New research out of Carnegie Mellon University has introduced a new AI-based technology that could better aid in understanding how the human brain works.
There are multiple technologies currently available for imaging brain function and dysfunction. While these have proved helpful to differing degrees, each one has its downsides. This new technology is set to improve in some areas that the existing ones have come rather short.
The innovative AI-based imaging approach brings speed and precision. It promises the capability to map out the brain’s fast-changing electrical activity with awesome speed and in high resolution. The technology also looks to enable low-cost dynamic brain imaging.
This research was conducted by the group of Professor Bin He, who has been working for decades to perfect non-invasive dynamic brain imaging tech. It appeared in the journal Proceedings of the National Academy of Sciences (PNAS).
Brain imaging techniques
According to scientists, electrical activity is spread out across the three-dimensional brain and is fast changing. Efforts have for long been made to map out this activity or connections with limited success.
There are many brain imaging tools currently available to cognitive neuroscientists for mapping – noninvasively – the brain’s structure and functions. Notable among these are electroencephalography (EEG), magnetic resonance imaging (MRI), and magnetoencephalography.
Each of these has its benefits and downsides. In the case of MRI, for example, it is not sufficiently fast to image brain dynamics. And while EEG does better than MRI, its spatial resolution has proven less than ideal.
EEG source imaging, however, tends to be less expensive and faster, compared to most other techniques. But it requires special training and skills to execute perfectly.
Scientists have also explored the option of electrophysiological source imaging. This involves translating scalp EEG recordings back to the brain. Signal processing and machine learning are then used to produce dynamic images of activity in the brain over time.
Dynamic imaging with AI
This new work pioneers a novel AI-assisted imaging technology capable of capturing brain dynamics with accuracy and speed. It is a culmination of decades-long work by He.
“Our group is the first to reach the goal of introducing AI and multi-scale brain models,” said He, who is a biomedical engineering professor at Carnegie Mellon University. “Using biophysically inspired neural networks, we are innovating this deep learning approach to train a neural network that can precisely translate scalp EEG signals back to neural circuit activity in the brain without human intervention.”
He’s group put the new technology to work in 20 healthy persons. The technique was used to image the sensory and cognitive brain responses of these subjects.
The team went further to also validate the approach in the identification of epileptogenic tissue in 20 people with drug-resistant epilepsy. Results from AI-based noninvasive imaging were compared to invasive measurements and outcomes of surgical resection.
In terms of accuracy and computational efficiency, the AI-based method did better than traditional source imaging methods.
“With this new approach, you only need a centralized location to perform brain modeling and training deep neural network,” He said. “After collecting data in a clinical or research setting, clinicians and researchers could remotely submit the data to the centralized well trained deep neural networks and quickly receive accurate analysis results.”
The researchers are now planning to carry out larger clinical trials. The results of those trials will determine if and how soon the new technology will be available for clinical use.