitized at 360 Hz.	One-dimensional CNN with conv $\times 3$, FC $\times 2$	MIT/BIH arrhythmia database	Other existing methods	Accuracy, sensitivity, specificity
[20]	Image resized to $175\times 200\times 26$	Two path CNN of seven layers (conv, ReLU, maxpool) $\times 2$, (conv, ReLU) $\times 2$, (conv, ReLU, maxpool), (conv, dropout) $\times 2$	Tsinghua University Hospital, Beijing and Fuzhou University Hospital, China	SIFT, KAZE	Accuracy $92.1\%$