Evaluation Metrics

AUROC ranges from 0 to 1. The higher the AUROC, the better the overall performance of the algorithm.

Single-lead

Sample size: approximately 47,000 cases, with an average length of 30 seconds per single-lead ECG.

Data source: data collected from HeartVoice equipment.

Annotation method: cardiologists/physicians from tertiary hospitals annotated the data.

Test results (AUROC)

Diagnosis conclusion	Label	AUROC
Sinus rhythm	SN	0.826
Sinus rhythm irregular	SNA	0.94
Sinus tachycardia	SNT	0.988
Sinus bradycardia	SNB	0.993
Premature ventricular contraction	PVC	0.975
Premature atrial contraction	PAC	0.967
Ventricular tachycardia	VT	0.748
Supraventricular tachycardia	SVT	0.968
Atrial fibrillation	AF	0.988
Atrial flutter	AFL	0.93
Atrioventricular re-entrant syndrome / pre-excitation syndrome	WPW	0.812
Junctional escape beat	AE	0.921
Supraventricular escape beat	JE	0.903
First-degree atrioventricular block	AVBI	0.967
Second-degree atrioventricular block	AVBII	0.876
Third-degree atrioventricular block	AVBIII	0.997
Left bundle branch block	LBBB	0.845
Right bundle branch block	RBBB	0.945

12-lead

Sample size: approximately 60,000 cases, with an average length of 30 seconds per 12-lead ECG.

Data source: 5 datasets.

Annotation method: unknown.

Test results (AUROC)

Multi-dataset comparison results

Class	PTB-XL	ningbo	chapman_shaoxing	georgia	CINC2021
SN	0.7884	0.913	0.9479	0.9041	0.7951
N	0.4798	0.679	0.7163	0.5912	0.6445
SNA	0.0757	0.1548	0.1394	0.0992	0.1423
SNT	0.5015	0.4005	0.6587	0.3873	0.5168
SNB	0.5141	0.8365	0.3818	0.5216	0.4443
PVC	0.9471	0.8265	0.5289	0.8583	0.7811
PVCB	0.7718	0.6102	0.7194	0.702	0.7548
PVCT	0.5893	0.674	0.9052	0.6974	0.6713
PACB	0.8662	0.8345	0.7377	0.8917	0.8157
PJC	0.9149	0.8474	0.536	0.8899	0.6969
PSCB	0.7549	0.7964	0.8777	0.4062	0.7847
VT	0.8015	0.863	0.9281	0.9923	0.8989
AFL	0.7101	0.642	0.7595	0.5575	0.7015
VF	0.6735	0.638	0.6908	0.8478	0.6601
VFL	0.6147	0.8093	0.7682	0.6563	0.644
WPW	0.5457	0.8818	0.7744	0.7065	0.6801
AE	0.5471	0.8778	0.5788	0.6499	0.5028
VE	0.6721	0.6964	0.6421	0.9552	0.612
JE	0.9221	0.8336	0.8059	0.8564	0.8292
SE	0.8066	0.8091	0.6815	0.7345	0.7907
AVBI	0.5163	0.6169	0.6443	0.7069	0.5473
AVBII	0.6926	0.9397	0.9574	0.786	0.806
AVBIII	0.7134	0.8862	0.3393	0.3961	0.6623
LBBB	0.6676	0.6584	0.36	0.7507	0.6651
ILBBB	0.6659	0.7961	0.5249	0.8208	0.7602
LAFB	0.6948	0.7628	0.5789	0.8507	0.6323
LPFB	0.5143	0.691	0.6791	0.6751	0.6168
RBBB	0.7293	0.7992	0.7402	0.6554	0.743

DeepLife

This DeepLife interface showcases core information for 11 common infectious diseases and the model’s recognition performance for these diseases:

ICD coding: the internationally standardized disease coding system, helping identify diseases precisely.

Positive/negative sample counts: the number of confirmed cases for the disease (positive) and the number of non-cases (negative).

AUC value: a metric measuring how well the model recognizes whether the patient has the disease (range 0–1). A higher score means better recognition.

For detailed data, download the reference material below.

Reference material (DeepLife data)

Next steps

ECG Basic API - ECG basic analysis
ECG Advanced API - ECG advanced features
Authentication - Get API Key
Usage examples - Complete integration example