Label-Efficient Deep Learning for ECG Delineation: A Multi-Dataset Benchmark against Widely Used Delineation Tools

Jeonghwa Lim; Minje Park; Yeongyeon Na; Yujin Eom; Soyeon Lim; Young Ho Lee; Yu Jeong Kim; Sunghoon Joo

cs.LGOct 6, 2026

Label-Efficient Deep Learning for ECG Delineation: A Multi-Dataset Benchmark against Widely Used Delineation Tools

Authors: Jeonghwa Lim, Minje Park, Yeongyeon Na, Yujin Eom, Soyeon Lim, Young Ho Lee, Yu Jeong Kim, Sunghoon Joo, +1 more

Organizations: VUNO Inc., Seoul, South Korea · Department of AI Mobility Engineering, Ajou University, Suwon, South Korea · C&Thoth Co., Ltd., Seoul, South Korea · Department of Biomedical Sciences, Chonnam National University Graduate School, and the Department of Cardiovascular Medicine, Chonnam National University Hospital, Gwangju, South Korea · Department of Cardiovascular Medicine, Chonnam National University Hospital, Gwangju, South Korea · Department of Cardiovascular Medicine, Chonnam National University Hospital, and the Department of Internal Medicine, Chonnam National University Medical School, Gwangju, South Korea

Abstract

Electrocardiogram (ECG) delineation, the identification of waveform boundaries, is a foundational step that translates raw ECG signals into clinically interpretable measurements. Deep learning has advanced this task but remains dependent on costly expert annotations. Label-efficient strategies such as self-supervised pretraining and semi-supervised learning are expected to ease this burden, yet it remains unclear whether they yield reliable delineation and whether the deep models they produce outperform the delineation tools used in practice. We address this in two stages. First, comparing self-supervised objectives with supervised or semi-supervised fine-tuning across one internal and four external datasets, we find that pretraining helps but the objective matters, and that the value of semi-supervised fine-tuning depends on the pretraining objective. Second, we benchmark the selected deep learning model against widely used open-source (NeuroKit2, Prominence, ECGdeli) and commercial (CalECG) tools using three complementary metrics. The model ranks best on every metric and dataset, outperforming the strongest tool by a clear margin on the rhythm-diverse set (mIoU 71.3 vs. 54.8%; averaged point-wise sensitivity 92.6 vs. 76.4%), and degrades the least from sinus to arrhythmia. A rhythm-stratified and point-wise analysis further characterizes the distinctive behavior of each tool, yielding practical guidance for tool selection. These results provide systematic, multi-dataset evidence that self-supervised pretraining is effective for ECG delineation and enables a label-efficiently trained deep learning model to outperform widely used delineation tools by leveraging abundant unlabeled data. This supports adopting such models in diverse, real-world clinical settings.

Figures & tables

Fig. 1: Study overview. Stage 1 pretrains an encoder on unlabeled ECG data with self-supervised learning (e.g., Masked Autoencoder) and fine-tunes it with a segmentation decoder (e.g., FCN) under supervised and semi-supervised learning (e.g., Mean Teacher). Stage 2 benchmarks the resulting model against existing delineation tools on multiple datasets using three complementary metrics: mIoU, measurement error on the PR, QRS, and QT intervals, and fiducial point localization accuracy.

Role	Source	#Subjects	#ECGs	Duration (labeled)	Sample rate	Lead type	#Samples
Training	QTDB	105	105	5.9–253.6 s	250 Hz	2-lead	718
	ISP	499	499	10 s	1000 Hz	12-lead	5,988
	PTB-XL	18,885	21,837	— (unlabeled)	500 Hz	12-lead	262,044
Evaluation	LUDB	200	200	10 s	500 Hz	12-lead	2,369
	Zhejiang	334	334	1.3–7.1 s	2000 Hz	12-lead	4,008
	RDB	2,399	2,399	10 s	500 Hz	12-lead	28,788

TABLE I: Characteristics of the ECG datasets used for developing and evaluating the delineation model. Lead configurations range from 12-lead (six limb and six precordial leads) to 6-lead (limb only) and 2-lead (arbitrary pairs, e.g., MLII and V1).

Pretraining	Fine-tuning	Internal	LUDB	Zhejiang	RDB	Avg. (%) $\uparrow$	Rank $\downarrow$
		mIoU (%) $\uparrow$				Summary
–	Supervised	$75.2\pm 0.2$	$64.8\pm 0.8$	$53.3\pm 2.4$	$60.5\pm 0.5$	$63.5$	$3.50$
–	Semi-supervised	$81.6\pm 0.2$	$70.0\pm 5.9$	$66.0\pm 5.3$	$67.9\pm 0.9$	$71.4$	$3.50$
MAE ∗	Supervised	$\mathbf{82.8\pm 0.3}$	$71.3\pm 2.3$	$\mathbf{75.2\pm 1.5}$	$\mathbf{71.9\pm 1.2}$	$\mathbf{75.3}$	$\mathbf{1.25}$
MAE ∗	Semi-supervised	$82.3\pm 0.5$	$72.8\pm 5.3$	$74.8\pm 2.1$	$71.1\pm 0.4$	$75.2$	$\mathbf{1.25}$
MoCo ∗	Supervised	$82.2\pm 0.1$	$69.5\pm 3.3$	$66.2\pm 3.7$	$71.6\pm 1.5$	$72.4$	$1.75$

TABLE II: Benchmarking results of learning strategies for ECG delineation. Per-dataset entries are mIoU, reported as mean

\pm

standard deviation over three seeds, where Internal denotes the QTDB

+

ISP test split. Under Summary, Avg. is the mIoU averaged across the four datasets for each fine-tuning strategy, and Rank is the average rank among the four pretraining strategies, aggregated over fine-tuning and datasets. The best value in each column is shown in bold .

	mIoU (%) $\uparrow$				Averaged interval error (ms) $\downarrow$				Averaged point-wise sensitivity (%) $\uparrow$
Method	Internal	LUDB	Zhejiang	RDB	Internal	LUDB	Zhejiang	RDB	Internal	LUDB	Zhejiang	RDB	Avg. Rank $\downarrow$
NeuroKit2 (CWT)	41.3	44.9	7.3	40.0	70.9	108.3	72.8	30.2	59.4	59.9	15.7	54.9	5.8
NeuroKit2 (DWT)	43.4	45.6	8.6	38.6	43.6	45.6	52.7	40.5	60.3	62.2	17.8	58.5	5.2
ECGdeli	67.5	59.4	41.9	54.8	24.6	29.3	30.4	30.0	82.4	80.6	56.3	76.4	3.4
Prominence	60.4	63.9	42.7	51.7	28.1	22.3	35.9	23.0	78.1	82.9	54.6	73.7	3.6
DL delineator (sup)	83.0	71.6	73.5	71.1	11.5	19.0	16.3	22.2	96.1	86.6	82.9	92.6	1.8

TABLE III: Comparison of the DL delineator with open-source delineation tools. The best value in each column is in bold , and the second-best is underlined . Avg. Rank is the mean rank across all columns. CalECG , configured for interval outputs only, is compared separately on interval error in the Supplementary Material (Table S10 ).

Method	Average	PR	QRS	QT
NeuroKit2 (CWT)	97.3	207.4	32.1	52.3
NeuroKit2 (DWT)	51.3	45.3	17.0	91.5
ECGdeli	32.9	47.0	18.0	33.6
CalECG	37.9	41.1	32.7	40.0
Prominence	25.4	29.6	25.0	21.7
DL delineator (sup)	15.0	13.8	12.2	19.1

TABLE IV: Comparison of the DL delineator on mECGDB with open-source and commercial delineation tools, evaluated by mean absolute error (ms) for the PR interval, QRS duration, QT interval, and their average. The best value is in bold and the second-best is underlined .

Dataset	Method	P on	P off	QRS on	QRS off	T on	T off	Total
LUDB	NeuroKit2 (CWT)	43.7	43.1	80.8	72.2	56.4	63.4	59.9
	NeuroKit2 (DWT)	73.3	63.8	66.2	81.8	32.5	55.8	62.2
	ECGdeli	71.8	87.2	97.3	85.1	70.7	71.7	80.6
	CalECG	74.0	-	67.7	69.1	-	61.5	68.1
	Prominence	79.2	80.8	91.9	85.8	79.1	80.7	82.9
	DL delineator (sup)	80.4	81.3	99.1	98.9	77.8	81.6	86.6

TABLE V: Comparison of the DL delineator on external test sets ( LUDB and Zhejiang ) with open-source and commercial delineation tools, evaluated by point-wise sensitivity (%) per fiducial point. ”Total” is the average over the available onset/offset points. The best value is in bold and the second-best is underlined .

Metric	Method	Sinus	Arrhythmia	Degradation
		Subgroup performance
mIoU (%) $\uparrow$	ECGdeli	66.3	38.7	41.6%
	Prominence	64.4	33.2	48.5%
	DL delineator (sup)	76.1	64.9	14.7%
	DL delineator (semi)	76.6	64.2	16.2%
Averaged interval error (ms) $\downarrow$	ECGdeli	28.9	40.1	38.8%

TABLE VI: Rhythm-stratified subgroup performance on RDB for the sinus and arrhythmia groups, with the relative performance degradation from sinus to arrhythmia (smaller is better). The best value is in bold and the second-best is underlined .

Fig. 2: Qualitative examples of ECG delineation results across methods and datasets. The rhythm classes shown reflect the characteristics of each dataset. Rows are grouped into the reference annotation (GT), rule-based tools, and the DL delineator ; shaded bands mark each method’s detected P (blue), QRS (orange), and T (green) waves on the same input signal. SR, sinus rhythm; PVC, premature ventricular contraction; AFIB, atrial fibrillation; SVT, supraventricular tachycardia.

Source	Train	Validation	Test	Total
QTDB	422	148	148	718
ISP	3,792	1,272	924	5,988
Internal ( QTDB $+$ ISP )	4,214	1,420	1,072	6,706

TABLE S1: The detail number of the Internal ( QTDB

+

ISP ) set sample.

Dataset	Rhythm	#ECGs
LUDB	Sinus rhythm	143
	Sinus tachycardia	4
	Sinus bradycardia	25
	Sinus arrhythmia	8
	Irregular sinus rhythm	2
	Abnormal rhythm (AFib, flutter)	18

TABLE S2: Rhythm composition of the LUDB and Zhejiang used for external test.

Group	Rhythm	#ECGs
Sinus	Sinus rhythm	400
	Sinus bradycardia	400
	Sinus tachycardia	140
	Sinus irregularity	399
Arrhythmia	Atrial flutter	400
	Atrial fibrillation	400

TABLE S3: Rhythm composition of the RDB used for the rhythm-stratified subgroup analysis. Recordings are grouped into sinus and arrhythmia categories.

Component	Setting
Framework	Mean Teacher (MT)
Teacher EMA decay	0.99
Consistency loss weight	1.0
Weak augmentation	Random resized cropping
Strong augmentation	RandAugment( $N=3,p=0.5$ )
Candidate transforms	Powerline noise

TABLE S4: Semi-supervised learning configuration.

Hyperparameter	MAE	MoCo	MERL
Epochs	800	300	50
Warm-up epochs	40	40	5
Batch size	2048	2048	2048
Learning rate	1.2e-3	1.2e-3	2.0e-4
Weight decay	0.05	0.1	1e-5
$(\beta_{1},\beta_{2})$	(0.9, 0.95)	(0.9, 0.999)	(0.9, 0.999)

TABLE S5: Optimization hyperparameters for self-supervised pretraining: MAE , MoCo , and MERL . All algorithms are trained with the AdamW optimizer and cosine learning rate scheduling.

Hyperparameter	Value
MAE
Masking ratio	0.75
Decoder embedding dimension	96
Decoder depth	4
Decoder attention heads	3
MoCo

TABLE S6: Model architecture and algorithm-specific hyperparameters for self-supervised pretraining.

Method	Time (ms) $\downarrow$
NeuroKit2 (CWT)	56.3 $\pm$ 11.9
NeuroKit2 (DWT)	63.6 $\pm$ 11.1
ECGdeli	141.0 $\pm$ 57.5
Prominence	20.0 $\pm$ 6.0
DL delineator	17.1 $\pm$ 1.9

TABLE S7: Computation time analysis on LUDB , reported as a mean

\pm

standard deviation. The best value is bolded.

Pretraining	Fine-tuning	Internal	LUDB	Zhejiang	mECGDB	RDB
–	Supervised	$16.2\pm 0.3$	$21.1\pm 0.7$	$33.1\pm 2.0$	$26.1\pm 0.9$	$23.4\pm 0.3$
–	Semi-supervised	$12.8\pm 0.4$	$20.8\pm 2.4$	$21.7\pm 3.0$	$16.2\pm 0.6$	$29.8\pm 4.3$
MAE	Supervised	$\mathbf{11.6\pm 0.2}$	$19.2\pm 1.7$	$16.5\pm 1.2$	$\mathbf{15.0\pm 0.2}$	$\mathbf{20.9\pm 1.2}$
MAE	Semi-supervised	$11.7\pm 0.3$	$18.4\pm 2.0$	$\mathbf{15.9\pm 0.6}$	$15.7\pm 0.7$	$22.7\pm 1.0$
MoCo	Supervised	$12.0\pm 0.3$	$21.3\pm 2.5$	$21.3\pm 1.0$	$15.2\pm 0.8$	$22.4\pm 1.2$
MoCo	Semi-supervised	$11.8\pm 0.3$	$\mathbf{17.2\pm 0.8}$	$16.9\pm 0.8$	$15.4\pm 0.7$	$22.3\pm 0.5$

TABLE S8: Benchmarking results of learning strategies for ECG delineation. Per-dataset entries are averaged interval error (ms), reported as mean

\pm

std over three seeds, where Internal denotes the QTDB

+

ISP test split. The best value in each column is shown in bold .

Pretraining	Fine-tuning	Internal	LUDB	Zhejiang	RDB
–	Supervised	$91.3\pm 0.2$	$82.5\pm 1.2$	$66.5\pm 2.6$	$87.1\pm 0.7$
–	Semi-supervised	$95.4\pm 0.2$	$86.5\pm 7.0$	$76.6\pm 6.2$	$90.8\pm 0.6$
MAE	Supervised	$96.1\pm 0.2$	$86.4\pm 3.0$	$85.4\pm 2.2$	$\mathbf{93.3\pm 0.6}$
MAE	Semi-supervised	$\mathbf{96.2\pm 0.3}$	$89.2\pm 3.3$	$\mathbf{85.9\pm 1.5}$	$92.6\pm 0.3$
MoCo	Supervised	$95.7\pm 0.1$	$85.1\pm 3.5$	$77.5\pm 4.2$	$92.4\pm 1.0$
MoCo	Semi-supervised	$96.1\pm 0.2$	$\mathbf{90.3\pm 3.5}$	$84.5\pm 0.7$	$92.9\pm 1.0$

TABLE S9: Benchmarking results of learning strategies for ECG delineation. Per-dataset entries are averaged point-wise sensitivity (%), reported as mean

\pm

std over three seeds, where Internal denotes the QTDB

+

ISP test split. The best value in each column is shown in bold .

	Averaged interval error (ms) $\downarrow$
Method	Internal	LUDB	Zhejiang
CalECG	37.9	20.1	58.2
DL delineator (sup)	11.5	19.0	16.3
DL delineator (semi)	12.0	16.8	15.3

TABLE S10: Comparison of the DL delineator with CalECG on averaged interval error. The best value is in bold and the second best is underlined .

		Sensitivity (%)		PPV (%)
Dataset	Method	40 ms	150 ms	40 ms	150 ms
Internal	NeuroKit2 (CWT)	59.4	73.7	66.1	82.1
	NeuroKit2 (DWT)	60.3	88.4	60.5	88.0
	ECGdeli	82.4	97.0	77.3	90.8
	Prominence	78.1	95.9	74.2	91.1
	DL delineator (sup)	96.1	99.0	95.6	98.6

TABLE S11: Comparison of the DL delineator with open-source delineation tools, evaluated by averaged point-wise total sensitivity and positive predictive value (PPV) (%) at tolerances of 40 and 150 ms. The best value in each column is in bold .

Dataset	Method	P on	P off	QRS on	QRS off	T on	T off	Total
LUDB	NeuroKit2 (CWT)	43.5	42.8	87.4	93.8	67.5	76.0	68.5
	NeuroKit2 (DWT)	62.4	53.6	75.1	92.3	32.7	56.5	62.1
	ECGdeli	55.1	66.9	96.8	84.6	63.2	64.1	71.8
	CalECG	83.8	-	95.0	96.9	-	84.5	90.1
	Prominence	64.6	65.9	91.0	84.9	79.0	80.7	77.7
	DL delineator (sup)	88.1	89.1	99.2	99.0	81.4	85.4	90.4

TABLE S12: Comparison of the DL delineator on external test sets ( LUDB and Zhejiang ) with open-source and commercial delineation tools, evaluated by point-wise positive predictive value (PPV, %) per fiducial point. ”Total” is the average over the available onset/offset points. The best value is in bold .

Dataset	Method	P on	P off	QRS on	QRS off	T on	T off
LUDB	NeuroKit2 (CWT)	-7.6 $\pm$ 9.4	0.1 $\pm$ 9.5	-8.8 $\pm$ 12.3	-12.0 $\pm$ 13.3	5.0 $\pm$ 17.4	6.1 $\pm$ 10.9
	NeuroKit2 (DWT)	3.5 $\pm$ 17.6	-3.0 $\pm$ 17.3	-3.6 $\pm$ 13.7	-3.1 $\pm$ 13.4	17.7 $\pm$ 20.1	-10.0 $\pm$ 13.2
	ECGdeli	-13.0 $\pm$ 11.3	16.8 $\pm$ 11.4	-7.2 $\pm$ 13.9	16.2 $\pm$ 16.0	-10.4 $\pm$ 19.9	4.5 $\pm$ 15.9
	CalECG	0.3 $\pm$ 11.5	-	8.1 $\pm$ 13.5	-2.3 $\pm$ 11.8	-	5.3 $\pm$ 12.1
	Prominence	1.9 $\pm$ 12.8	3.1 $\pm$ 13.9	-0.0 $\pm$ 15.9	-11.4 $\pm$ 18.1	2.8 $\pm$ 18.2	13.2 $\pm$ 14.8
	DL delineator (sup)	-5.4 $\pm$ 14.3	7.6 $\pm$ 15.1	-8.4 $\pm$ 10.1	10.4 $\pm$ 11.4	4.0 $\pm$ 20.1	12.5 $\pm$ 14.1

TABLE S13: Comparison of the DL delineator on LUDB and Zhejiang external test sets with widely used open-source and commercial delineation tools, evaluated by point-wise localization error (ms) per fiducial point at the 40 ms tolerance. The signed mean error and its standard deviation are reported.

		Sinus group					Arrhythmia group
Metric	Method	SR	SB	ST	SI	Avg.	AF	AFIB	AT	SVT	Avg.	Deg.%
mIoU (%) $\uparrow$	ECGdeli	67.8	67.2	63.2	66.9	66.3	38.8	42.1	45.3	28.8	38.7	41.5
	Prominence	67.2	65.4	58.9	66.3	64.4	33.1	38.1	38.6	22.8	33.2	48.6
	DL delineator (sup)	76.4	77.0	75.0	75.9	76.1	61.1	67.6	63.5	67.4	64.9	14.7
	DL delineator (semi)	77.3	78.0	74.2	76.9	76.6	59.4	67.8	63.4	66.3	64.2	16.2
Averaged interval error (ms) $\downarrow$	ECGdeli	28.3	30.1	28.4	28.6	28.9	34.1	29.8	34.4	62.0	40.1	65.8

TABLE S14: Rhythm-stratified subgroup analysis on the RDB , comparing the DL delineator with the two strongest tools (Prominence, ECGdeli). Rhythms are grouped into sinus (SR, SB, ST, SI) and arrhythmia (AF, AFIB, AT, SVT). For each metric, we report the group mean performance and its relative degradation (Deg.%) from sinus to arrhythmia. The best value is in bold .

Fig. S1: Qualitative P/QRS/T delineation on a representative 2.5 s Lead II segment from LUDB . Rows are grouped into the reference annotation (GT), rule-based tools, and the DL delineator ; shaded bands mark each method’s detected P (blue), QRS (orange), and T (green) waves on the same input signal.

Fig. S2: Qualitative P/QRS/T delineation on a representative 2.5 s Lead II segment from Zhejiang . Rows are grouped into the reference annotation (GT), rule-based tools, and the DL delineator ; shaded bands mark each method’s detected P (blue), QRS (orange), and T (green) waves on the same input signal.

Fig. S3: Qualitative P/QRS/T delineation on a representative 2.5 s Lead II segment from RDB . Rows are grouped into the reference annotation (GT), rule-based tools, and the DL delineator ; shaded bands mark each method’s detected P (blue), QRS (orange), and T (green) waves on the same input signal. SR, sinus rhythm; AFIB, atrial fibrillation; SVT, supraventricular tachycardia.

Explore similar work

Decoder Design Matters for ECG Delineation
Sep 15, 2026Joseph Scharpf, William Han, Chaojing Duan +3Electrocardiogram Classification U-Net
Carnegie Mellon University · Allegheny Health Network · University of Colorado
HeartBeatAI: An Interpretable and Robust Deep Learning Framework for Multi-Label ECG Arrhythmia Detection
May 23, 2026Shubham Gupta, Nikhil Panwar, Partha Pratim RoyElectrocardiogram Classification Arrhythmia Classification
Department of Computer Science and Engineering, Indian Institute of Technology (ISM) Dhanbad, Dhanbad, Jharkhand, India. · Department of Computer Science and Engineering, Indian Institute of Technology Roorkee, Roorkee, Uttarakhand, India.
ECG-NAT: A Self-supervised Neighborhood Attention Transformer for Multi-lead Electrocardiogram Classification
May 13, 2026Mahsa Gazeran, Sayvan Soleymanbaigi, Fatemeh Daneshfar +2Electrocardiogram Classification Self-Supervised Learning
Department of Computer Engineering, University of Kurdistan, Sanandaj, Iran · Department of Mathematics and Operational Research, University of Mons, Mons, Belgium

Sep 15, 2026, cs.LGJ/K move · Enter open · S save

Decoder Design Matters for ECG Delineation

Joseph Scharpf, William Han, Chaojing Duan, Michael A. Rosenberg+2

Carnegie Mellon University · Allegheny Health Network · University of Colorado

Electrocardiogram Classification U-Net Delineation Semi-Supervised Learning Annotations

Open paper AskPDF arXiv

Electrocardiogram (ECG) delineation identifies the boundaries of P waves, QRS complexes, and T waves, providing structural annotations that can guide AI models in learning to interpret ECGs. However, training accurate delineation models requires manual annotations that are scarce and time-consuming to obtain. Recent work addresses this limitation through semi-supervised learning (SSL), but the design of the architecture, particularly the decoder, has received less attention. To this end, we propose R-U-Net, an ECG delineation model that pairs a ResNet-18 encoder with a U-Net decoder. On SemiSegECG, R-U-Net outperforms the strongest evaluated ResNet-18 + fully convolutional network (FCN) head baseline in each of the 16 in-domain settings by 3.3-13.0 mIoU and achieves 82.6 mIoU in the cross-domain setting, an improvement of 8.1 mIoU. Controlled ablations show that decoder design contributes more to performance gains than the evaluated SSL methods, motivating further exploration of architectures for ECG delineation. All code is open-source at github.com/ELM-Research/ECG-Delineation.

Abstract

Explore similar work

Decoder Design Matters for ECG Delineation

HeartBeatAI: An Interpretable and Robust Deep Learning Framework for Multi-Label ECG Arrhythmia Detection

ECG-NAT: A Self-supervised Neighborhood Attention Transformer for Multi-lead Electrocardiogram Classification