[Paper Review] Signal Quality Assessment of Photoplethysmogram Signals using Quantum Pattern Recognition and lightweight CNN Architecture

Tamaghno Chatterjee, Aayushman Ghosh+1 more|arXiv (Cornell University)|Feb 1, 2022

Non-Invasive Vital Sign Monitoring4 citations

TL;DR

This paper proposes a lightweight 2D CNN model combined with a novel Quantum Pattern Recognition (QPR) technique to classify photoplethysmogram (PPG) signals as 'good' or 'bad' quality. The method transforms 5-second PPG segments into 20×500 pixel grayscale images via QPR, which are then classified by the slim CNN architecture, achieving 98.3% accuracy, 99.3% sensitivity, and 98.9% F1-score on the University of Queensland dataset.

ABSTRACT

Photoplethysmography (PPG) signal comprises physiological information related to cardiorespiratory health. However, while recording, these PPG signals are easily corrupted by motion artifacts and body movements, leading to noise enriched, poor quality signals. Therefore ensuring high-quality signals is necessary to extract cardiorespiratory information accurately. Although there exists several rule-based and Machine-Learning (ML) - based approaches for PPG signal quality estimation, those algorithms' efficacy is questionable. Thus, this work proposes a lightweight CNN architecture for signal quality assessment employing a novel Quantum pattern recognition (QPR) technique. The proposed algorithm is validated on manually annotated data obtained from the University of Queensland database. A total of 28366, 5s signal segments are preprocessed and transformed into image files of 20 x 500 pixels. The image files are treated as an input to the 2D CNN architecture. The developed model classifies the PPG signal as `good' or `bad' with an accuracy of 98.3% with 99.3% sensitivity, 94.5% specificity and 98.9% F1-score. Finally, the performance of the proposed framework is validated against the noisy `Welltory app' collected PPG database. Even in a noisy environment, the proposed architecture proved its competence. Experimental analysis concludes that a slim architecture along with a novel Spatio-temporal pattern recognition technique improve the system's performance. Hence, the proposed approach can be useful to classify good and bad PPG signals for a resource-constrained wearable implementation.

Motivation & Objective

• To address the challenge of motion artifact-corrupted PPG signals in wearable health monitoring systems.
• To improve signal quality assessment (SQA) accuracy beyond rule-based and conventional ML methods.
• To develop a computationally efficient deep learning model suitable for deployment on resource-constrained wearable devices.
• To validate the proposed framework on both clean, manually annotated data and real-world noisy data from the Welltory app.
• To demonstrate the superiority of QPR-processed images combined with a lightweight CNN for SQA tasks.

Proposed method

• Transformed 1D PPG signals into 20×500 pixel grayscale images using a novel Quantum Pattern Recognition (QPR) algorithm to extract spatio-temporal patterns.
• Designed a lightweight 2D CNN architecture based on a 'Slim module' micro-architecture to reduce model complexity and parameter count.
• Trained the CNN to classify images as 'good' or 'bad' quality PPG signals using cross-entropy loss and Adam optimization.
• Preprocessed 28,366 five-second PPG segments from the University of Queensland database, applying standard normalization and artifact detection.
• Validated performance using five evaluation metrics: sensitivity, specificity, F1-score, and accuracy.
• Compared the proposed model against three baseline methods: SQIs+SVM, HOG+MLP, and 1D-CNN, using the same dataset and evaluation protocol.

Experimental results

Research questions

• RQ1 Can a lightweight CNN architecture combined with QPR-processed PPG images outperform existing rule-based and ML-based SQA methods?
• RQ2 How does the proposed method perform on real-world noisy PPG data collected from mobile apps like Welltory?
• RQ3 Does the integration of quantum-inspired pattern recognition enhance the discriminative power of deep learning models in PPG quality assessment?
• RQ4 To what extent does the proposed model maintain high performance while minimizing computational complexity for wearable deployment?
• RQ5 How does the performance of the proposed model compare to state-of-the-art deep learning models using the same dataset and evaluation metrics?

Key findings

• The proposed QPR-based lightweight CNN achieved 98.3% accuracy, 99.3% sensitivity, 94.5% specificity, and 98.9% F1-score on the University of Queensland dataset.
• The model outperformed all baseline methods, including SQIs+SVM (84.4% accuracy), 1D-CNN (95.8% accuracy), and HOG-MLP (85% accuracy).
• On the noisy Welltory app dataset, the model maintained high performance, demonstrating robustness in real-world conditions.
• Compared to prior DL work (Chen, 2021), the proposed method achieved higher sensitivity (99.3% vs. 96.6%) and specificity (94.5% vs. 69.7%) using QPR images.
• When evaluated on STFT images, the proposed model still outperformed Chen et al. (2021) in all metrics, including accuracy (94.0% vs. 93.6%) and F1-score (96.2% vs. 96.0%).
• The model's performance was superior despite using a smaller dataset (28,366 segments) than Chen et al. (58,040 segments), indicating strong generalization and efficiency.