Sleep Well
Insomnia, defined by difficulties in falling asleep, staying asleep, or waking up prematurely, is a common health issue in modern society, with approximately half of adults experiencing its symptoms at some point in their lives. The effects of insomnia are far-reaching, leading to psychological and physical deterioration. Individuals suffering from insomnia face an elevated risk of developing cardiovascular diseases, mental health disorders such as depression and anxiety, and cognitive decline. On a broader scale, insomnia contributes to decreased workplace productivity and an increased risk of work-related accidents due to fatigue and impaired concentration. This imposes a significant socioeconomic burden, affecting public health systems and economic productivity. Traditionally, sleep quality has been assessed using self-reported tools such as the Pittsburgh Sleep Quality Index (PSQI). However, these assessments are often limited by their subjective nature, making them prone to bias and inconsistency. As a result, these tools may fail to accurately capture true sleep behaviours, limiting their usefulness in diagnosing sleep disorders. In contrast, Polysomnography (PSG) provides a more objective and comprehensive evaluation of sleep by recording various physiological parameters. However, PSG is resource-intensive, requiring expert analysis, and thus impractical for widespread use, particularly given the prevalence of sleep disorders like insomnia.
To address these limitations, a project has been launched with the goal of developing automated tools for PSG analysis using artificial intelligence (AI). The project seeks to provide a scalable, objective, and accurate method for evaluating sleep quality, as well as to develop digital therapeutics aimed at treating insomnia. With support from the CIND Award, the project has made significant strides. The funding enabled the appointment of two research assistants, Hoachen Liu and Timo Hromadka, who are collaborating with awardee Sam Nallaperuma-Herzberg on the analysis of PSG data and the creation of sleep therapeutics, respectively.
One of the project's major achievements is the development of a novel AI-based model for automatic sleep staging using single-channel EEG data. This approach focuses on temporal dependencies in sleep patterns, which enables the model to outperform both simple and complex baseline models. It has been validated on large datasets containing both healthy individuals and patients with insomnia, demonstrating superior performance in accurately classifying sleep stages. This innovation holds great potential for improving the diagnosis and treatment of sleep disorders. This work has been accepted for presentation at the Sleep Europe Congress 2024.
In parallel, the project is progressed on the development of digital therapeutics, specifically focusing on sleep-inducing music as a therapeutic intervention. Research supports the use of music therapy to promote relaxation and improve sleep quality. The developed preliminary model is capable of generating 20-second samples of sleep-inducing music, with early evaluations using the Fréchet Audio Distance (FAD) metric showing promising results in terms of audio quality. Plans are in place to conduct qualitative evaluations with human participants.
Looking forward, the project team aims to secure further funding from external sources such as UK Research and Innovation (UKRI) to advance the research and development of these technologies. The project’s long-term goal is to translate these innovations into clinical practice, providing healthcare professionals with tools to objectively evaluate sleep and offer digital therapeutics to treat sleep disorders like insomnia. The team expresses gratitude to CIND for its support in launching this project, as well as for its continued guidance and advice.
Dr. Sam Nellaperuma-Herzberg