Princeton University neuroscientists joined forces with Intel computer scientists to map the human mind in real time, developing the next generation in brain imaging analysis.
In a lab at the Princeton Neuroscience Institute, test subjects look at pictures, watch movies and listen to The Moth Radio Hour as scientists track their brain activity via functional magnetic resonance imaging (fMRI).
The researchers’ goal: to read their subjects’ minds in real time, as they are thinking, feeling and reacting to the stimuli.
This task would have been impossible just a few years ago. Reading a single scan was time consuming — reading a multitude of scans meant blowing out a system’s storage and processing capabilities.
New technologies are changing that.
Researchers at Princeton University and Intel Labs have developed software that enables cognitive neuroscientists to map the mind in real time. The software decodes neural data and reveals how brain activity affects learning, memory and other cognitive functions.
“The research that we’re doing uses fMRI human brain imaging to view what’s going on inside the brain as people think and feel,” said Jonathan Cohen, a neuroscience professor and co-director of the Princeton Neuroscience Institute.
Monitoring the brain in real time could improve the diagnosis and treatment of brain disorders, Cohen said. The scans show images of the brain based on changes in blood oxygen level, which indicate mental activity in different regions of the brain.
While neuroscientists want to learn more about the human brain, Intel computer scientists want to use the brain research to design more efficient software algorithms to process large amounts of data.
“We hope to apply what we learn from these studies of human cognition to machine learning and artificial intelligence,” said Ted Willke, senior principal engineer at Intel Labs.
“This may lead to safer autonomous driving, quicker drug discovery and earlier detection of cancer.”
Decoding Brain-Loads of Data
Prior to this collaboration, researchers spent days analyzing fMRI scans.
They translated the brain’s neural activity and blood flow in each voxel (a pixel-sized cube) into thoughts and emotions, based on sensory stimuli.
The goal was to create a mind map to represent different cognitive states and processes. But the process was painstakingly slow, as researchers worked through one variable at a time. A typical scan contains up to 1 million voxels.
“When you’re talking about neural activity, you want to understand it rapidly while someone is watching a movie or listening to a story or completing a task,” said Willke.
The researchers wanted not just to speed up the process of analyzing a single brain scan, they wanted to analyze brain-loads of data in real time.
“We need all the computational power that we can get, so the faster that we can decode people’s thoughts and return feedback to them, the more effective the neurofeedback is going to be,” said Ken Norman, a psychology professor at the Princeton Neuroscience Institute.
Computer Cluster as Therapy
Willke and the Intel Mind’s Eye team are now exploring other ways technology could be used in neuroscience, including neurocognitive therapy for mental illnesses.
Cognitive neuroscience deals with the study of the prefrontal cortex where cognition happens. This is where ideation, creativity and logical reasoning live.
“When that prefrontal cortex goes wrong, it leads to things like depression, anxiety disorders, PTSD, psychopathy, bipolar disorder and schizophrenia,” Willke said.
Willke and his colleagues believe that by understanding the specifics of what’s going wrong in the prefrontal cortex, better treatments can be developed.
For instance, as the database of brain scans grows, scientists will be become more familiar with normal and abnormal function. They will also be able to observe the effect of cognitive training on cognitive function, offering valuable feedback on what works and what doesn’t.
In addition, the technology itself may offer new therapeutic solutions.
“The kind of thing that we’ll eventually be able to use this for is to better diagnose and eventually, hopefully, treat psychiatric disorders,” said Cohen.
Not only can the new software be programmed to analyze brain function as subjects react to stimuli, but it also can present new therapeutic stimuli — images or other media — to shift brain function from unhealthy to healthier patterns.
“It might just be that the next $10 billion drug is a computer cluster,” said Willke, referring to the processing power that makes such real-time diagnosis and treatment an option.
Already, there is evidence that such neurocognitive therapy could be successful.
In an initial feasibility study, Princeton researchers exposed subjects to a set of overlaid images of neutral scenes and sad faces. Participants were asked to pay attention to the scenes. Previous research by Princeton and others had already established that depressed people tended to focus on negative attractors such as images of people screaming or crying, or photos of disasters such as oil spills.
The subjects were then provided with real-time feedback as they viewed the images. If they got distracted by the sad faces, the software decoded this from their brain scans and made the task more difficult — the proportion of scene stimulus was reduced by fading it out. As the subjects continued to receive feedback, they began to shift their focus toward the less negative images, even after the reinforcement ended.
In fact, the positive shift appeared to last for weeks after the initial treatment in the study.
“We think that this may be the first computer-based neurocognitive therapy for depression, PTSD and other mental illnesses,” said Willke.
Key to Unlocking Drug Discovery and Beyond
The Princeton-Intel collaborative has open-sourced their Brain Imaging Analysis Kit (BrainIAK) to help other researchers around the globe process fMRI data.
“Other researchers will be able to benefit from our advances and make advances of their own,” said Nicholas Turk-Browne, a professor of psychology at Princeton.
The Mind’s Eye team hopes to apply findings from their studies to machine learning and AI in other industries, including pharmaceutical research and development, Willke said.
Using a process called virtual screening, researchers look for drugs that work as “keys” to a specific biological “lock.” Chemical databases already contain millions of keys, but supercomputing allows researchers to sift through these massive databases to find a shortlist of drugs most likely to fit the locks.
Speeding up this process could result in higher quality drugs and less costly drug failures, Willke said.
Additionally, this research may lead to further development of autonomous cars.
A better understanding of human cognition, including how the brain solves problems, may lead to improvements in the AI that enables self-driving cars to solve problems and make decisions about the environment in which they operate.
The Mind’s Eye team will continue to peer into the human mind, finding new ways to apply machine learning and AI to mental health, autonomous driving and beyond.