Optimizing deep learning models to predict obstructive sleep apnea from abdominal breathing

The project will aim to accurately predict obstructive sleep apnea from measurements of abdominal breathing
Master
Sleep apnea test

Obstructive sleep apnea is a potentially serious sleep disorder. It causes breathing to repeatedly stop and start during sleep. Nearly 1 billion people world wide are estimated [1] to suffer from undiagnosed obstructive sleep apnea. Sleep apnea has also been linked to stroke [2], neurodegenerative diseases [3], and even pancreatic and kidney cancer [4].

This project aims to optimise deep learning models to predict obstructive sleep apnea from abdominal breathing data. The data is obtained from 8000 overnight recordings taken as part of the Sleep Health Heart Study(sleepdata.org). We have developed deep learning models based on residual neural networks and long short-term neural networks to classify sleep apnea based on abdominal breathing measurements.

We want to go a step further and try to achieve a prediction accuracy of over 90%. This will be done by optimizing our deep learning models with hyper parameter tuning, varying input data length, and comparison of different deep learning architectures. The student will use time series classification deep learning models such as residual neural networks [5] and long short-term memory neural networks [6] to classify abdominal breathing data as apnea or non-apnea.

Goal

1) To optimise deep learning models. to predict sleep apnea from abdominal breathing data
2) Achieve over 90% accuracy on test data
3) Test model with real-world data garnered by the FLOW sensor (http://www.sweetzpot.com/flow)

Learning outcome

    1) Handling and pre-processing large time series datasets efficiently
    2) Deep learning models for time series data classification
    3) Strategies for hyper-parameter tuning to achieve optimal prediction performance
    4) Experience working in an interdisciplinary setup along with sleep doctors

    Qualifications

      1) Bachelors in any scientific discipline with strong data analysis component
      2) Strong mathematical foundations necessary to understand deep learning algorithms
      3) Intermediate to advanced Python Programming experience
      4) Great interest in developing a complete product to screen people for obstructive sleep apnea

      Supervisors

      • Sagar Sen
      • Pierre Bernabet, PhD student
      • Nils Henrik Holmedahl, LHL-sykehuset, Gardemoen

      Collaboration partners

      LHL-sykehuset, Gardemoen

      References

      1) Benjafield, A., et al. "Global prevalence of obstructive sleep apnea in adults: estimation using currently available data." B67. RISK AND PREVALENCE OF SLEEP DISORDERED BREATHING. American Thoracic Society, 2018. A3962-A3962.

      2) Yaggi, H. Klar, et al. "Obstructive sleep apnea as a risk factor for stroke and death." New England Journal of Medicine 353.19 (2005): 2034-2041.

      3) Daulatzai, Mak Adam. "Evidence of neurodegeneration in obstructive sleep apnea: relationship between obstructive sleep apnea and cognitive dysfunction in the elderly." Journal of neuroscience research 93.12 (2015): 1778-1794.

      4) Gozal, David, Sandra A. Ham, and Babak Mokhlesi. "Sleep apnea and cancer: analysis of a nationwide population sample." Sleep 39.8 (2016): 1493-1500.

      5) Fawaz, Hassan Ismail, et al. "Deep learning for time series classification: a review." Data Mining and Knowledge Discovery 33.4 (2019): 917-963.

      6) Van Steenkiste, Tom, et al. "Automated Sleep Apnea Detection in Raw Respiratory Signals using Long Short-Term Memory Neural Networks." IEEE journal of biomedical and health informatics (2018).

        Contact person