Automatic Segmentation of Actigraphy Data Utilising Gradient Boosting Algorithm

As the popularity of decentralised clinical trials increases, there is a need to have a tool enabling remote assessment of sleep, while having good consistency with the golden standard, i.e. with polysomnography (PSG). This study aims to introduce a new approach to sleep assessment that utilises the...

Full description

Saved in:
Bibliographic Details
Published in:2021 44th International Conference on Telecommunications and Signal Processing (TSP) pp. 399 - 402
Main Authors: Mikulec, Marek, Mekyska, Jiri, Sigmund, Jan, Galaz, Zoltan, Brabenec, Lubos, Moravkova, Ivona, Rektorova, Irena
Format: Conference Proceeding
Language:English
Published: IEEE 26.07.2021
Subjects:
actigraphy
Boosting
Correlation
Correlation coefficient
machine learning
polysomnography
Sensitivity
Signal processing
Signal processing algorithms
sleep
sleep diary
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:As the popularity of decentralised clinical trials increases, there is a need to have a tool enabling remote assessment of sleep, while having good consistency with the golden standard, i.e. with polysomnography (PSG). This study aims to introduce a new approach to sleep assessment that utilises the modelling of actigraphy data by a gradient boosting algorithm. The method is compared to a conventional baseline technique in terms of sleep/wake stages detection accuracy in a dataset containing 55 recordings of actigraphy and PSG (acquired from 28 subjects). In addition, we explored how well the outputs of the new method agree with data acquired via sleep diaries in another dataset including 150 recordings (22 subjects). With 97% sensitivity and 73%specificity, the new method significantly outperformed the baseline one in modelling the PSG ground truth. On the other hand, it had a lower agreement with the patient-reported outcomes. The results suggest that a combination of both approaches could be a good alternative to the golden standard in remote sleep assessment studies.
DOI:10.1109/TSP52935.2021.9522650