By directly modulating abnormal activities, neural implants have the potential to help treat brain disorders like Parkinson’s disease and epilepsy. Xilin Liu, a researcher at the University of Toronto, is working with microelectronics and artificial intelligence to make this emerging technology safer and more intelligent.
The method that is employed by Liu’s team to incorporate neural implants into small silicon chips is the same method that is used to fabricate chips that are used in modern smartphones and computers. They were able to reduce the device’s physical dimensions as well as its power consumption with the help of this technology, which is known as CMOS, which stands for complementary metal-oxide semiconductor. This allowed them to reduce the risks associated with the initial surgical procedure as well as the implant’s use over the long term.
Liu is a member of the CRANIA neurotechnology centre, which is a joint venture between the University of Toronto and the University Health Network. This centre brings together neuroscientists, data and material scientists, and clinicians with electrical and computer engineers to develop new neurotechnology. Together, they investigate strategies to enhance brain health and map out alternative therapy options, particularly for individuals who do not react favourably to the pharmaceuticals that are now available.
In a project that was completed not too long ago, Liu and his colleagues attempted to make use of the potential of AI in order to enhance the clinical efficacy of the implants and reduce the harmful consequences of excessive stimulation.
The group turned to a form of artificial intelligence known as deep learning (DL), which consists of algorithms that, after they have been taught, are able to extract deep-level information when presented with novel data. These models outperformed standard algorithms when it came to determining the best timing, and they showed to be especially effective when it came to locating hidden biomarkers, which are frequently disregarded by conventional methods.
However, because to the high computational cost of deep learning models, it can be difficult to integrate them, which is especially problematic when taking into consideration how important it is for all processing to execute locally within the implants.
Liu and his team devised methods for training and optimising the models for each patient’s condition in order to cut down on the amount of computing work that was required. A recent case study shown that deep learning can identify epileptic seizures in low-power neural implants in a manner that is equivalent to state-of-the-art algorithms that are executed on high-performance computers. This work was eventually published in the Journal of Neural Engineering in the year 2021.
Liu notes that up to one billion people around the world suffer from a variety of brain illnesses, thus he believes that the technology that his team has developed can be utilised in a wide variety of clinical applications in addition to treating epilepsy.
Graduate students who sign up for Liu’s brand-new neuromodulation class, which will be offered for the first time this coming fall, are going to learn about the importance of being collaborative and keeping an open mind.
Chronic pain, depression, and dementia are some of the conditions that will be targeted in the future. Already, Liu is considering the ways in which neuromodulation treatments can be of assistance to Alzheimer’s disease patients.
Edge deep learning for neural implants: a case study of seizure detection and prediction, Journal of Neural Engineering (2022).