Meeting News

Artificial intelligence identifies long QT syndrome when ECG did not

Michael J. Ackerman, MD, PhD
Michael J. Ackerman

BOSTON — Artificial intelligence using deep neural networks identified patients with congenital long QT syndrome whose ECG reading was normal, according to findings presented at the Heart Rhythm Society Annual Scientific Sessions.

Despite QT prolongation being a major feature of long QT syndrome, approximately 25% of patients with congenital long QT syndrome have a normal QTc at rest, according to the study background.

Michael J. Ackerman, MD, PhD, director of Mayo Clinic’s Genetic Heart Rhythm Clinic and the Windland Smith Rice Sudden Death Genomics Laboratory at Mayo Clinic, and colleagues investigated whether AI using deep neural networks is better than QTc alone at identifying patients with concealed long QT syndrome.

The researchers applied AI to data from lead one of a 12-lead mobile ECG (KardiaMobile and KardiaBand, AliveCor).

They analyzed ECGs from 1,048 patients with long QT syndrome and 1,010 patients initially identified as normal. The derivation cohort consisted of 72% of patients, with the rest comprising the validation cohort.

In the overall cohort, the area under the curve for the deep neural network was 0.91 vs. 0.84 for QTc alone, Ackerman and colleagues found.

Among patients with long QT syndrome but a normal resting QTc value, defined as less than 450 ms, the deep neural network distinguished long QT syndrome from normal with an AUC of 0.83 vs. 0.67 for QTc alone, according to the researchers.

Ackerman and colleagues also reported that the deep neural network was able to distinguish the three main genotypic subtypes from one another in patients with long QT syndrome (overall accuracy, 79%; AUC for LQTS1 vs. LQTS2 and LQTS3 = 0.86; AUC for LQTS2 vs. LQTS1 and LQTS3 = 0.91; AUC for LQTS3 vs. LQTS1 and LQTS2 = 0.87).

“Building on our previous work using Mayo Clinic’s proprietary T-wave fingerprint software, it is stunning that our ‘AI brain’ is distinguishing one patient who has a potentially life-threatening syndrome, [long QT syndrome], but a normal QTc, from a normal patient with the same QTc value by just staring at a single lead,” Ackerman said in a press release. – by Erik Swain

Reference:

Bos JM, et al. Abstract B-PO02-168. Presented at: Heart Rhythm Society Annual Scientific Sessions; May 9-12, 2018; Boston.

Disclosure: Ackerman reports he receives royalty payments from AliveCor, Blue Ox Health and StemoniX and has financial ties with Audentes Therapeutics, Boston Scientific, Gilead Sciences, InVitae, Medtronic, Myokardia and St. Jude Medical.

Michael J. Ackerman, MD, PhD
Michael J. Ackerman

BOSTON — Artificial intelligence using deep neural networks identified patients with congenital long QT syndrome whose ECG reading was normal, according to findings presented at the Heart Rhythm Society Annual Scientific Sessions.

Despite QT prolongation being a major feature of long QT syndrome, approximately 25% of patients with congenital long QT syndrome have a normal QTc at rest, according to the study background.

Michael J. Ackerman, MD, PhD, director of Mayo Clinic’s Genetic Heart Rhythm Clinic and the Windland Smith Rice Sudden Death Genomics Laboratory at Mayo Clinic, and colleagues investigated whether AI using deep neural networks is better than QTc alone at identifying patients with concealed long QT syndrome.

The researchers applied AI to data from lead one of a 12-lead mobile ECG (KardiaMobile and KardiaBand, AliveCor).

They analyzed ECGs from 1,048 patients with long QT syndrome and 1,010 patients initially identified as normal. The derivation cohort consisted of 72% of patients, with the rest comprising the validation cohort.

In the overall cohort, the area under the curve for the deep neural network was 0.91 vs. 0.84 for QTc alone, Ackerman and colleagues found.

Among patients with long QT syndrome but a normal resting QTc value, defined as less than 450 ms, the deep neural network distinguished long QT syndrome from normal with an AUC of 0.83 vs. 0.67 for QTc alone, according to the researchers.

Ackerman and colleagues also reported that the deep neural network was able to distinguish the three main genotypic subtypes from one another in patients with long QT syndrome (overall accuracy, 79%; AUC for LQTS1 vs. LQTS2 and LQTS3 = 0.86; AUC for LQTS2 vs. LQTS1 and LQTS3 = 0.91; AUC for LQTS3 vs. LQTS1 and LQTS2 = 0.87).

“Building on our previous work using Mayo Clinic’s proprietary T-wave fingerprint software, it is stunning that our ‘AI brain’ is distinguishing one patient who has a potentially life-threatening syndrome, [long QT syndrome], but a normal QTc, from a normal patient with the same QTc value by just staring at a single lead,” Ackerman said in a press release. – by Erik Swain

Reference:

Bos JM, et al. Abstract B-PO02-168. Presented at: Heart Rhythm Society Annual Scientific Sessions; May 9-12, 2018; Boston.

Disclosure: Ackerman reports he receives royalty payments from AliveCor, Blue Ox Health and StemoniX and has financial ties with Audentes Therapeutics, Boston Scientific, Gilead Sciences, InVitae, Medtronic, Myokardia and St. Jude Medical.

    See more from Heart Rhythm Society Scientific Sessions