Machine learning to improve multimodal biomarkers in Frontotemporal Dementia

If someone told you that you would definitely get dementia, and potentially at a young age, what would you want to know next? Many people might ask "When?" and "Is there anything that can be done?"

Frontotemporal dementia (FTD) is an uncommon disease that often strikes at a much younger age than Alzheimer's disease. One out of every three cases is caused by a specific faulty gene. If a parent carries this faulty gene, then there's a 50/50 chance it will be passed on to a child.

Currently, there is no reliable way to determine when an individual carrying this faulty gene will begin to experience symptoms. In research, we estimate this based on the age when their parent or other family members developed symptoms, but this is not very accurate1. Some people at-risk will choose to find out if they carry a faulty gene.

For those who do, having a better understanding of how and when the disease will develop could be helpful for themselves and their families.

Understanding when someone could start to experience symptoms would also be important for treatment. While there are currently no approved treatments to slow the disease there are promising treatments in development. These drugs may provide more benefit if they are given before individuals have symptoms, as they could address the underlying cause of the disease before long-term, irreversible damage is caused.

However, before these drugs can be prescribed, they first must be tested in controlled clinical trials to see if they are safe and effective.

A central challenge in all trials is to select the right markers to catch the changes produced by the treatment to determine its effectiveness. In rare diseases, it is an additional challenge to enrol enough participants to accurately detect these changes. This difficulty is amplified when we want to enrol presymptomatic participants but do not know when a person will develop symptoms.

This proposal addresses these issues. I will use data collected as part of the Genetic Frontotemporal dementia Initiative (GENFI). This large multi-centre study follows individuals from families known to have a gene for FTD, most of them without symptoms.

Of the participants without symptoms, roughly half of the participants will carry the gene and develop FTD, while the other half are non-carriers who act as control group. GENFI collects many types of data that measure various aspects of the disease, including blood, cerebrospinal fluid (the fluid that surrounds the brain), brain images and clinical/cognitive tests.

I will use machine learning techniques to combine these data and create new measures to understand the disease. These measures will provide an estimate of when someone is likely to develop symptoms.

The brain often operates in networks that span different brain regions. Many forms of dementia interfere with these networks. I will develop a signature of the brain networks that are disrupted in FTD, called the network failure quotient.

I will determine how brain imaging changes over time can be detected in each individual. I will also build up a picture of the many different processes which are occurring that ultimately cause someone to have symptoms. With these new tools, we hope the next time we tell someone they will definitely get dementia, we can tell them when, and we can offer them something to do about it.