2024
A multilabel deep learning model for the detection of conduction and rhythm disorders from PDF outputs of a widely available portable ECG Device
Sangha V, Dhingra L, Khera R. A multilabel deep learning model for the detection of conduction and rhythm disorders from PDF outputs of a widely available portable ECG Device. European Heart Journal 2024, 45: ehae666.3437. DOI: 10.1093/eurheartj/ehae666.3437.Peer-Reviewed Original ResearchDeep learning modelsPDF outputArtificial intelligenceLearning modelsConvolutional neural networkPortable devicesPortable ECG deviceMultilabel modelEfficientNet-B3Neural networkECG outputWearable ECGHeld-out subsetCurrent algorithmsSingle-lead ECGSynthetic ECGECG deviceWearableClinical ECGClinical labelsAlgorithmCommercial devicesDetection of conductivityNo current algorithmPDFArtificial intelligence applied to electrocardiographic images for scalable screening of cardiac amyloidosis
Sangha V, Oikonomou E, Krumholz H, Miller E, Khera R. Artificial intelligence applied to electrocardiographic images for scalable screening of cardiac amyloidosis. European Heart Journal 2024, 45: ehae666.3436. DOI: 10.1093/eurheartj/ehae666.3436.Peer-Reviewed Original ResearchATTR-CMBone scintigraphy scansTransthyretin amyloid cardiomyopathyPositive predictive valueAI-ECG algorithmCardiac amyloidosisScintigraphy scanAmyloid cardiomyopathyAI-ECGSex-matchedDevelopment cohortMyocardial remodelingUnder-diagnosedUnder-treatedMatched controlsPredictive valueUnder-recognizedTransthyretin stabilizersConvolutional neural networkPatientsECGArtificial intelligenceHospitalPrevalenceTransthyretinBiometric contrastive learning for data-efficient deep learning from electrocardiographic images
Sangha V, Khunte A, Holste G, Mortazavi B, Wang Z, Oikonomou E, Khera R. Biometric contrastive learning for data-efficient deep learning from electrocardiographic images. Journal Of The American Medical Informatics Association 2024, 31: 855-865. PMID: 38269618, PMCID: PMC10990541, DOI: 10.1093/jamia/ocae002.Peer-Reviewed Original ResearchLabeled training dataContrastive learningECG imagesLabeled dataTraining dataDeep learningProportions of labeled dataArtificial intelligenceSelf-supervised contrastive learningTraditional supervised learningConvolutional neural networkHeld-out test setSupervised learningPretraining strategyBiometric signatureImageNet initializationPretraining approachNeural networkImageNetAI modelsImage objectsTest setLearningDetect atrial fibrillationEquivalent performance
2023
Detection of left ventricular systolic dysfunction from single-lead electrocardiography adapted for portable and wearable devices
Khunte A, Sangha V, Oikonomou E, Dhingra L, Aminorroaya A, Mortazavi B, Coppi A, Brandt C, Krumholz H, Khera R. Detection of left ventricular systolic dysfunction from single-lead electrocardiography adapted for portable and wearable devices. Npj Digital Medicine 2023, 6: 124. PMID: 37433874, PMCID: PMC10336107, DOI: 10.1038/s41746-023-00869-w.Peer-Reviewed Original ResearchArtificial intelligenceRandom Gaussian noiseNoisy electrocardiogramGaussian noiseElectrocardiogram (ECGWearable devicesSingle-lead electrocardiogramPortable devicesSNRWearableNoiseDevice noiseRepositoryAI-based screeningIntelligenceDetectionDevicesNoise sourcesVentricular systolic dysfunctionModelElectrocardiogramSingle-lead electrocardiographyTraining
2022
Automated multilabel diagnosis on electrocardiographic images and signals
Sangha V, Mortazavi BJ, Haimovich AD, Ribeiro AH, Brandt CA, Jacoby DL, Schulz WL, Krumholz HM, Ribeiro ALP, Khera R. Automated multilabel diagnosis on electrocardiographic images and signals. Nature Communications 2022, 13: 1583. PMID: 35332137, PMCID: PMC8948243, DOI: 10.1038/s41467-022-29153-3.Peer-Reviewed Original ResearchConceptsConvolutional neural networkArtificial intelligenceApplication of AISignal-based dataSignal-based modelElectrocardiographic imagesECG imagesGrad-CAMImage-based modelsNeural networkDiagnosis modelECG signalsImagesClinical labelsValidation setLabelsExternal validation setMultilabelIntelligenceNetworkApplicationsModelBroad useSetBroader setting