Written by Niall Bourke (Research Assistant, Imperial College London)
Edited by Lucia M. Li
Researchers are constantly exploring new ways to try and understand the effects of traumatic brain injury on the brain. Dr James Cole used Machine Learning methods to study this question. Machine Learning is when computers are used to look for patterns in large amounts of data. The computer then derives rules based on these patterns, and then applies these rules to make predictions about other data.
A similar pattern recognition technique is used on websites such as Facebook or Google – these websites analyse your internet activity and personal information and suggest what friends to connect with or adverts to be displayed to you. For example, if many of your Facebook friends are from a particular workplace, it will suggest more people from that workplace to connect with. Or, if you often visit websites to do with pets, you will see more adverts about related things, like pet insurance.
We know that brain volume changes over time with normal aging – our brains shrink as we get older, such that older people have brains of smaller volume than compared with younger people. Some diseases, such as chronic alcoholism, can make the brain shrink faster than normal. This would mean that their ‘Brain Age’ was higher than their chronological age. A TBI may also influence the normal process of an aging brain. There may be a one off change from the injury – slightly older (stable additive effects), or this could potentially result in an accelerated processes (interactive effects) - see Figure.
There were two phases to the study.
The first phase was training the computer to identify patterns from MRI scans of 1,537 healthy participants of various ages. Each participant’s brain volume was calculated to give estimated normal brain volumes for each chronological age.
The second phase was the testing phase. Here, the computer was given new data (MRI scans) and it applied the rules from phase 1 to make predictions. In this case, the computer was asked to calculate the age of the participant’s brain based on his/her brain volumes, the ‘Brain Age’. Calculations were made for 99 TBI patients with cognitive problems, and 113 healthy controls. The ‘Brain Age’ generated by the computer could then be compared to a participant’s actual age to calculate a ‘predicted age difference’
For healthy participants, the ‘Brain Age’ calculated by the computer matched the actual chronological age very well. When this method was applied to the TBI patients it was found that the predicted ‘Brain Age’ was higher than the chronological age. On average, TBI patients were estimated to have a ‘Brain Age’ that was 4 years older than their actual age. This difference was related to the severity of the injury. Patients with mild injuries did not differ significantly from healthy controls, whereas there was a significant difference for patients who had a more severe injury. The cause of injury did not appear to influence how much a participant’s ‘Brain Age’ and actual age differed.
Also, in patients whose ‘Brain Age’ was different to their actual age, the bigger this difference, the worse their cognitive function, particularly for memory and processing speed.
Discussion &cautionary notes
This is the first study to show how TBI may lead to an acceleration of the brain shrinkage that is seen in normal ageing. This highlights some of the chronic issues often seen following an injury. It is currently unclear the exact processes involved that lead to this problem. This could be explored with follow up studies that track patients and cognitive changes and brain volumes over time.
Take home points
Using a large enough sample to train a computer to recognise patterns, it is possible to make predications when new data is provided. Using this method, Dr Cole and colleagues showed that TBI can result in a difference between actual age and ‘Brain Age’, and that this difference is related to cognitive problems. In the future, this measure could potentially be developed to become a clinical measure to assess the effectiveness of treatments.