Deep Learning Predicts Neurodegenerative Disorders from Brain Scans
What if I told you that a computer could help warn you about potential diseases you risk facing, simply from images of your brain alone? Well, that is exactly true – a recent study shows that a new deep neural network model can accurately predict the brain age and propensity to certain neurodegenerative diseases based on deviations from healthy brain age trajectories.
Traditionally brain scans have been used by doctors to determine brain age. Since deviations from normal brain aging usually manifest in abnormal brain development, doctors tried to use brain age as a biomarker for certain diseases such as Alzheimer’s, Type 2 Diabetes, stroke, and other seizure disorders. However, they often struggled to do this accurately because it required the combination and correlation of multiple points of brain structure and function to make accurate predictions.
Deep learning has emerged as a powerful tool in medical image analysis, allowing the modeling of complex relationships from raw brain scans and expression of correlations with major disease families with utmost efficiency and accuracy.
“While clinicians can only grossly estimate or quantify the age of a patient based on their EEG [more on that later], this study shows an artificial intelligence model can predict a patient’s age with high precision,” said lead author Yoav Nygate, senior AI engineer at EnsoData, the company that produced the deep learning algorithm.
When put to the test, researchers at the American Academy of Sleep Medicine (AASM) found that the deep learning algorithm was extremely accurate with a margin of error of only 4.6 years.
Let’s break down the basics of the study and the science behind it -
Brain Age
To start off, a person’s “brain age” can be described as their cognitive function, or level of brain strength, written into tangible, numerical data. This is different from your regular, or chronological age because, while it differs from person to person, certain aspects of your brain become stronger as you age. An example of one such aspect is abstract thinking or language and problem-solving abilities.
Brain age is typically calculated from MRI scans or EEG, or electroencephalography. EEG is a monitoring method aimed at recording electronic activity in your head. This type of monitoring is often referred to as an “electrophysical” monitoring, as scientists look for changes in electrical signals in a patient’s brain. Thousands upon thousands of data points are created as the electrodes placed on a patient’s head record the synapses, or electrical messages being sent between neurons, and graph them into a large paper.
This brain age is then used to find something called a “Brain Age Index,” (BAI), which is a numerical representation of the deviation from normal brain growth. The study here aimed at seeing how low or high BAI’s corresponded with diseases such as epilepsy, sleep disorders, and more.
To do this, members of EnsoData took these EEG readings and created a deep learning model that has learned highly generalizable neuroimaging features to accurately identify brain age and other biomarkers. This is a relatively recent development in neuroimaging because deep learning models require large amounts of data to build reliable self-learning error correction models and varied data was not available till recently.
But wait, what exactly is deep learning?
Deep Learning and Neural Networks
Deep learning is a specialized subset of machine learning (ML) which, in turn, is a subset of artificial intelligence (AI) field. Artificial intelligence is the development of computer systems able to perform tasks that normally require human intelligence. Within AI, ML gives computers the ability to learn from recognizing data patterns without being explicitly programmed.
A basic example of a ML algorithm is linear regression. For example, imagine you want to predict your income after higher education. After a software programmer defines an equation and inputs hundreds of points of training data of people’s years of education and their associated income, the algorithm can construct an equation of a line. It can then use that to predict your income given your years of higher education. In short, a linear regression ML algorithm can construct a line of best fit by giving random data.
Deep learning takes this a step further and is a more sophisticated evolution of ML. Deep learning algorithms analyze data with a logic structure like how a human would draw conclusions. The underpinning of deep learning is a layered structure of algorithms called a neural network. The design of a neural network is inspired by the biological neural network of the human brain, enabling a highly nuanced process of learning that’s far more advanced than standard machine learning models.
Deep Learning vs Machine Learning
The main difference of deep learning from ML is how the deep learning model can self-correct its results by automatically selecting appropriate aspects of the input data without being explicitly fed the input.
In the previous example, a programmer would have to manually enter the two “features,” or data categories (years of higher education and income), but with a neural network, the computer itself finds these or other features after being trained with a correction function and massive amount of data. For example, when Tesla cars try to identify a stop sign, they use deep learning. Basic ML models can identify relevant properties of the stop sign after being fed millions of images and can learn which images are what after being manually fed what they all have in common – they are all red, have 8 straight edges, etc. In contrast, neural networks are capable of automatic feature engineering, in which they can self-identify what features, or attributes, are important to identify a stop sign.
Artificial Neural Network gives back to the human brain that inspired it
Coming back to brain age, EnsoData created a new deep learning model using the raw EEG files to predict the brain age and at-risk disease profiles. The neural network developed had several neurons per layer and several layers to triangulate various combinations of multiple data points, like cortical thickness, surface curvature, volume of gray matter etc, from thousands of EEG images. The different layers of the neural network were programmed to learn about each attribute. Interestingly, taken individually, none of these features could predict the brain age but that is where the power of deep learning comes into play. Not only that, but the model was also trained with nearly 126,241 sleep studies, and was then tested on nearly 8 thousand studies, making the predictions extremely accurate and far more capable than what neuroscientists have been able to do traditionally.
The study found that the deep learning model showed a clear relationship between brain age and many disorders including epilepsy, seizures, stroke, disoriented breathing, and low sleeping efficiency. But what’s more is that the study found that patients with diabetes, depressive syndromes, or high levels of fatigue also presented an elevated BAI compared to the healthy population.
In other words, the AI deviation model presented here could clearly indicate specific neurodegenerative diseases based on brain age calculated via a simple brain scan. What is notable here is that such a simple and precise indicator for neurodegenerative diseases has not existed so far and the introduction of AI is rapidly going to change that.
A Biomarker for Neurodegenerative Diseases like Blood Pressure for Stroke
Nygate sums it up best when he says: “The results in this study provide initial evidence for the potential of utilizing AI to assess the brain age of a patient. Our hope is that with continued investigation, research, and clinical studies, a brain age index will one day become a diagnostic biomarker of brain health, much like high blood pressure is for risks of stroke and other cardiovascular disorders.”
This study is one of many which shows how AI can significantly help improve the lives of people. It also shows how deep learning algorithms of AI, modeled after the neural network of the human brain is now ‘giving back’ by helping further healthcare endeavors in all its aspects.
Sources and other References:
https://neurosciencenews.com/ai-eeg-brain-age-18597/
https://www.med.upenn.edu/cbica/brain-age.html
Image Credit: https://icdn.digitaltrends.com/image/digitaltrends/eeg-headset.jpg