Enhanced Wrist-Based Arterial Stiffness Prediction Via Improved MFCC Features from PPG Signals

Arterial stiffness is a critical indicator of cardiovascular risk; however, traditional assessment methods require specialized equipment and professional medical personnel, limiting their widespread adoption. With the growing popularity of smartwatches, wrist-worn photoplethysmography (PPG) signals offer a convenient data source for early screening of arterial stiffness. This study proposes an improved Mel-frequency cepstral coefficient (MFCC) feature extraction method, optimized through a modified Mel scale to more effectively extract arterial stiffness-related features from wrist-based PPG signals. Wrist PPG data were collected from 444 participants, using carotidfemoral pulse wave velocity (cfPWV) as the gold standard for arterial stiffness. Among 104 features, the improved MFCC features showed the strongest correlation with cfPWV (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$r=-0.361$</tex>), second only to basic physiological indicators such as age and BMI. Incorporating these novel MFCC features, the model accuracy for binary classification of cfPWV (threshold =10 m/s) increased from 0.8750 to 0.9091. The results indicate that improved MFCC features derived from wrist PPG signals significantly enhance the accuracy of arterial stiffness prediction, demonstrating potential for daily health monitoring and large-scale population screening, and paving the way for future deployment in wearable health systems.