New machine learning tool that examines brain MRIs

Good brain health is often a good indicator of chronological age. Current convolutional neural networks (CNNs) are able to accurately predict the age of a healthy patient from structural MRI scans of the brain.

Image Credit:

A team of researchers based at the School of Biomedical Engineering & Imaging Sciences, King’s College London, have created a new machine learning tool that looks at brain MRI scans and can predict brain age.

Indeed, the team has designed a screening tool that can automatically identify brains that appear older than chronological age in real time, using standard clinical MRI scans.

Research published in Neuroimaging demonstrates that the brain undergoes a loss of volume during the natural aging process. Since the volume loss is correct for the patient’s age, the new machine learning tool can accurately determine the patient’s age.

Detecting diseased brains

When a patient’s MRI shows a brain showing signs of disease – such as dementia – there is usually a disproportionate volume loss, and the majority of existing brain age models have been optimized for scans that do not are not part of a patient’s routine examination or are dependent on costly pre-treatment phases that limit in situ use.

So the King-based team set out to develop a framework that demonstrated the feasibility of being used as a screening tool in standard hospital exams to identify older-looking brains in real time.

This in turn could help improve the patient journey by expediting care (including intervention where appropriate) and potentially accelerating the development of disease-modifying drugs through better clinical trial enrollment.

Dr Thomas Booth, Lecturer in Neuroimaging at King’s College London

The team believes its tool can bridge the gap between analysis and expert panel review, if available, by automating these processes.

Using the machine learning-based Neuroradiology Report Classifier, the team was able to produce a large dataset containing 23,302 “radiologically for normal age” MRI results from two of London’s largest hospitals, Guy’s and St. Thomas’ NHS Foundation Trust and King’s College Hospital, using pre-existing neuroradiology reports.

The team then adopted a new approach. The scans were applied to an additional deep learning image algorithm with little computational pre-processing of the large data set they had acquired. Typical types of analysis obtained from a third institution were then introduced alongside an open source dataset.

Model testing

Booth and his team then began testing their model on scans that displayed disproportionate volume loss in the brain and assessed the brain age framework they had constructed through a series of heat maps that highlighted highlights the areas of brain volume loss.

Using such a large dataset, the team was able to appropriately train the machine learning model due to the model’s ability to generalize out-of-sample data.

Currently, older looking and abnormal brains are detected some time after analysis at the time of reporting. The most accurate reports will be in centers where there are neuroradiologists, but few centers have neuroradiologists.

Dr Thomas Booth, Lecturer in Neuroimaging at King’s College London

This indicates that the model could have significant effects on patient care and drug development, especially in small neurology centers where there is a lack of neuroradiology experts.

Automatic detection of volume loss in real time could pave the way for screening for common problems associated with neurodegeneration in scans obtained for a variety of reasons.

Subsequently, the diagnosis of, for example, early-stage dementia or Alzheimer’s disease has the potential to improve patient care by introducing early social and medical interventions. Patients could also be recruited into drug trials at a much earlier stage.

This framework could be used to leverage large hospital databases to offer invaluable new resources for testing, clinical validation, and training of medical image analysis tools beyond data-based applications. brain age.

References and further reading

Wood, D., Kafiabadi, S., et al., (2022) Accurate brain age patterns for routine clinical MRI examinations. NeuroImage, [online] 249, p.118871. Available at: (2022) New screening tool developed to automatically identify older-looking brains typical of dementia. [online] Available at:

Disclaimer: The views expressed herein are those of the author expressed privately and do not necessarily represent the views of Limited T/A AZoNetwork, the owner and operator of this website. This disclaimer forms part of the terms of use of this website.

Sherry J. Basler