A Cascaded Convolutional Neural Network for Assessing Signal Quality of Dynamic ECG

Table 5

Comparisons with previous studies on public dataset.

Model	7-SQI SVM [3]	13-SQI SVM [10]					13-SQI SVM [10]^+	Baseline			Cascaded CNN
Level	Bad or good	Clean	Minor	Moderate	Severe	Extreme	—	Low	Mild	Severe	Low	Mild	Severe
Se (%)	86.3	56.9	59.3	56.2	46.7	45.3	—	86.3	82.2	87.2	92.9	89.1	85.6
Sp (%)	94.8	92.5	65.9	87.9	97.1	99.1	—	89.9	88.5	96.9	96.1	93.9	97.5
Ac (%)	94.6		57.3				88.07		85.7			91.8

Trained on simulated dataset and tested on public dataset; SVM with 13 signal quality assessment indices (SQIs) classified ECG signals into five levels (clean, minor, moderate, severe, and extreme noise). ⁺SVM with 13 SQIs was trained and tested on public dataset, for which only accuracy of five classifications was given.