System, method and apparatus for detecting an evoked response signal
Inventors
Kurtz, Isaac • Steinman, Aaron • ROWLANDS, Stephen Allan
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-10327664-B2
Publication Date
2019-06-25
Expiration Date
Abstract
A method for detection of an evoked response signal in noise including: generating a plurality of stimuli; receiving a noisy signal related to an evoked response to the plurality of stimuli; divide the noisy signal into a plurality of responses to the plurality of stimuli; estimate a statistic matrix for the plurality of responses; shrink the statistic matrix; calculate weights based on an inverse of the shrunk statistic matrix; apply weights to the plurality of responses to construct a final response; and output the final response. An apparatus having an input device configured to receive data related to a plurality of stimuli; and a processor configured to: receive a noisy signal and divide the noisy signal into a plurality of responses; estimate a statistic matrix; shrink the statistic matrix; calculate weights based on an inverse of the shrunk statistic matrix; and apply weights to the plurality of responses to construct a final response.
Core Innovation
The invention relates to detection of an evoked response signal in noise by generating a plurality of stimuli and receiving a noisy signal related to the evoked response signal to the plurality of stimuli. The noisy signal is divided into a plurality of responses to the plurality of stimuli, weights are calculated for the plurality of responses, and sets of responses are identified. For each identified set, responses are combined as a revised response, a new weight is calculated for the revised response, and the responses are removed from the plurality of responses.
A final response is constructed by weight averaging the revised responses and responses not identified in a set. The final response represents the evoked response or is indicative of whether or not an evoked response has been detected. An apparatus and a system implement the same core processor functionality using an input device or stimulus generator, respectively, and output the final response.
The invention optionally includes hierarchical processing in which each response is decomposed into sub-responses, and the set-based shrinking, weighting, combining, removing, and final construction operations are applied across sub-response sets. Dependent refinements specify identification of response sets by calculating weights based on an estimated statistic matrix and a shrinking process, including shrinkage logic using a shrinkage list and a mask constructed from diagonal elements and non-diagonal elements.
The dependent refinements further include wavelet decomposition to determine a frequency band and support scale-by-scale processing using the sub-responses.
Claims Coverage
The patent contains three independent claims: a method claim, an apparatus claim, and a system claim, each covering an evoked-response detection workflow that iteratively constructs revised responses from selected response sets and forms a final evoked-response estimate via weight averaging. Dependent claims add statistic-matrix-based identification and shrinkage using mask-based constraints and optional hierarchical sub-response decomposition with wavelet-based frequency band processing.
Iterative revised response construction by weight averaging
Generates a plurality of stimuli, receives a noisy signal related to an evoked response signal, divides the noisy signal into a plurality of responses to the plurality of stimuli, calculates weights for the plurality of responses, identifies sets of responses, combines each identified set as a revised response, calculates a new weight for the revised response, removes the responses from the plurality of responses, and constructs a final response by weight averaging the revised responses and responses not identified in a set for output as the evoked response or indicative of detection.
Statistic-matrix based identification and shrinking of response sets
Identifies set(s) of responses using estimation of a statistic matrix, selects correlated response sets whose combination reduces noise, and shrinks the statistic matrix by removing or combining correlated response sets to form a shrunk statistic matrix.
Mask-based shrinkage with diagonal construction and zeroing non-selected non-diagonal elements
Constructs a mask for the statistic matrix by forming a diagonal matrix of the same dimension as the statistic matrix and setting non-diagonal elements corresponding to responses not included in a shrinkage list to 0, thereby producing and returning a shrunk statistic matrix.
Inverse shrunk statistic matrix weights
Calculates new weights using an inverse of the shrunk statistic matrix.
Hierarchical sub-response decomposition with repeated set-based processing
Decomposes each response into multiple sub-responses with subsets of information, performs the calculation, identification, combination, removing, and construction operations for each set of sub-responses, and combines the final responses from each set of sub-responses.
Wavelet decomposition for frequency-band processing
Determines a frequency band using wavelet decomposition for sub-response grouping and scale-by-scale processing.
Across the independent claims, the core inventive subject matter is iterative identification of response sets, formation of revised responses with updated weights, and construction of a final evoked-response estimate via weight averaging. Dependent inventive features further narrow the detection by introducing statistic-matrix-based shrinking with mask construction, computing weights from the inverse of the shrunk statistic matrix, and optionally applying hierarchical sub-response decomposition with wavelet decomposition.
Stated Advantages
Reduced powerline noise artifacts (e.g., 60 Hz) in detection results.
Reduced alpha-band EEG noise artifacts in detection results.
Improved statistical distinction of MMN in awake versus anesthesia conditions.
Documented Applications
Detection of auditory brainstem response (ABR), including Wave V detection timing described in the document’s comparative results.
Detection of mismatch negativity (MMN), including awake versus anesthesia comparisons and a MMN time window described in the document’s comparative results.
Detection/detection-like processing for auditory steady state responses (ASSR) as part of the document’s entity set referenced in the summary.
Infant hearing screening and auditory thresholds are mentioned as context in the document.
Interested in licensing this patent?