An Orchestra of Complexity

multicolored image of a brain

How Artificial Intelligence is transforming epilepsy and stroke treatment

By Alhaji Janneh

Intelligence has long been thought to reside exclusively in the human brain. However, that assumption is being challenged by the rise of artificial intelligence (AI), or the ability of computers to perform tasks that typically require human intelligence. Computers learn by analyzing patterns in data and applying what they learn to new data sets.

Researchers, such as Leonardo Bonilha, M.D., Ph.D., a professor in the Department of Neurology at MUSC, and resident now wonder whether AI could teach us something about the brain and its disorders. Specifically, Bonilha and his team are working to develop AI that can better predict epilepsy seizures and language recovery after stroke.

Currently, epilepsy medications only help to prevent seizures but don’t cure the disease. The only real cure is neurosurgery to remove the brain tissue from which the seizure activity originates, but it doesn’t work for all patients. Bonilha’s laboratory is interested to know whether AI could better predict which patients with epilepsy would benefit from surgery by learning from MRI and other imaging data of past patients.

In his laboratory, researchers such as neurology fellow Ezequiel Gleichgerrcht, M.D., Ph.D.,use an advanced form of MRI (diffusion tensor imaging) to map how information flows through the brain’s networks, also known as the connectome, and how that correlates to seizure activity. This sort of imaging produces large quantities of data, and Bonilha believes that the AI his team has created, a form of deep learning, can discover complex patterns in those data that humans wouldn’t necessarily be able to identify. These patterns could hold the key to predicting which patients are likely to benefit from neurosurgery.

“The connectome is complex just like an orchestra,” explained Bonilha. “If one instrument stops working well, it will disrupt the entire flow of the music. That’s why we’re using AI to better understand the complexity and make predictions for neurosurgical outcomes.”

After having been trained on a data set from past patients, the AI should be able to predict which new patients are most likely to benefit from neurosurgery. When fully developed and tested, this form of AI could lead to significant changes in epilepsy treatment.

In addition to improving epilepsy care, Bonilha’s team also uses deep learning to better target treatments for language recovery after stroke. Approximately 20% of all stroke patients experience language impairment or aphasia, which is the inability to process language, including language production and understanding.

As with epilepsy, Bonilha’s laboratory uses an advanced form of MRI to understand how stroke disrupts the brain’s networks related to speech production. With AI, they can also decipher which areas of the brain remain intact after stroke and could potentially aid with language recovery.

“For aphasia and language recovery, we don’t fully understand which part of the brain recovers and we can’t always predict who will do well with language recovery after stroke,”said Bonilha. “Therefore, we are hoping to use AI to predict which patients are more likely to recover their speech and benefit from speech therapy.”

Like any tool, however high-tech, AI has its limitations. Most current AI approaches require large amounts of data, and researchers are still figuring out the quantity needed for a good prediction.As AI advances, researchers will try to make accurate predictions with less data input, which will be especially significant for rare diseases such as epilepsy, for which fewer data are available.

For many indications,however, studies are showing that AI can equal or even surpass predictions by humans. But AI is not intended to replace physicians but rather to provide them more information on which to draw to make the best clinical decisions possible.

“The idea is for AI to make us better at treating diseases,”said Bonilha. “For example, AI can be great at landing an airplane, but I would also prefer to have a pilot there in case anything goes wrong.