Non-Invasive Monitoring of Transcatheter Heart Valve Using Photoplethysmography and Machine Learning.

Area of Science:

• Biomedical Engineering
• Cardiovascular Devices
• Artificial Intelligence in Medicine

Background:

• Transcatheter aortic valve implantation (TAVI) is a key treatment for aortic stenosis, but concerns about transcatheter heart valve (THV) durability and leaflet thrombosis limit its use in younger patients.
• Current monitoring methods for THV function are often invasive or intermittent, hindering early detection of potential complications.

Purpose of the Study:

• To develop and validate a non-invasive, continuous monitoring system for THV function using photoplethysmography (PPG) sensors and machine learning.
• To assess the system's ability to differentiate between healthy and impaired leaflet motion.

Main Methods:

• An in vitro mock circulatory loop was established with a compliant aortic phantom and a self-expanding Evolut FX THV.
• Two PPG sensors were utilized to capture flow signals under diverse hemodynamic conditions.
• Endoscopic imaging provided geometric orifice area (GOA) measurements for valve performance assessment.
• Machine learning models (regression and classification) were trained using PPG-derived metrics and flow variables.

Main Results:

• The regression model achieved an R² of 0.83, predicting geometric orifice area with an RMSE of 7.18 mm² and MAE of 5.58 mm².
• The classification model accurately identified reduced leaflet motion with 95% accuracy, 0.89 precision, and 0.91 recall.
• The study successfully demonstrated the PPG-based system's capability in monitoring THV performance.

Conclusions:

• Photoplethysmography (PPG) sensors combined with machine learning provide an effective non-invasive approach for monitoring transcatheter heart valve (THV) function.
• This method holds promise for the early detection of leaflet dysfunction, potentially enhancing the long-term management and outcomes for TAVI patients.