Approved Research

A generative modeling for normative aging brain MRIs and neurological abnormality detection

Lay summary

Human brain changes with normal aging and incidence of brain diseases also rise with age. In order to provide aging people with medical treatments at the right timing, expert routine inspection of brain structure and function in medical images, which is an important proxy to examine brain changes with increasing age, is important for determining if the changes are normal or not. However, this expert inspection can be very time-consuming and expensive. There have been researchers' efforts to automate this manual inspection of brain MR images using computer algorithms, but it has been hampered by high computing demand and limited amount of brain MR image datasets which are crucial for developing accurate automated computer algorithms. Due to the recent advent of super-computers and big public brain MRI datasets such as the UK Biobank data, developing artificial intelligence (AI) based automated tools for examining and creating brain MR images is being rapidly accelerated.

In this study, we propose developing computer algorithms that can automatically generate a healthy brain medical image of a subject, in which its brain appearance is normal for the subject's specific age. The computer algorithm will automatically create a normal-for-age healthy brain medical image when the subject's demographic and clinical information and real brain medical image are given to the computer algorithm. Then the difference between the AI-generated healthy brain image and the real brain image will be automatically analyzed to help physicians decide if the subject's brain condition is healthy or abnormal for the specific age. This appearance gap between real brain MRI image and computer-generated healthy brain MRI image will serve as a medical feature indicating potential risk of having brain diseases and the severity of them. The new AI based computer algorithm would enable to discover new medical indicators that signify potential risk of a subject to develop further aging-related brain diseases.

We will validate the new computer algorithm with both healthy subjects and patients with brain disease to determine if there is any meaningful relation between the brain appearance gap and medical symptoms in patients with brain diseases. The outcome of the study will help improving the accuracy of routine brain MR exams and reducing the associated expert time and cost, which would be beneficial for managing aging population more effectively. The project will be conducted over multiple years as the UK Biobank dataset continues including more new subjects.