Design of Knowledge Driven and Data Driven Algorithms for Neurodegenerative Diseases

Parkinson’s disease (PD) is one of the most important neurodegenerative disorders with major impact on quality life, morbidity and mortality and there are still not valid and world-recognized biomarkers for identification of early and late stages of the disease.

Sleep disturbances are common symptoms of PD, and some studies have found that there is a strong relation between a specific sleep disorder (REM sleep behavior disorder - RBD) and Parkinsonim. This relation suggests the idea that sleep disturbances might precede the clinical diagnosis of PD.

Therefore, analysis of sleep carried out with polysomnography (which typically includes the recording of electroencephalography, electrooculography, electromyography, electrocardiography, respiratory airflow and peripheral pulse oximetry) might be used for early disease identification.

Previous studies have shown that automatic data-driven and knowledge-driven algorithms are able to identify altered eye movements during sleep, altered sleep stability, and a decrease and change of morphology of spindles in patients suffering from PD and RBD when compared to healthy subjects. However, none of these biomarkers can be considered as a stand-alone that is able to identify the diseased condition.

Therefore, there is the necessity of further investigations in this topic. Improvements might be obtained by combining the previous mentioned biomarkers, by researching and studying new potential biomarkers, by analyzing more data from all over the world in order to study the usability and variability of such biomarkers across cities and countries, and by developing innovative and robust algorithms that can improve sleep analysis and work properly for diseased subjects as well as in noisy conditions.

In this context, the current Ph.D. project has the aim of performing data analysis on national and international clinically recorded sleep studies in healthy and subjects suffering from PD, RBD and other disturbances. The analysis of these sleep studies has the objective of discovering optimal features for PD identification thanks to the design of new robust algorithms based on advanced mathematical tools.

The selected features will be then submitted to advanced new machine learning algorithms to classify and differentiate disease and normal events. Moreover, a purely data-driven approach will also be pursued due to the scientific hypothesis that data-driven definition of features can be closer to the real physiological brain activities, and may outperform state-of-the-art manual clinical (partly ambiguous) standards for classifying sleep studies.

The new algorithms that will be developed during the Ph.D. project have the aim of achieving high accuracy and performances that can be world-wide recognized as clinically applicable.

The project is co-funded by Rigshospitalet Glostrup and DTU.