Photoplethysmography (PPG) uses optical measurements to determine the volumetric changes of body fluids, for example blood. This non-invasive method is used both to aid diagnosis and the ongoing monitoring of various diseases, such as high blood pressure or diabetes.
In Europe, it is expected that approximately 38 million people will suffer from diabetes by 2030, making the development of affordable medical devices essential to guarantee high quality healthcare.
PPG measuring devices are inexpensive and portable, allowing for long-term monitoring of patients and extensive data collection. Machine learning (ML) methodologies are used to extract the key information from these large data sets.
However, currently there are no standard practice guidelines regarding the processing of PPG signals, and therefore it is unclear how the uncertainty of ML algorithms will affect the generated diagnostic and therapeutic output.

This project aims to publish new benchmark data sets covering a range of diagnostic problems and develop new methods for quantifying the uncertainty of various ML methodologies. The optimisation of the metrological infrastructure will help develop new standardised procedures for manipulating PPG data sets, which the project will publish in a good practice guide.
Additionally, an open-source software repository will be made available to allow health care providers across Europe to use ML methodologies for PPG data sets.

Developing new standardised methods for using ML algorithms in the data analysis of PPG signals can help European healthcare to have access to affordable diagnostic tools and long-term monitoring devices for diseases.