Parkinson’s disease (PD) is a rapidly growing neurological disorder worldwide, with an estimated doubling of patients to 12 million by 2040. As a progressive condition, PD leads to severe impairments in the late stages, and novel treatments to slow or halt the underlying neurodegeneration are being developed. However, testing their efficacy is hindered by the lack of sensitive and reliable endpoints. Currently, the progression of PD is mainly evaluated by clinical experts who measure the symptom severity when patients visit the clinic, but this method is limited by intra- and inter-rater variability, as well as observer effects and common response fluctuations associated with symptomatic medication.
Our vision is that remote monitoring, combined with trustworthy AI, has the potential to transform clinical trials and care for people with Parkinson’s disease, by providing objective and frequent insights into how patients function in their home environment. The AI for Parkinson Lab will focus on two different systems for remote monitoring: wearable sensors and video recordings. Both systems can be used to capture hallmark motor (e.g., tremor, bradykinesia, and gait impairments) and non-motor symptoms (e.g., sleep, autonomic impairments) of PD. The challenge is to extract reliable information about these PD symptoms from the heterogeneous real-life raw sensor signals, which not only reflect PD-specific signs but also behavioral and environmental factors.
We now have access to multiple worldwide unique datasets which include wearable sensor data and video recordings of large groups of PD patients. An important example is the Personalized Parkinson Project (PPP), which originated from the existing collaboration between Verily Life Sciences and Radboudumc. PPP is a single-center cohort study including 550 early PD patients who are followed for up to 3 years. In addition to elaborate annual in-clinic assessments (including expert evaluation of symptoms, video recordings, fMRI, ECG, and collection of biosamples), patients are monitored in real-life using a multi-sensor smartwatch, which continuously collects raw sensor data from multiple sensor types during the full 3-year follow-up period.
The AI for Parkinson Lab aims to use this and other datasets to develop AI-based tools to process raw wearable sensor and video data into digital biomarkers to monitor Parkinson’s disease in real-life. We will investigate whether such biomarkers are indeed more reliable and sensitive to change compared to current clinical assessments and to optimally use a large amount of unlabeled data, we will investigate the value of promising developments in unsupervised and semi-supervised deep learning. To support the applicability of the developed tools, we will investigate the perspective of patients, clinicians, and other stakeholders (e.g. regulatory bodies, pharmaceutical industry) during all stages of the projects.
