Summary: Researchers at the University of Pennsylvania apply cryptographic methods to decode motor neuron activity and predict limb movements.
Source: University of Pennsylvania.
Breaking the German Enigma cipher was a pivotal achievement in World War II. By combining clues from espionage with analytical and computational techniques, codebreakers discovered rules that converted scrambled characters into readable German, producing intelligence that saved lives and shortened the war.
Inspired by that history, a team of researchers from the University of Pennsylvania, the Georgia Institute of Technology, and Northwestern University has developed a cryptography-inspired method to decode activity in motor neurons. Using only general knowledge about typical arm movements, their approach can infer the direction in which rhesus macaques move their arms from recorded brain activity.
The same cryptanalytic idea could ultimately be applied to decode more complex muscle activation patterns for advanced prosthetic control or, potentially, to help restore communication for people with severe paralysis by decoding the neural patterns associated with speech and expression.
The study was led by Konrad Kording, a Penn Integrates Knowledge Professor with appointments in the Department of Neuroscience at Penn’s Perelman School of Medicine and the Department of Bioengineering in the School of Engineering and Applied Science, together with Eva Dyer, who was a postdoctoral researcher in Kording’s lab and is now an assistant professor in the Department of Biomedical Engineering at Georgia Tech and Emory University. The team also collaborated with Lee Miller’s laboratory in the Department of Physiology at Northwestern University.
The researchers reported their findings in Nature Biomedical Engineering.
In their experiments, the team recorded electrical activity from several hundred neurons related to arm movement in three rhesus macaques. Implanted electrodes captured spike trains as the monkeys performed reaching tasks: a target would appear at different positions around a central starting point, and the monkeys moved their arms to reach it. The recorded neuronal spikes were tied to the movements but were not accompanied by explicit pairing of neural patterns with labeled movements for training.
Conventional brain-computer interfaces (BCIs) typically rely on supervised learning. In that setup, neural firing patterns are paired with known movements during a training phase, so that a decoder learns to map neural activity to motor outputs and then reconstruct movements from ongoing signals.
By contrast, the cryptography-based approach avoids supervised training. As Kording explains, “In cryptography, ‘supervised learning’ would be called a ‘known plaintext attack’—you have both the encrypted and unencrypted message and only need to deduce the transformation. In this study, we aimed to decode brain activity using a movement model from the encrypted signal alone.”
The team began with the raw firing patterns of individual neurons and sought a consistent mathematical mapping from those patterns to the monkeys’ actual arm movements. The key insight was to exploit statistical regularities in natural movements. Just as frequency patterns and distributional properties of letters and vowels helped codebreakers crack Enigma, statistical structure in movement trajectories provides constraints that make decoding possible without explicit supervised labels.
“The algorithm tests many possible decoders until it produces outputs that resemble typical, biologically plausible movements,” Kording said. “Scaling the method remains a challenging computational problem, but our results demonstrate proof of concept that cryptanalysis techniques can be applied to neural decoding.”
This unsupervised, cryptanalysis-style decoding is particularly promising for BCIs that control prosthetic limbs. Removing or reducing the need for long calibration sessions would make prosthetic systems easier and faster to use. Calibration can be especially problematic when the user cannot perform or imagine actual limb movements—for example, when a limb is amputated rather than paralyzed—because imagined movements do not always produce the same neural signatures as executed movements.
Beyond prosthetics, the researchers envision applications for people who are “locked-in.” “A patient could be asked to produce neural patterns associated with specific concepts or words,” Kording noted. “Instead of training a decoder for every possible word in a large vocabulary, we could transform the patient’s neural patterns until the outputs match the statistical structure of language.”
Looking ahead, the researchers expect continued improvements in neural recording technology to increase the usefulness of their technique. Work toward electrode arrays capable of sampling hundreds of thousands to a million neurons simultaneously would provide richer datasets and improve decoding performance.
Kording also emphasized ethical considerations: “As recording capabilities grow, this technique becomes more powerful. We must be careful about potential misuse and the possibility of applying such tools without consent. What intelligence agencies could do with this technology is concerning.”
Although still in early stages, the team believes a cryptanalysis approach matches the neural decoding problem well. Natural selection shaped neural representations in living brains, and those representations display structure and redundancies that make unsupervised decoding feasible—similar to the “mistakes” or regularities that allowed Enigma to be cracked.
Funding: This work was supported by the National Institute of Neurological Disorders and Stroke through grants R01 NS053603 and R01 NS074044.
Source: University of Pennsylvania—news release prepared by Evan Lerner.
Publisher: Organized by NeuroscienceNews.com.
Image credit: Researchers.
Original research: “A cryptography-based approach for movement decoding” by Eva L. Dyer, Mohammad Gheshlaghi Azar, Matthew G. Perich, Hugo L. Fernandes, Stephanie Naufel, Lee E. Miller, and Konrad P. Körding, published in Nature Biomedical Engineering (published online December 12, 2017).
University of Pennsylvania. “The Enigmatic Brain: Using WWII Code-Breaking Techniques to Interpret Brain Data.” NeuroscienceNews, December 13, 2017.
Abstract
A cryptography-based approach for movement decoding
This study demonstrates a method inspired by cryptanalysis that decodes motor-related neural activity without supervised labels by leveraging statistical patterns in natural movements. The approach maps neuronal firing patterns to plausible movement trajectories, offering a proof of concept for unsupervised neural decoding with potential applications in brain-computer interfaces and assistive technologies.