�鶹��

The rumbling and banging of the MRI scanner in the ears of a person undergoing a scan is a sign that the invisible will soon become visible. As is well known, MRI scans give doctors and researchers a unique view inside the brain, and with advanced analytical tools they extract information from the scans and interpret it. But sometimes the most important details are missed. A new brain atlas is intended to remedy this.

The new atlas is called NextBrain and has just been presented by an international team of researchers in an article in Nature.

The new atlas is a digital 3D map of the human brain, based on photographs and analyses of brain tissue from deceased individuals. The model represents a plausible depiction of what a healthy average brain looks like in detail. The atlas is intended to be used as a spatial reference for MRI scans.

Research Fellow Oula Puonti from the Danish Research Centre for Magnetic Resonance (DRCMR) at Amager and Hvidovre Hospital is supported by the Lundbeck Foundation, and he has contributed to the development of the atlas. He has also helped to develop a method that makes the use of NextBrain simpler and faster.

He explains how the new atlas differs from others:

“Unlike other comparable models, NextBrain covers the entire brain, and we can carry out analyses of brain structures at a much finer level of detail. For example, previously we have largely only been able to work with the overall structure of the hippocampus, which is our memory centre, but now we can include several substructures. And we can analyse changes in both structure and function, depending on the type of MRI scans we are dealing with.”

The atlas takes account of both biological variation and the many different image formats and qualities that doctors and researchers work with. This makes the model nuanced and statistically robust.

The development of NextBrain was led by Associate Professor of Radiology Juan Eugenio Iglesias from Massachusetts General Hospital in Boston.

Tested on brains with Alzheimer’s and signs of ageing

Five cerebral hemispheres from deceased individuals form the foundation of the new brain atlas, which is intended to help researchers carry out more in-depth, precise and statistically robust analyses of MRI brain scans.

The five hemispheres were cut into around 10,000 slices, which were prepared and placed between two glass plates. They were then photographed under a microscope, and using probability theory and machine learning, the images were assembled into a 3D model representing an average adult brain. The brain was subsequently divided into 333 regions that are of particular relevance for doctors and researchers to study.

Based on the model, a purpose-built computer program can automatically perform a segmentation, which adapts the MRI scans to the atlas coordinate system and the 333 regions. This allows structures to be compared and enables researchers to measure and investigate their anatomy and function.

The researchers behind NextBrain tested their tools on, among other things, a dataset of just under 400 brain scans from people with and without Alzheimer’s disease.

Using the new tool, it was possible to distinguish with 90.3 per cent accuracy between brains with and without changes resulting from Alzheimer’s disease. This is 4–5 per cent better than the most widely used standard tools.

In another test, high-quality MRI scans of the brains of 705 living individuals aged between 36 and 90 were compared with NextBrain. Using the atlas and its associated analytical tools, the researchers were able to demonstrate in detail how the volume of the various substructures gradually changes as we age.

Increasing the brain atlas’ speed from days to 20 minutes

Since the launch of NextBrain, Oula Puonti has helped to speed up segmentation using the atlas, as the original model was heavy and cumbersome to work with.

“Because there are so many 3D data points, the atlas requires substantial computing power, and it took a long time to run analyses using the original method. So we set out to make it easier and faster to work with,” he explains.

The solution was a combination of different technologies and methods which together make the tool faster and even more accurate. Among other things, an artificial intelligence system was trained to identify contrasts at lightning speed, enabling it to efficiently sort differences in the imaging data.

To increase speed, the researchers also used GPUs, which are processors designed to handle large datasets. The process moved from stepwise segmentation to segmentation in a single run, and the results speak for themselves. For example, a segmentation that previously took a week can now be completed in 90 minutes. Another that used to take two or three days is now ready in 20 minutes without a GPU and in under five minutes with a GPU.

“We can now analyse large volumes of brain scans in a much shorter time, and with a higher level of detail and depth than before. This could have major significance for neuroscience,” says Oula Puonti.

Although NextBrain is primarily intended for research, Oula Puonti also sees great potential in using the atlas for diagnosis and treatment.

“Because the atlas includes all the substructures, it can hopefully be used to investigate many different diseases,” he says.

“A dream scenario would be for it to be used to identify more sensitive biomarkers for diseases such as Alzheimer’s and Parkinson’s. That would allow diagnoses to be made earlier and enable better monitoring of disease progression, as well as evaluation of how different treatments work.”

The atlas is open source and freely available, so anyone keen to explore the brain in greater detail can simply get started.