Approved Research

Fair Learning for Large MRI databases

Lay summary

Brain imaging has been a great tool for neuroscience and medical research (psychiatry, neurology). The size of brain imaging databases has grown significantly in the past 15 years, reaching tens of thousands of scans. This has been very beneficial to neuroscience and clinical research. However, as the number of scans grows, data analysis becomes harder, with many different types of scans, and results that can be affected by factors such as age, gender, or the scanner used to acquire the data. This can lead to bias results, that are sometimes hard to compare between studies and difficult to generalize to the entire population.

In this 3-year project, in order to solve this problem, we are developing machine learning algorithms that can consider possible sources of bias and produce results that are not altered by these biases. We are using the very large collection of brain scans from the UK Biobank database in order to develop, test and train these algorithms. By doing this, we hope to help research in neuroscience, neurology, or psychiatry to produce better results.