Convolutional neural network-based method for the real-time detection of reflex syncope during head-up tilt test

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Reflex syncope (RS) is a common condition resulting from autonomic nervous system dysregulation.
• Current diagnosis via head-up tilt test (HUTT) is time-consuming and can induce symptoms.
• Existing automated methods often rely on manual feature extraction, which is sensitive to noise and methodology.

Purpose of the Study:

• To develop a real-time deep learning system for detecting reflex syncope.
• To eliminate the need for manual feature extraction in syncope detection.
• To create a more efficient and convenient diagnostic tool for RS.

Main Methods:

• An end-to-end deep learning architecture utilizing residual and squeeze-and-excitation blocks was employed.
• The system analyzes raw blood pressure signals for real-time RS risk assessment.
• A dataset of 1348 patients (1291 normal, 57 with RS) was used for development and validation.

Main Results:

• The deep learning model achieved an area under the ROC curve of 0.972.
• At a threshold of 0.75, the method demonstrated 94.74% sensitivity and 94.27% specificity.
• RS was detected an average of 165.35 seconds prior to its clinical occurrence.

Conclusions:

• The proposed deep learning method surpasses conventional techniques for reflex syncope detection.
• Requiring only blood pressure monitoring enhances the method's applicability and convenience.
• This advancement facilitates the development of safer, more efficient, and patient-friendly RS diagnostic systems.