Summary: By using artificial neural networks to analyze neuroimaging data, researchers are able to accurately determine biological age.
source: Max Planck Society
The biological age of a person can be accurately determined through brain images using the latest artificial intelligence techniques, the so-called artificial neural networks.
Until now, it was not clear what traits these networks use to infer age. Researchers at the Max Planck Institute for Cognitive and Brain Sciences have now developed an algorithm that reveals that age estimation is due to a whole range of features in the brain, providing general information about a person’s health status.
The algorithm could thus help detect tumors or Alzheimer’s disease more quickly and allow conclusions to be drawn about the neurological consequences of diseases such as diabetes.
Deep neural networks are an artificial intelligence technology that is already enriching our daily lives on many levels: The artificial networks, modeled after the way real neurons work, can understand and translate language, interpret texts, and recognize objects and people in images. But they can also determine a person’s age based on an MRI scan of their brain.
True, it will be easier to find out the age by asking a person. However, determining the age of the machine also gives you an idea of what a healthy brain looks like at different stages of life.
If the network estimates the biological age of the brain based on the scan to be higher than it really is, this may indicate a possible disease or injury.
Previous studies, for example, have found that the brains of people with certain diseases, such as diabetes or severe cognitive impairment, appear to have more years under their belts than they actually do. In other words, the brains were in worse biological condition than one might assume based on the age of these people.
Although artificial neural networks can accurately determine biological age, until now it was not known what information from the brain visualizes their algorithms to do so. Scientists in the field of AI research also refer to this as the “black box problem”.
According to this, you push an image of the brain into the model, the “black box”, and let it process it – and in the end you only get its answer. However, due to the complexity of the networks, it was previously not clear how to generate this response.
An algorithm to interpret the results of artificial intelligence
So scientists at the Max Planck Institute for Human Cognitive and Brain Sciences in Leipzig wanted to unlock the black box: what does the model look at to arrive at its findings, brain age? To do this, they worked with the Fraunhofer Institute for Communication in Berlin to develop a new interpretation algorithm that can be used to analyze age estimates of networks.
Simon M., a Max Planck Institute candidate and first author of the basic study, which has now appeared in the journal, explains NeuroImage.
“We can now identify regions and characteristics of the brain that indicate a higher or lower biological age.”
This showed that artificial neural networks use, among other things, white matter to make predictions. Accordingly, they specifically look at the number of small cracks and scars that pass through the nerve tissue in the brain. They also analyze the width of the grooves in the cerebral cortex or the size of the cavities, the so-called ventricles.
Previous studies have shown that the older a person gets, the larger the grooves and ventricles, on average. The interesting thing is that artificial neural networks arrived at these results on their own – without giving them this information. During the training phase, all they had were brain scans and a person’s real life years.
“Of course, the increased age estimation can also be interpreted as an error in the model,” said Veronica Witt, leader of the research group. “But we were able to show that these deviations are biologically significant.”
For example, researchers have confirmed that diabetics have increased brain life. They were able to show that patients had more lesions in the white matter.
The future role in medical diagnosis
It is already clear that artificial neural networks will play an increasingly important role in medical diagnosis. Knowing what these algorithms inform will become increasingly important: in the future, brain scans can be automatically analyzed by different networks, each specialized in specific areas — one draws conclusions about Alzheimer’s disease, another about tumors, and another about possible psychiatric disorders.
“Then the doctor not only receives notes that certain diseases may be present. Hoffman also explains that she sees which areas of the brain underlie the diagnosis.
Corresponding features are directly marked in the MRI image by the algorithms in each case, and thus can be more easily detected by medical professionals – who in turn can draw immediate conclusions about the severity of the disease.
It will also be easier to detect misdiagnoses: if the analysis is based on biologically implausible areas, such as errors made when creating the image, the doctor can immediately detect them. Thus, the research team’s interpretation algorithm can also help improve the accuracy of the artificial neural networks themselves.
In a follow-up study, the researchers now want to investigate in more detail why their models also look at brain features that have so far played little role in aging research — for example, neural networks also focus on the cerebellum. How the aging processes progress there in healthy and sick people has been a mystery to scientists.
About this AI and biological age research news
original search: open access.
“Towards the interpretability of deep learning models for multimodal neuroimaging: finding structural changes in the aging brainWritten by Simon M. Hoffman et al. NeuroImage
Towards the interpretability of deep learning models for multimodal neuroimaging: finding structural changes in the aging brain
Brain age (BA) estimates based on deep learning are increasingly used as a neuroimaging biomarker of brain health; However, the basic neurological features remained unclear.
We combined sets of convolutional neural networks with layer-wise related spread (LRP) to discover brain features that contribute to BA.
They were trained on magnetic resonance imaging (MRI) data for a population study (n = 2637, 18–82 years), our models accurately estimated age based on single and multiple modalities, regionally restricted images and whole-brain images (mean absolute errors 3.37–3.86 years).
We found that BA estimates capture aging in both small and large-volume changes, revealing gross enlargements of the ventricles and subarachnoid spaces, as well as white matter lesions, and atrophy seen throughout the brain. The difference from expected aging reflected cardiovascular risk factors and the acceleration of aging was most pronounced in the frontal lobe.
Applying LRP, our study demonstrates how superior deep learning models detect brain aging in healthy and at-risk individuals throughout adulthood.