Neural Signal Decoding for Prosthetic Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current decoding algorithms for neural-controlled prostheses are limited in providing real-time, graded control of multiple degrees of freedom, often relying on invasive methods and requiring complex computational power, while also being prone to noise and cross-talk between channels, which restricts the functionality of prosthetic limbs in amputees.
Innovation Solution
The development of a system that includes multiple single-channel decoders and a demixer, utilizing band-pass filters, nonlinear reshaping functions, and adaptive learning architectures to decode neural and muscle signals, enabling real-time estimation and control of motor intent with minimal noise and cross-talk, allowing for simultaneous control of multiple degrees of freedom in prosthetic devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If surface EMG recording is used for prosthetic control, then the decoding method is simple, but the control is limited to one degree of movement at a time and cannot provide appropriate information to control many lost degrees of freedom
Solution Approach 1:
The patent segments the neural control problem into multiple independent decoders, each handling a specific degree of freedom. Multiple decoders are trained separately on neural signals corresponding to different movement types, allowing each to specialize in decoding one aspect of motor intent while collectively providing comprehensive multi-DOF control capability.
Solution Approach 2:
The patent creates a universal decoding framework that can handle multiple degrees of freedom using the same basic decoder architecture. By training multiple instances of the decoder on different neural signal datasets corresponding to different movements, the system achieves versatile control across multiple prosthetic degrees of freedom without requiring fundamentally different decoding methods for each.
2Adaptability or versatility
If target muscle re-innervation (TMR) surgical procedure is used, then high level amputation can be addressed with surface EMG electrodes, but the surgical procedure is complicated and decoding algorithms become more complicated
Solution Approach 1:
The patent employs adaptive learning algorithms that automatically adjust decoding parameters based on the specific neural signal characteristics of each patient. The system performs self-calibration by learning the relationship between neural signals and intended movements during a training period, eliminating the need for complex manual decoding algorithm configuration and reducing the burden of complicated surgical procedures.
Solution Approach 2:
The patent changes the parameters of the decoding algorithm dynamically based on the patient's specific condition and neural signal characteristics. By adjusting decoding parameters adaptively rather than using fixed complex algorithms, the system can handle high-level amputations effectively while keeping the decoding process manageable and adaptable to individual patient needs.
3Measurement precision
If linear classifier is used for movement class identification, then the decoder can identify movement class, but the user is not able to control the degree by which to open or close hand or control how fast or slow the robotic arm moves
Solution Approach 1:
The patent transitions from static movement class classification to dynamic continuous control by using multiple decoders that output continuous control signals. Each decoder produces a graded output that varies continuously with the user's neural intent, enabling dynamic control of movement degree, speed, and other parameters rather than discrete movement classes.
Solution Approach 2:
The patent implements feedback mechanisms where the decoded control signals are continuously adjusted based on the relationship between neural activity and actual movement execution. The system uses feedback from the prosthesis state and neural signal patterns to modulate the control output, enabling graded control of movement parameters such as hand opening degree and arm movement speed.
4Quantity of substance
If multiple neural interface electrodes are used, then access to more neural signals is obtained, but noise and cross-talk between neural signals from multiple electrodes increases
Solution Approach 1:
The patent extracts relevant neural signal features from multiple electrode recordings while discarding noise and cross-talk components. By focusing on specific signal characteristics that correlate with intended movements and filtering out irrelevant variations, the system effectively separates useful neural information from harmful noise and cross-talk signals.
Solution Approach 2:
The patent introduces intermediate processing stages including signal filtering, feature extraction, and decoder training that act as mediators between the raw multi-electrode neural signals and the final control output. These intermediary processing steps clean and organize the neural signals, reducing noise and cross-talk while preserving the essential motor intent information.
Data Source
AI summary
Systems and methods for decoding neural and/or electromyographic signals are provided. A system can include at least one single channel decoder. Optionally, the system can include a demixer in operable communication with the single channel decoders. Each single channel decoder can include a filter to attenuate noise and sharpen spikes in the neural and/or electromyographic signals, a detection function to identify spikes, and a demodulator to get a real-time estimate of motor intent.