Earlier diagnosis of atypical Parkinsonian conditions from MRI using machine learning

Disease area(s):	Neuroscience; Multiple system Atrophy, Corticobasal syndrome, Progressive Supranuclear Palsy
Data sources:	University Hospitals Plymouth NHS Trust (UHPNT); Parkinson’s Progression Markers Initiative (PPMI);PROSPECT-M study
Project stage:	Data collation
Ethical approval:	Granted (Reference Number: 21/PR/0918)
Principal Investigator:	Stephen Mullin
Lead Researcher:	Megan Courtman
Funder(s):	Rotary Club Holsworthy;
Reference:

Summary

The atypical Parkinsonian syndromes Progressive Supranuclear Palsy, Multiple System Atrophy and Corticobasal Syndrome are rapidly progressive neurological conditions with a poor prognosis. Initially they may present almost identically to Parkinson’s disease. A number of promising treatments to slow down or halt the progression of these diseases are currently being tested in drug trials. Using Artificial Intelligence, we aimed to train a model to identify features on MRI brain scans which may allow earlier diagnosis of the three conditions.

A dataset of routinely collected imaging was collated. The imaging for atypical Parkinsonian syndromes comprised a very small number of scans, which we aimed to augment using data from a national research library. However, the MRI protocols were not compatible with our routinely collected dataset. Machine learning models were trained to differentiate the small number of cases from control/Parkinson’s disease scans. Various methods of dealing with imbalanced data were trialled; however, the number of cases present was unfortunately too small to build an effective machine learning model.