Study Shows that ML Algorithm Accurately Diagnoses Parkinsonian Syndromes

Researchers at the Francis Crick Institute, in collaboration with UCL Queen Square Institute of Neurology and tech company Faculty AI, have harnessed machine learning to accurately predict Parkinson’s disease subtypes using patient-derived stem cell images. Published in Nature Machine Intelligence, their breakthrough revealed that computer models can classify four subtypes with an impressive 95% accuracy, offering promising prospects for personalized medicine and precise drug development.

Parkinson’s disease, a neurodegenerative ailment affecting mobility and cognition, exhibits varying symptoms and progressions among individuals due to diverse underlying mechanisms. Until now, the lack of a reliable subtype differentiation method has resulted in generic diagnoses and limited access to tailored treatments and support.

Parkinson’s disease primarily involves protein misfolding and impaired mitochondrial clearance, with some cases linked to genetic mutations. To create a ‘human model of brain disease in a dish,’ researchers transformed patients’ cells into stem cells and induced four distinct Parkinson’s subtypes. High-resolution imaging, including the labeling of key cell components like lysosomes, was employed. A computer program was trained to recognize each subtype and predict them even with unseen images.

Mitochondria and lysosomes emerged as crucial indicators for subtype classification, affirming their roles in the disease’s development. Other aspects of cell imagery also contributed, though not yet fully understood.

James Evans, a Crick and UCL PhD student and co-first author alongside Karishma D’Sa and Gurvir Virdi, stated that advanced image techniques and AI enabled the evaluation of a greater number of cell features, extracting more information than conventional methods. The team aims to expand this approach to uncover the contributions of cellular mechanisms to other Parkinson’s subtypes.

Sonia Gandhi, assistant research director and group leader of the Neurodegeneration Biology Laboratory at the Crick, emphasized the challenge of identifying mechanisms in living patients, hindering precise treatment. Their algorithm, built on patient-derived neuron models and extensive imagery, offers hope for pinpointing disease subtypes during a patient’s lifetime. Furthermore, it could enable drug testing in stem cell models, predicting a patient’s likely response before clinical trials, potentially revolutionizing personalized medicine.

This project took shape during pandemic disruptions, with the team learning coding skills in Python, which they now apply to ongoing projects. James Fleming, Chief Information Officer at the Crick, highlighted the success of this collaboration with Faculty AI, illustrating the potential of AI to drive new insights and investments into AI and software engineering.

The team’s next goals involve understanding disease subtypes in individuals with different genetic mutations and exploring the classification of sporadic Parkinson’s cases without genetic markers.

Author: Neurologica