Neural Interface Calibration Using Sparse Tensor Decomposition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current direct neural interface systems face challenges with high prediction error and computational cost due to the inefficiency of existing multi-way analysis methods, particularly in the calibration step, which limits their performance in generating accurate command signals for external devices and is not suitable for practical, self-paced applications in real-life environments.
Innovation Solution
A method that involves acquiring electrophysiological signals as an N+1-way tensor and applying a penalized Alternating Least Squares algorithm to determine sparse weight vectors, maximizing covariance between the score and output vectors, thereby reducing computational cost and improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional BCI systems use a limited number of features extracted from EEG or ECoG signals to generate command signals, then the system complexity is reduced, but the prediction accuracy and information exploitation efficiency deteriorate
Solution Approach 1:
The patent transforms the feature extraction approach by changing from traditional spectral amplitude features to time-frequency representation using wavelet transform. This parameter change in signal representation enables better exploitation of neural information while maintaining computational feasibility through the use of tensor decomposition methods.
Solution Approach 2:
The patent introduces a new dimension by organizing signal data into tensor structures that capture spatial, temporal, and spectral information simultaneously. This multi-dimensional representation allows the system to exploit more information from the signals without proportionally increasing system complexity, as the tensor decomposition efficiently processes the high-dimensional data.
2Quantity of substance
If PARAFAC decomposition is applied to EEG signal tensor before classification, then the dominant signal components are retained, but event-related components with low amplitude are lost and non event-related information is retained
Solution Approach 1:
The patent incorporates feedback by using the output classification results to guide the tensor decomposition process. The decomposition is performed in a way that prioritizes components relevant to the classification task, allowing the system to iteratively refine which signal components are retained based on their discriminative value for the intended classification.
Solution Approach 2:
The patent modifies the decomposition approach by changing from standard PARAFAC to a classification-oriented decomposition that considers the relevance of components for discrimination. This parameter change in the decomposition methodology ensures that event-related components, even with low amplitude, are preserved while non-event-related information is filtered out.
3Ease of operation
If a cue-paced approach is used where subjects wait for external cues, then the system operation is simplified and synchronized, but the productivity and naturalness of interaction in real-life environments deteriorate
Solution Approach 1:
The patent implements self-service by enabling the BCI system to automatically detect and respond to user intentions without requiring external cues or synchronized timing. The system autonomously identifies relevant neural patterns and generates commands, allowing users to interact naturally in real-life environments without waiting for system prompts or maintaining strict timing protocols.
4Productivity
If self-paced BCI systems are implemented without external stimuli, then the productivity and naturalness of interaction improve, but the prediction error increases due to lack of synchronization and concentrated attention
Solution Approach 1:
The patent applies preliminary action by pre-processing the neural signals through wavelet transform and organizing them into tensor structures before classification. This preliminary organization of data in a way that highlights relevant patterns enables the system to maintain high prediction accuracy even without external synchronization cues, as the structural preparation of the data compensates for the lack of temporal alignment.
Data Source
Figure 1~3
Figure 4~5
Figure 6
AI summary
A method of calibrating a direct neural interface system comprising the steps of: a. Acquiring electrophysiological signals electrophysiological signals representative of a neuronal activity of a subject's brain over a plurality of observation time windows and representing them in the form of a N+1-way tensor (X), N being greater or equal to two, called an observation tensor; b. Acquiring data indicative of a voluntary action performed by said subject during each of said observation time windows, and organizing them in a vector or tensor (y), called an output vector or tensor; and c. Determining a (multi-way) regression function of said output vector or tensor on said observation tensor; wherein said step c. includes performing multilinear decomposition of said observation tensor on a "score" vector (t), having a dimension equal to the number of said observation time windows, and N "weight" vectors (w1, w2, w3), characterized in that said "weights" vectors are chosen such as to maximize the covariance between said "score" vector and said output vector or tensor subject to a sparsity-promoting constraint or penalty. A method of operating a direct neural interface system for interfacing a subject's brain (B) to an external device (ED), said method comprising such a calibration step.