Detecting QT prolongation from a single-lead ECG with deep learning.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Technology

Background:

• Antiarrhythmic drug loading often requires prolonged hospitalization for QT prolongation monitoring.
• Current monitoring methods are resource-intensive and limit patient mobility.

Purpose of the Study:

• To develop and validate a deep learning model (QTNet) for inferring QT intervals from Lead-I ECG.
• To assess QTNet's capability in detecting drug-induced QT prolongation during Dofetilide loading.

Main Methods:

• QTNet, a deep neural network, was trained on over 3 million ECGs and validated on millions more from internal and external datasets.
• The model estimates QT intervals from Lead-I ECG, a common lead in ambulatory monitors.
• Performance was evaluated using mean absolute error, Pearson correlation, and sensitivity/specificity for detecting QT prolongation.

Main Results:

• QTNet achieved low mean absolute errors (12.63ms internal, 12.30ms external) and high Pearson correlation coefficients (0.91 internal, 0.92 external) for QT interval estimation.
• The model demonstrated 87% sensitivity and 77% specificity in detecting Dofetilide-induced QT prolongation.
• A negative predictive value exceeding 95% was observed for detecting drug-induced QT prolongation under specific pre-test probability conditions.

Conclusions:

• Deep learning models can accurately infer QT intervals from Lead-I ECG, facilitating remote patient monitoring.
• QTNet shows promise for tracking drug-induced QT prolongation, potentially enabling out-of-hospital care and reducing healthcare burdens.
• This technology could revolutionize antiarrhythmic drug management and patient safety.