Few challenges of developing brain-machine interfaces

Jean-Charles Nigretto
3 min readJan 25, 2021

--

Today I’ve read a paper about the main challenges when developing brain-machine interfaces. The paper’s source can be found at the bottom of this article.

Yesterday I’ve read a white paper written by Elon Musk and describing how its company Neuralink was handling the development of a brain-machine interface. Willing to continue learning about this topic, today I’ve read another paper about brain-machine interfaces. This time, the authors focus on the physiological properties of brain-machine interface input signals and spend some time describing the challenges of designing brain-machine interfaces.

Brain-machine interface (or BMI) is a field that focuses on direct interactions between the brain and external devices. That being said, BMI seems more centered around interactions from the brain (using brain data as an input) to external devices. Applications could include controlling robotic prostheses or limbs. To do so, some captors (mainly electrodes) must be able to capture brain “signals”, understand them and send back the translated movement to a machine. These different steps pose a lot of challenges that are partially described in today’s paper. I will here talk about challenges related to brain signals.

The brain constantly receives, processes, and sends information. This activity can be analyzed through different measures like looking at how much and how hard neurons fire spikes or how does the blood flow in our brain change over time. The most easily available signals are electrical signals, mainly from neurons action potentials (spikes). These action potentials could start from a single neuron and quickly trigger consecutive action potentials from surrounding neurons. Because of that, understanding the meaning of neuron’s action potentials require to get a good spatial and temporal resolution of neurons spikes. Unfortunately, this electrical information is highly attenuated by our cerebrospinal fluid, our skull, and our scalp. These limitations require recording electrodes to be very close (potentially inside) our brain in order to accurately map neurons spikes.

When searching for a specific brain signal (for example, primary motor cortex single spikes have been associated with arm movements), many more challenges appear. First, the signals of interest must be robust, which means that it must persist as long as we will use our device. If it appears that the signals we’ve identified only last for minutes only, this means our designed device won’t be that useful. This characteristic is described as the signal longevity in the article.

Another crucial feature is signal stability: we must be able to accurately receive and process the signal even when it changes. The more frequent changes occur, the most stable the signal has to be in order to be effectively used. An unstable signal will require frequent device calibrations and may greatly impact signal analysis accuracy.

After we’ve identified signals with high enough longevity and stability, we know need to see how strong it is (its amplitude). The amount of relevant signal we get is directly linked to the number of degrees of liberties our device can handle: the stronger the signals, the more things we can do from them.

The last challenge this article will cover is related to the digital sampling frequency. The more data we want to analyze and the more power we need. This may not be an issue if we're able to analyze data externally but as explained above, electrical signals rapidly intensity through fluids and tissues which means that we are often working very close to the brain. This proximity poses limitations in terms of how much heating we can generate, how much we can deteriorate the battery of our receivers and how much power we can use. This precise issue was also well documented in yesterday’s paper from Elon Musk which insisted on the difficulty of performing live neuron action potential detection with both high accuracy and low power consumption.

Source: Slutzky, M.W. and Flint, R.D., 2017. Physiological properties of brain-machine interface input signals. Journal of neurophysiology, 118(2), pp.1329–1343.

--

--