Mini brain machine interface for real time neural processing
EPFL Unveils Miniaturized Brain-Machine Interface
Introduction
Breakthrough Technology
EPFL's research team has unveiled a state-of-the-art miniaturized brain-machine interface, enabling direct brain-to-text communication on diminutive silicon chips.
Overview of Brain-Machine Interfaces (BMIs)
Historical Context
Brain-Machine Interfaces (BMIs) are increasingly recognized as a viable means of restoring communication and control for those with severe motor impairments. Historically, these systems have been cumbersome, energy-intensive, and constrained in their practical use.
New Development
Scientists at EPFL have created the first high-performance Miniaturized Brain-Machine Interface (MiBMI), presenting a remarkable small, low-power, highly accurate, and versatile technology.
Technological Advancements
Publication and Presentation
Published in the most recent IEEE Journal of Solid-State Circuits and showcased at the International Solid-State Circuits Conference, the MiBMI significantly improves the efficiency and scalability of brain-machine interfaces, paving the way for fully implantable devices.
Potential Impact
This technology has the potential to greatly enhance the quality of life for patients suffering from conditions like amyotrophic lateral sclerosis (ALS) and spinal cord injuries.
Design and Practicality
The MiBMI's compact design and low power consumption are pivotal, making it ideal for implantable applications. Its minimal invasiveness enhances safety and practicality for clinical and real-world use.
Integration and Innovation
This fully integrated system, where recording and processing occur on two ultra-compact chips totaling 8mm², represents the latest innovation in low-power BMI devices developed at Mahsa Shoaran's Integrated Neurotechnologies Laboratory (INL) within EPFL's IEM and Neuro X institutes.
Functionality and Performance
Translation of Neural Signals
"MiBMI enables the translation of complex neural signals into readable text with exceptional accuracy and minimal power usage. This development moves us closer to practical, implantable solutions that could greatly improve communication for those with severe motor impairments," says Shoaran.
Brain-to-Text Conversion
Brain-to-Text conversion entails interpreting neural signals produced when an individual envisions writing letters or words. This process uses electrodes implanted in the brain to capture neural activity related the motor functions of handwriting.
Real-Time Processing
The MiBMI chipset processes these neural signals in real time, converting the intended hand movements from the brain into digital text. This technology enables individuals, particularly those with locked-in syndrome and severe motor impairments, to communicate by merely thinking about writing, with the interface translating their thoughts into readable text displayed on a screen.
Achievements and Capabilities
Current Performance
"Although the chip has yet to be integrated into a functional BMI system, it has demonstrated its capability by processing data from earlier live recordings, including those from the Shenoy lab at Stanford, and achieving 91% accuracy in converting handwriting activity into text," says lead author Mohammed Ali Shaeri.
Character Decoding
With the ability to decode up to 31 characters, the chip surpasses other integrated systems. "We believe we can decode up to 100 characters, but we are still awaiting a handwriting dataset with a larger variety," says Shaeri.
Data Processing
Current brain-machine interfaces collect data from electrodes implanted in the brain and then send these signals to an external computer for decoding. The MiBMI chip, however, performs both recording and real-time processing of data, featuring a 192-channel neural recording system and a 512-channel neural decoder.
Future Prospects
Neurotechnological Innovation
This neurotechnological advancement exemplifies remarkable miniaturization, integrating specialized knowledge in circuits, neural engineering, and AI. It is particularly significant in the rapidly growing neurotech startup sector within the BMI space, where miniaturization and integration are central. EPFL's MiBMI offers promising prospects for the future of the field.
New Data Analysis Strategy
Researchers had to develop a new data analysis strategy to mange the large volume of information form the miniaturized BMI's electrodes. They discovered that each letter's brain activity, as visualized by the patient, contians unique markers called distinctive neural codes (DNCs).
Efficiency of Processing
The microchip processes only around a hundred bytes of distinctive neural codes (DNCs) per letter, rather than the thousands of bytes typically required. This streamlining results in a fast, accurate system with low power consumption and shorter training periods, simplifying BMI usage and making it more accessible.
Collaborations and Future Research
Ongoing Collaborations
Collaborations with team at EPFL's Neuro-X and IEM Institutes, including Gregoire Courtine, Silvestro Micera, Stephanie Lacour, and David Atienza, are paving the way for the next generation of integrated BMI sustems. Shoaran, Shaeri, and their team are also investigating additional applications for the MiBMI system beyond handwriting recogniation.
Expanding Applications
"We are working with various research teams to evaluate the system across different applications, including speech decoding and movement control. Our aim is to create a flexible BMI that can be adapted to a range of neurological conditions, offering diverse solutions for patients," states Shoaran.
Labels: brain machine interface, brain-to-text