Table 5.

Comparative evaluation of the deep learning model’s performance in ECG waveform delineation across four classes: P-wave, QRS complex, T-wave, and “N/W” (non-waveform regions) using QTDB. The table reports sensitivity (equivalent to recall), precision, and F1-score metrics for each class, indicating whether input signals are filtered. The best values are highlighted in bold. The * symbol denotes references utilizing datasets other than MIT-BIH arrhythmia, while the • symbol indicates metrics not available in the references.

Metric	Sensitivity				Precision				F1-score				Filtering
Class	N/W	P	QRS	T	N/W	P	QRS	T	N/W	P	QRS	T
Ref. [30]	0.95	0.90	0.95	0.92	0.95	0.92	0.94	0.90	0.95	0.91	0.94	0.91	No
Ref. [31]	0.93	0.92	0.95	0.93	0.95	0.90	0.90	0.92	0.95	0.91	0.94	0.93	No
Ref. [11] (%)	84.39	76.80	85.54	82.40	95.11	87.83	96.24	91.43	89.43	81.95	90.58	86.68	No
Ref. [11] (%)	98.75	96.53	99.70	96.81	98.36	89.74	99.19	95.44	98.55	93.01	99.45	96.12	Yes
Ref. [18] (%)	94.39	92.66	95.29	92.30	•	•	•	•	94.20	93.17	95.22	92.52	No
Ref. [19]* (%)	•	94.52	97.40	92.94	•	94.03	97.25	94.92	•	94.27	97.32	93.92	No
Our approach (%)	98.48	98.13	97.44	98.29	98.36	97.82	97.55	98.67	98.42	97.98	97.49	98.48	No