Machine Learning tools for Healthcare and Sports Performance Applications

A recent CAMERA collaboration workshop revealed the opportunity for Dr Campbell’s CAMERA funded work on Bayesian non-parametric modelling (such as that with the Foundry on rotoscoping tools for the Creative Industries) to be applied for statistical analysis in a number of healthcare applications. Market examples of novel computational tools available to solve modelling problems include the TensorFlow toolkit from Google. Whilst innovative, frameworks such as TensorFlow are limited as they are designed for machine learning (ML) experts with a requirement that all models and assumptions be expressed indirectly through formal programming languages.

Organisations and individuals working in healthcare related sectors that are interested in the analysis of human motion data have a demonstrable need to translate the findings of their observations into statistical models that can then be used to test theories, diagnoses and treatment recommendations, but lack the ML skills to do so. Presently the data collected by healthcare researchers and practitioners have to be passed on to machine learning experts for development into statistical models for analysis, introducing both a time cost and possibility of implementation errors due to a lack of domain knowledge by the programmer and a lack of computer science knowledge by the domain expert.

This project will build a software tool allowing health researchers to analyse their data using state-of-the-art Bayesian methods without having to learn formal programming languages or operate through the bottleneck of a software programmer.

The project will be a collaboration between the Departments of Computer Science and Health, and will involve engagement with a working group of external organisations including the Department of Clinical Neurosciences at the University of Cambridge, Smart Acoustics Ltd and the Royal United Hospital for Rheumatic Diseases (RNHRD). The design of the software tool will be directly influenced by their requirements in the first instance. We would then like to generalise to ensure the software is sufficiently flexible to help with analysis in a whole range of fields, including more commercially focussed sectors providing revenue generation opportunities from licensing, such as the Creative Industries.

The project presents an opportunity to embed University of Bath research into the pipeline of public and private sector healthcare providers and creative industries post production pipelines, as part of a substantial and wide ranging impact case study.