Table 6 Comparison of state-of-the-art methods for detecting LVH using ECG.

From: Left ventricular hypertrophy detection using electrocardiographic signal

Study (year)

Method

Adopted features

Detection performances

Limitation

ACC

PRE

SEN

SPE

Others

ECG criteria

 Ref.16

2019

Cornell Product criteria

Multi-lead ECG

–

–

–

–

AUC = 0.62

Patients with age < 65 y were excluded

 Ref.17

2019

Combined criteria

12-lead ECG

–

0.401

0.379

0.915

AUC = 0.65

For a untreated hypertension cohort

 Ref.18

2021

Peguero-Lo Presti criteria

12-lead ECG

–

0.665

0.519

0.821

AUC = 0.7

Patients with age < 70 y were excluded

 Ref.15

2021

Peguero–Lo Presti

Multi-lead ECG

0.68

0.12

0.29

0.73

NPV = 0.89

 

Cornell voltage

0.86

0.24

0.12

0.95

NPV = 0.89

Cornell product

0.86

0.12

0.04

0.96

NPV = 0.89

Sokolow–Lyon voltage

0.81

0.13

0.12

0.89

NPV = 0.89

Sokolow–Lyon product

0.86

0.13

0.04

0.96

NPV = 0.89

Ref.12

2021

NCRCHS#1

criterion with multiple linear regression

3-lead ECG

–

–

0.90

0.36

AUC = 0.74

 

Ref.23

2021

CHCM#2

3-lead ECG

0.705

–

0.743

0.687

  

 Ref.14

2021

RaVL voltage-duration product

Lead aVL ECG

–

0.756

0.674

0.546

AUC = 0.64

In older individuals with left bundle branch block

Sokolow–Lyon criteria

3-lead ECG

–

0.75

0.261

0.818

AUC = 0.54

Machine learning models

 Ref.27

2018

Random forest

ECG data

0.661

–

0.58

0.709

  

 Ref.20

2019

BART#3-LVH criteria

26 features#4

–

0.299

0.29

0.946

AUC = 0.829

Participants without cardiovascular disease at enrollment

 Ref.21

2020

Decision tree with logistic regression

6 ECG features

0.733

–

0.816

0.693

  

 Ref.24

2020

Deep neural network

87 ECG features

0.736

0.73

0.667

0.782

  

 Ref.25

2020

Ensemble neural network#5

12-lead ECG signals + demographic features#6

0.851

–

0.613

0.896

AUC = 0.868

 

 Ref.26

2021

Convolution neural network

12-lead ECG

–

–

0.96

0.34

AUC = 0.653

 

 Ref.22

2021

GLMNet#7

34-feature 12-lead ECG

–

–

–

–

AUC = 0.873

 

This study

BPN

24-feature 12-lead ECG with ECG beat seg-mentation

0.961

0.958

0.966

0.956

 

Participants without arrhythmia

  1. ACC Accuracy, PRE Precision, SEN Sensitivity, SPE Specificity, AUC area under ROC curve, NVP negative predictive value.
  2. #1NCRCHS = Northeast China Rural Cardiovascular Health Study.
  3. #2CHCM = Cardiac Hypertrophy Computer-based Model.
  4. #3BART = Bayesian Additive Regression Trees.
  5. #426 features include age, sex, height, systolic and diastolic blood pressures and 21 ECG features.
  6. #5Ensemble neural network = convolutional neural network + deep neural network.
  7. #6demographic features include age, sex, weight and height.
  8. #7GLMNet = penalized logistic regression with the ElasticNet penalty.
  9. The bold and underlined values indicate the maximum and minimum values, respectively, of each column for ECG criteria or machine learning models.
Back to article page