Kernel, a start-up claiming to offer “Neuroscience as a Service” has been in the news lately because of their recently announced $53M Series C fund raising. What hasn’t been as widely covered is what their services are and how they work.
Fortunately Kernel has published a fairly extensive breakdown of their research and the resulting products they plan to bring to market. Kernel is focused on measuring signals in the brain to better understand the brain and advance neuroscience more quickly, eventually hoping to be able to read and write cognition. Traditional tools for measuring brain activity like functional Magnetic Resonance Imaging fMRI and Electroencephalography (EEG) are costly, bulky and difficult to use in real-life situations that would produce more meaningful data.
To solve this problem, Kernel has developed two different technologies for measuring brain activity it calls Flow and Flux, which it hopes will scale our understanding of the human brain, leading to advances in brain health and overall brain performance to name just a few potential benefits. Kernel believes that using both technologies in conjunction will result in even faster advances in how brain activity is measured.
Kernel Flow is a technology for measuring hemodynamics, that is the flow of blood, in the brain. This is similar to an FMRI, only without needing the patient to lie perfectly still in a room-sized machine. Flow by contrast is a wearable headset which allows the wearer to engage in normal activities, and the data is processed in a connected computer (currently tethered via USB, though wireless data transfer certainly seems feasible given the relative transfer rates of USB and WiFi standards).
Kernel Flux uses advanced devices that measure magnetic fields called optically pumped magnetometers or OPMs, while still allowing the subject to move their head normally and comfortably. The flux system is also able to distinguish background magnetic fields not coming from the subject (such as the magnetic fields created by power lines or other parts of the human body) and filter them out, resulting in a higher signal to noise ratio.
Kernel detailed two experiments to demonstrate the fidelity of its neural activity measurements, “Speller” and “Speak ID.” Both experiments described are reproductions of previously published experiments using sensors that seems to have been the precursors to its Kernel Flux technology.
In the Speller experiment, subjects were shown 30, eight-letter words (240 total letters), and asked to spell each word after it was displayed by looking at a visual keyboard. Using 20 OPMs, placed on the subjects head, the researchers were able to measure the brain waves of the participants, then run them through a statistical method known as canonical correlation analysis (CCA). With this analysis, the researchers were able to correctly predict which letter the subject was looking at an average of 80% of the time. In subsequent experiments, they were able to nearly reproduce the performance of 20 sensors using only 4 sensors.
The Sound ID experiment used 16 OPM sensors again placed on the heads of the subjects, while the subject selected 60 second snippets of recorded speech. Again using CCA analysis on the real-time data fed from the sensors the researchers were able to determine which snippet the subject was listening to with 100% accuracy in most cases in under 20 seconds. The experiment was also run using snippets of songs instead of speech, and while the results were less accurate and predictions took longer, they were able to predict 8 out of 10 songs correctly in some subjects.
Kernel has promised to share results from their Flow and Flux hardware soon. Until then, we’ll be wondering what new insights might be gleaned from these novel ways of collecting data directly from the brain.