Dr. Feng Liu is currently an assistant professor in the Department of Systems Engineering here at Stevens, and his latest project is making and developing AI models that help “decode” the brain with the help of multiple other already existing medical tools, such as an electroencephalogram (EEG) and different types of magnetic resonance imaging (MRI). His project helps explain how the brain works in explicit detail, as well as how other things like addiction and Alzheimer’s affect the brain and disrupt it.
His time at Harvard Medical School taught him that math can be used to serve clinical needs, and he still uses that in his current work. He has worked with teams at other locations in the U.S., like Rutgers or Georgia Tech, and teams internationally in places such as Canada, Germany, Taiwan, and Australia for projects. The Stevens motto “Inspired by humanity, powered by technology,” inspires him and aligns with the projects he does.
He got the idea for his most recent project while doing his Ph.D., when he started working with brain imaging data, such as an EEG and functional magnetic resonance imaging (fMRI), and then he started “decoding” brain states under anesthesia using those tools. He then realized that he could use AI to explain and interpret those brain waves to be used for medical research and needs. Pairing EEG with fMRI data allowed Liu and his lab to non-invasively map internal brain activity compared to surgically invasive measuring techniques. His lab utilizes Graph Neural Networks (GNN) to decode how brain regions communicate and how those pathways are changed because of disease. Graph neural networks turn scientific information, such as neural pathways and activity, into graphs to process, interpret, or model them for researchers to use in their studies and projects.
His lab also uses large language models (LLMs), like EpiSemoLLM, to interpret the patients’ descriptions of seizure symptoms to help medical professionals treat epileptogenic zones, which are zones in the brain that tend to induce epilepsy. Liu’s team builds models that can help find where seizures start and help treat patients who don’t respond well to medication. When testing an LLM’s ability to interpret seizure symptoms from patients’ descriptions, its performance was similar to that of human experts, showing its efficiency and reliability. He is also expanding into the territory of sleep disorders, such as sleep apnea, to help determine severity, making diagnoses more accessible for everyone. In terms of working with addiction, Liu is trying to use LLMs to be able to effectively predict and prevent relapse events and identify their brain state at that point.
Liu’s system provides a novel medium to study the electrical properties of biological cells and tissue through “3D visualization of the electrophysiological source.” This form of imaging shows doctors and medical providers where the issue is, so they know how to properly treat it with other tools they have, whether that be medication or surgery. These tools allow doctors to be able to medically diagnose a range of brain disorders in a non-invasive way, making the medical industry significantly safer and more accessible.
