Reconstruction of the respiratory rate using wrist-based PPG, accelerometer and gyroscope
Koivisto, Piita (2025-06-16)
Reconstruction of the respiratory rate using wrist-based PPG, accelerometer and gyroscope
(16.06.2025)
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
avoin
Julkaisun pysyvä osoite on:
https://urn.fi/URN:NBN:fi-fe2025062473238
https://urn.fi/URN:NBN:fi-fe2025062473238
Tiivistelmä
Breathing is an important indicator of serious illnesses. Traditional measurement methods are often uncomfortable and not suitable for everyday use. Consequently, there is a growing need for accurate and easily adoptable solutions for continuous respiratory rate monitoring. Wearable approaches that utilize biosignals have emerged as a promising way to address this need. However, biosignals often contain a significant amount of noise, which can reduce the accuracy of the measurements.
A wide range of methods have been proposed for respiratory rate estimation, including both traditional signal processing techniques and modern machine learning approaches. In this thesis, we utilize accelerometer, gyroscope, and photoplethysmography (PPG) signals collected from the wrist to develop a pipeline for accurate respiratory rate estimation. The goal is to produce a pipeline that combines waveform analysis techniques, such as the Fast Fourier Transform and modulation extraction methods, to enable efficient estimation without relying solely on computationally expensive machine learning algorithms. However, we do employ machine learning methods to assess signal quality as accurately as possible, since signal quality has a significant impact on subsequent analysis. An additional objective is to assess whether the fusion of estimates from multiple modalities yields improved accuracy relative to the use of PPG alone. We adopted a pipeline proposed in previous research as a foundation and explored ways to improve its performance.
The pipeline based on PPG produced results comparable to those of other methods reported in the literature; however, it did not outperform the most recent deep learning approaches. Although the fusion method achieved similar accuracy, it was more computationally intensive. Therefore, relying solely on the PPG pipeline appears to be a more efficient and accurate approach. Notably, the fusion pipeline performed better than the baseline pipeline adopted from previous work.
A wide range of methods have been proposed for respiratory rate estimation, including both traditional signal processing techniques and modern machine learning approaches. In this thesis, we utilize accelerometer, gyroscope, and photoplethysmography (PPG) signals collected from the wrist to develop a pipeline for accurate respiratory rate estimation. The goal is to produce a pipeline that combines waveform analysis techniques, such as the Fast Fourier Transform and modulation extraction methods, to enable efficient estimation without relying solely on computationally expensive machine learning algorithms. However, we do employ machine learning methods to assess signal quality as accurately as possible, since signal quality has a significant impact on subsequent analysis. An additional objective is to assess whether the fusion of estimates from multiple modalities yields improved accuracy relative to the use of PPG alone. We adopted a pipeline proposed in previous research as a foundation and explored ways to improve its performance.
The pipeline based on PPG produced results comparable to those of other methods reported in the literature; however, it did not outperform the most recent deep learning approaches. Although the fusion method achieved similar accuracy, it was more computationally intensive. Therefore, relying solely on the PPG pipeline appears to be a more efficient and accurate approach. Notably, the fusion pipeline performed better than the baseline pipeline adopted from previous work.