Deep learning-based photoplethysmography classification for peripheral arterial disease detection: a proof-of-concept...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiovascular Diagnostics

Background:

• Peripheral arterial disease (PAD) is a significant vascular condition.
• Accurate and accessible diagnostic methods for PAD are crucial.
• Photoplethysmography (PPG) offers a potential non-invasive signal source.

Purpose of the Study:

• To evaluate a deep learning (DL) classification method for detecting PAD using toe PPG signals.
• To assess the diagnostic performance of a DL-based PPG (DLPPG) technique.
• To determine the feasibility of DLPPG as a screening tool for peripheral arterial disease.

Main Methods:

• A pretrained AlexNet model was fine-tuned using transfer learning for a 2-class PAD detection problem.
• PPG spectrogram images from 214 participants were used as input data.
• k-fold cross-validation (k=5 and k=10) was employed to evaluate diagnostic performance.

Main Results:

• The DLPPG method achieved an overall test accuracy of 88.9% (sensitivity 86.6%, specificity 90.2%).
• High sensitivity was observed for detecting major PAD (100.0%) and mild-moderate PAD (83.0%).
• The diagnostic performance was consistent across k=5 and k=10 cross-validation folds.

Conclusions:

• The DL-based PPG classification demonstrates substantial agreement with the ankle-brachial pressure index (ABPI) for PAD diagnosis.
• This novel, automated approach requires minimal pre-processing, enhancing its clinical utility.
• DLPPG presents a potentially valuable, low-cost, and portable diagnostic solution for PAD in various healthcare settings.