Detection of Mental Stress From Photoplethysmography for Under- Resourced Healthcare Systems

Mental stress adversely affects overall well-being, and prolonged exposure poses serious health risks and is underdiagnosed, especially in societies with under-resourced healthcare systems. Photoplethysmography (PPG) sensors are widely available and cost-effective, making them a practical modality for stress assessment. This study proposes SADDNet, a lightweight deep learning framework for objective stress detection from PPG signals, using a PPG device fabricated using an inexpensive sensor. The model was trained and evaluated using subject-independent leave-one-subject-out cross-validation on two datasets: CATSA (50 participants) and WESAD (15 participants), achieving 92.25% accuracy, 94.91% F1-score, and 0.895 AUC, and 98.11% accuracy, 97.33% F1-score, and 0.989 AUC, respectively. Additionally, evaluation on PPG signals acquired using the low-cost Max30102 sensor achieved 84.72% accuracy, 85.71% F1-score, and an AUC of 0.876 on a Raspberry Pi 5, demonstrating the model’s robustness and suitability for real-world, low-cost embedded platform deployments. SADDNet outperformed several convolutional and transformer-based state-of-the-art models, while requiring fewer parameters and faster inference. Cosine similarity-based explainability further confirmed that the model’s decisions aligned with physiological insights. These findings show the potential of developing such low-cost devices for real-time emotional stress monitoring.