Table 1 Open in new tab Summary of...

Step-by-step approach

CV diagnostic method

Study

Sample size

AI-based method

Problem addressed

Performance metric of the AI-based method

Comparison

First-line screening tool

Anamnesis

Rahman et al., 2013²⁴

470

Naïve Bayes
SVM
RF

Determine whether the AHA screening questionnaire correctly screens athletes if compared with ECG and ECHO

Accuracy—RF = 0.553

Auscultation

Viviers et al., 2017³³

131

Computer-assisted auscultation system

Determine whether a computer-assisted auscultation system has the ability to detect the presence of structural murmur if compared with a sports physician auscultation

Computer-assisted auscultation system—sensitivity = 100%, specificity = 50%

Physician auscultation—sensitivity 66.7%, specificity 66.7%

ECG

Długosz et al., 2018⁵¹

160

Use ECGs to estimate the level of cTnI in amateur athletes
Detect CAD in athletes

CAD detection—AUC = 0.91

Lombardi et al., 2018⁵⁰

Linear discriminant analysis

Determine whether patients with idiopathic ventricular arrhythmias with left bundle branch block and inferior axis morphology arrhythmia originated from the aortic sinus cusps or the right ventricular outflow tract

Accuracy = 0.947

Adetiba et al., 2017⁴⁸

ANN

Automatic heart defect detection (tachyarrhythmia, bradyarrhythmia, and HCM) for athletes

Accuracy = 0.9

ECG (wearables)

Adetiba et al., 2019⁴⁹

ANN

Develop a wearable-ECG that can be worn by athletes to help automatically detect defects

Accuracy = 1

Castillo-Atoche et al., 2022¹¹³

56 542 samples from 487 patients

CNN

Automatically detect arrhythmias in athletes in real time

Accuracy = 0.939

Second-line CV evaluation

EST

Qammar et al., 2022⁵⁵

ML algorithms

Correctly classify BP during EST in active population

ECHO

Narula et al., 2016⁷³

77 athletes and 62 HCM patients

Investigate the diagnostic value of a ML framework that incorporates speckle-tracking echocardiographic data for automated discrimination of HCM from physiological hypertrophy in athletes

Sensitivity = 96%, specificity = 77%

E/A (sensitivity = 79%, specificity = 77%), e′ (sensitivity = 86%, specificity = 82%), longitudinal strain (sensitivity = 68%, specificity = 77%)

Huang et al., 2022⁷²

598

Agglomerative hierarchical clustering
Multiple regression analysis

Identify athlete groups with similar characteristics
Investigate the validity of sport-specific adaption for evaluating athlete’s hearts

Third-line CV evaluation

CMR

Bernardino et al., 2020⁸¹ [NO_PRINTED_FORM]

77 controls and 89 athletes

Logistic regression
Principal component analysis
Statistical shape analysis

Highlight areas of the heart that undergo cardiac remodelling due to endurance exercise

Full-CV risk of athlete

Anthropometric data + demographic data + biomedical data + ECG

Barbieri et al., 2020¹²⁷

26 002

DT
Logistic regression

Classify whether an athlete is at cardiovascular risk or not

AUC = 0.78

Step-by-step approach

CV diagnostic method

Study

Sample size

AI-based method

Problem addressed

Performance metric of the AI-based method

Comparison

First-line screening tool

Anamnesis

Rahman et al., 2013²⁴

470

Naïve Bayes
SVM
RF

Determine whether the AHA screening questionnaire correctly screens athletes if compared with ECG and ECHO

Accuracy—RF = 0.553

Auscultation

Viviers et al., 2017³³

131

Computer-assisted auscultation system

Determine whether a computer-assisted auscultation system has the ability to detect the presence of structural murmur if compared with a sports physician auscultation

Computer-assisted auscultation system—sensitivity = 100%, specificity = 50%

Physician auscultation—sensitivity 66.7%, specificity 66.7%

ECG

Długosz et al., 2018⁵¹

160

Use ECGs to estimate the level of cTnI in amateur athletes
Detect CAD in athletes

CAD detection—AUC = 0.91

Lombardi et al., 2018⁵⁰

Linear discriminant analysis

Accuracy = 0.947

Adetiba et al., 2017⁴⁸

ANN

Automatic heart defect detection (tachyarrhythmia, bradyarrhythmia, and HCM) for athletes

Accuracy = 0.9

ECG (wearables)

Adetiba et al., 2019⁴⁹

ANN

Develop a wearable-ECG that can be worn by athletes to help automatically detect defects

Accuracy = 1

Castillo-Atoche et al., 2022¹¹³

56 542 samples from 487 patients

CNN

Automatically detect arrhythmias in athletes in real time

Accuracy = 0.939

Second-line CV evaluation

EST

Qammar et al., 2022⁵⁵

ML algorithms

Correctly classify BP during EST in active population

ECHO

Narula et al., 2016⁷³

77 athletes and 62 HCM patients

Investigate the diagnostic value of a ML framework that incorporates speckle-tracking echocardiographic data for automated discrimination of HCM from physiological hypertrophy in athletes

Sensitivity = 96%, specificity = 77%

E/A (sensitivity = 79%, specificity = 77%), e′ (sensitivity = 86%, specificity = 82%), longitudinal strain (sensitivity = 68%, specificity = 77%)

Huang et al., 2022⁷²

598

Agglomerative hierarchical clustering
Multiple regression analysis

Identify athlete groups with similar characteristics
Investigate the validity of sport-specific adaption for evaluating athlete’s hearts

Third-line CV evaluation

CMR

Bernardino et al., 2020⁸¹ [NO_PRINTED_FORM]

77 controls and 89 athletes

Logistic regression
Principal component analysis
Statistical shape analysis

Highlight areas of the heart that undergo cardiac remodelling due to endurance exercise

Full-CV risk of athlete

Anthropometric data + demographic data + biomedical data + ECG

Barbieri et al., 2020¹²⁷

26 002

DT
Logistic regression

Classify whether an athlete is at cardiovascular risk or not

AUC = 0.78

Table 1

Open in new tab

Summary of studies analysing various artificial intelligence-based cardiovascular diagnostic techniques in athletes

Step-by-step approach	CV diagnostic method	Study	Sample size	AI-based method	Problem addressed	Performance metric of the AI-based method	Comparison
First-line screening tool	Anamnesis	Rahman et al., 2013²⁴	470	Naïve Bayes SVM RF	Determine whether the AHA screening questionnaire correctly screens athletes if compared with ECG and ECHO	Accuracy—RF = 0.553	NA
	Auscultation	Viviers et al., 2017³³	131	Computer-assisted auscultation system	Determine whether a computer-assisted auscultation system has the ability to detect the presence of structural murmur if compared with a sports physician auscultation	Computer-assisted auscultation system—sensitivity = 100%, specificity = 50%	Physician auscultation—sensitivity 66.7%, specificity 66.7%
	ECG	Długosz et al., 2018⁵¹	160	DT LR	Use ECGs to estimate the level of cTnI in amateur athletes Detect CAD in athletes	CAD detection—AUC = 0.91	NA
		Lombardi et al., 2018⁵⁰	26	Linear discriminant analysis	Determine whether patients with idiopathic ventricular arrhythmias with left bundle branch block and inferior axis morphology arrhythmia originated from the aortic sinus cusps or the right ventricular outflow tract	Accuracy = 0.947	NA
		Adetiba et al., 2017⁴⁸	40	ANN	Automatic heart defect detection (tachyarrhythmia, bradyarrhythmia, and HCM) for athletes	Accuracy = 0.9	NA
	ECG (wearables)	Adetiba et al., 2019⁴⁹	40	ANN	Develop a wearable-ECG that can be worn by athletes to help automatically detect defects	Accuracy = 1	NA
	ECG (wearables)	Castillo-Atoche et al., 2022¹¹³	56 542 samples from 487 patients	CNN	Automatically detect arrhythmias in athletes in real time	Accuracy = 0.939	NA
Second-line CV evaluation	EST	Qammar et al., 2022⁵⁵	19	ML algorithms	Correctly classify BP during EST in active population	NA	NA
	ECHO	Narula et al., 2016⁷³	77 athletes and 62 HCM patients	SVM RF ANN	Investigate the diagnostic value of a ML framework that incorporates speckle-tracking echocardiographic data for automated discrimination of HCM from physiological hypertrophy in athletes	Sensitivity = 96%, specificity = 77%	E/A (sensitivity = 79%, specificity = 77%), e′ (sensitivity = 86%, specificity = 82%), longitudinal strain (sensitivity = 68%, specificity = 77%)
	ECHO	Huang et al., 2022⁷²	598	Agglomerative hierarchical clustering Multiple regression analysis	Identify athlete groups with similar characteristics Investigate the validity of sport-specific adaption for evaluating athlete’s hearts	NA	NA
Third-line CV evaluation	CMR	Bernardino et al., 2020⁸¹ [NO_PRINTED_FORM]	77 controls and 89 athletes	Logistic regression Principal component analysis Statistical shape analysis	Highlight areas of the heart that undergo cardiac remodelling due to endurance exercise	NA	NA
Full-CV risk of athlete	Anthropometric data + demographic data + biomedical data + ECG	Barbieri et al., 2020¹²⁷	26 002	DT Logistic regression	Classify whether an athlete is at cardiovascular risk or not	AUC = 0.78	NA

Step-by-step approach	CV diagnostic method	Study	Sample size	AI-based method	Problem addressed	Performance metric of the AI-based method	Comparison
First-line screening tool	Anamnesis	Rahman et al., 2013²⁴	470	Naïve Bayes SVM RF	Determine whether the AHA screening questionnaire correctly screens athletes if compared with ECG and ECHO	Accuracy—RF = 0.553	NA
	Auscultation	Viviers et al., 2017³³	131	Computer-assisted auscultation system	Determine whether a computer-assisted auscultation system has the ability to detect the presence of structural murmur if compared with a sports physician auscultation	Computer-assisted auscultation system—sensitivity = 100%, specificity = 50%	Physician auscultation—sensitivity 66.7%, specificity 66.7%
	ECG	Długosz et al., 2018⁵¹	160	DT LR	Use ECGs to estimate the level of cTnI in amateur athletes Detect CAD in athletes	CAD detection—AUC = 0.91	NA
		Lombardi et al., 2018⁵⁰	26	Linear discriminant analysis	Determine whether patients with idiopathic ventricular arrhythmias with left bundle branch block and inferior axis morphology arrhythmia originated from the aortic sinus cusps or the right ventricular outflow tract	Accuracy = 0.947	NA
		Adetiba et al., 2017⁴⁸	40	ANN	Automatic heart defect detection (tachyarrhythmia, bradyarrhythmia, and HCM) for athletes	Accuracy = 0.9	NA
	ECG (wearables)	Adetiba et al., 2019⁴⁹	40	ANN	Develop a wearable-ECG that can be worn by athletes to help automatically detect defects	Accuracy = 1	NA
	ECG (wearables)	Castillo-Atoche et al., 2022¹¹³	56 542 samples from 487 patients	CNN	Automatically detect arrhythmias in athletes in real time	Accuracy = 0.939	NA
Second-line CV evaluation	EST	Qammar et al., 2022⁵⁵	19	ML algorithms	Correctly classify BP during EST in active population	NA	NA
	ECHO	Narula et al., 2016⁷³	77 athletes and 62 HCM patients	SVM RF ANN	Investigate the diagnostic value of a ML framework that incorporates speckle-tracking echocardiographic data for automated discrimination of HCM from physiological hypertrophy in athletes	Sensitivity = 96%, specificity = 77%	E/A (sensitivity = 79%, specificity = 77%), e′ (sensitivity = 86%, specificity = 82%), longitudinal strain (sensitivity = 68%, specificity = 77%)
	ECHO	Huang et al., 2022⁷²	598	Agglomerative hierarchical clustering Multiple regression analysis	Identify athlete groups with similar characteristics Investigate the validity of sport-specific adaption for evaluating athlete’s hearts	NA	NA
Third-line CV evaluation	CMR	Bernardino et al., 2020⁸¹ [NO_PRINTED_FORM]	77 controls and 89 athletes	Logistic regression Principal component analysis Statistical shape analysis	Highlight areas of the heart that undergo cardiac remodelling due to endurance exercise	NA	NA
Full-CV risk of athlete	Anthropometric data + demographic data + biomedical data + ECG	Barbieri et al., 2020¹²⁷	26 002	DT Logistic regression	Classify whether an athlete is at cardiovascular risk or not	AUC = 0.78	NA

ANN, automatic neural network; AUC, area under the curve; BP, blood pressure; CAD, coronary artery disease; CMR, cardiac magnetic resonance; CNN, convolutional neural network; cTnI, cardiac troponin; DT, decision tree; ECG, electrocardiogram; ECHO, echocardiography; EST, exercise stress test; HCM, hypertrophic cardiomyopathy; LR, logistic regression; ML, machine learning; NA, not available; RF, random forest; SVM, support vector machine.

This Feature Is Available To Subscribers Only