UCL Centre for Medical Image Computing


Tech developed at CMIC is behind the discovery of new pathological variants of Alzheimer's disease

24 June 2021

CMIC researchers have played a key role in a recent study that leveraged the power of computer science and big data to redefine Alzheimer’s disease.

Alzheimer’s disease is a devastating age-related disease that slowly dissolves brain tissue over decades, resulting in loss of mental capacity and eventual death. Symptoms typically appear after age 65, with the most common being memory loss. However, a variety of symptoms have been linked with the disease, including affected visual processing (your eyes can see, but your brain doesn’t register) and executive dysfunction (inability to perform complex tasks), often with relatively unaffected memory. Such clinical variants are well known, but the causes of this variation are poorly understood.

Previous studies have used autopsy or neuroimaging within clinical variants to show that the hallmark pathology of Alzheimer’s disease — tau — can deviate from the stereotypical spatial pattern. These studies are limited to end-stage or late-stage disease. Ignoring disease progression in this way confuses spatial variation in tau spread with temporal severity of tau spread. Disentangling these to understand both spatial and temporal aspects of Alzheimer’s disease pathology has only recently been made possible through a combination of data-driven disease progression modelling technology developed at CMIC, and the availability of large neuroimaging datasets.

CMIC researchers in the UCL Progression Of Neurodegenerative Disease group (POND) develop computational models of disease progression, such as the SuStaIn algorithm (Young at al., Nature Communications 2018), that leverage known disease characteristics together with unsupervised machine learning techniques to identify and quantify heterogeneous disease patterns. SuStaIn is a key part of the suite of medical image computing software developed at CMIC and is quickly gathering traction with researchers around the world.

Together with lead author Jacob Vogel from McGill University in Canada, CMIC researchers applied SuStaIn to the largest ever assembled dataset of tau-PET images from five centres to reveal four spatiotemporal subtypes of pathology spread in Alzheimer’s disease. Each subtype was found in each centre, and the subtypes were validated both longitudinally and in a separate dataset that used a different tau-PET radiotracer. The four data-driven subtypes had different symptom profiles and also varied in other ways, including age.

The concept of “typical Alzheimer’s disease” may be a thing of the past.

Lund University

Further reading: (Vogel et al., Nature Medicine 2021)