DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Responsive to communications on 05/05/2026
Claims 1-2, 4-6, 9-16, and 18-20 amended
Claims 3, 7-8 and 17 original
Claims 1-20 pending
Claims 1-4, 6-18, and 20 are rejected
Claims 5 and 19 are objected to
Final Action
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Interview
Examiner thanks applicant for interview conducted on March 11, 2026. Examiner confirms discussions in interview pertains to 101, 102, and 112 topics with interview summary record attached in the application file.
Response to Arguments
Response to 112(b)
Regarding the rejections of claims 4-8 and 18-20 the claims have been amended to overcome the rejection. Examiner confirms amendments to the claim overcome the previous 112(b) rejection and the rejection is withdrawn.
Response to 101
Applicant argues 101 rejection of claims 1-13 and 15-20 is rendered moot by amendments to the claims.
1.Issue 1: Applicant argues that amended independent claim 1 recites operations which are performed by processing circuitry through layers of an electrophysiological signal classification model. Applicant argues that this is a model which is based on training signals collected by the acquisition device. Applicant argues that these features cannot be categorized as corresponding to a mental process.
Rule: Whether the claimed features can be categorizes as corresponding to a mental process is a Step 2A prong one inquiry. The MPEP 2106.04 (ii)(1) states “In Prong One examiners evaluate whether the claim recites a judicial exception, i.e. whether a law of nature, natural phenomenon, or abstract idea is set forth or described in the claim.” The MPEP 2106.04(a) outlines examples of abstract ideas to be “Mathematical concepts – mathematical relationships, mathematical formulas or equations, mathematical calculations, and Mental processes – concepts performed in the human mind (including an observation, evaluation, judgment, opinion)”
Analysis: In Step 2A prong 1 the examiner considers whether the claims recite a judicial exception, with some examples being mathematical concepts or mental processes. The examiner believes that certain limitations in the claim, such as “extracting by processing circuitry through a temporal convolution layer of an electrophysiological signal classification model a time feature based on the multi-channel electrophysiological signal” constitutes actions which can reasonably be performed in the mind. The action being performed is “extracting … a time feature,” which is interpreted by the examiner as encompassing an observation done by a user of a signal. The operations being performed by processing circuity which relates a signal classification model must then be considered under Step 2A prong 2, and is not relevant towards analysis conducted as to whether or not the claim recites a mental process.
Conclusion: The examiner does not find the argument regarding step 2A prong 1 to be convincing, please see below for considerations under step 2A prong 2.
2. Issue 2: Applicant argues that even if claim 1 recites judicial exceptions, that the amended features of “extracting … “ and “generating …” improves the computational performance and accuracy of determining classifications information based on multi-channel electrophysiological signals. Applicant also cites specification where this improvement is discussed.
Rule : The “2024 Guidance Update on Patent Subject Matter Eligibility, Including on Artificial Intelligence” states An important consideration in determining whether a claim improves technology is the extent to which the claim covers a particular solution to a problem or a particular way to achieve a desired outcome, as opposed to merely claiming the idea of a solution or outcome.” “the USPTO identifies other examples of claims that improve technology and are not directed to a judicial exception from Federal Circuit decisions: … Claims to a cardiac monitoring device that analyzes the variability in the beat-to-beat timing for atrial fibrillation and atrial flutter to more accurately detect the occurrence of these cardiac conditions were directed to an improvement in cardiac monitoring technology and not an abstract idea,CardioNet, LLC v. InfoBionic, Inc., 955 F.3d 1358, 1368-69 (Fed. Cir. 2020).”
Analysis: As cited and outlined by the applicant in the specifications, the above claims are used to improve classification accuracy of the electrophysiological signal. The claim has been amended to include the specific way the model accomplishes this task, such that the model does not cover all ways of performing the judicial exception with a model “through a temporal convolution layer … through network embedding … using matrix multiplication processing …. “through a spatial convolution layer … through at least a non-linear layer and a fully connected layer.” The above amended limitations provide a particular way to achieve the outcome of classification processing. This claim closely matches the claims to a cardiac monitoring device that analyzes the variability in the beat to beat timing for atrial fibrillation which is directed to an improvement in cardiac monitoring. The combination of the specific machine learning model used was found to be an improvement towards brain monitoring technology rather than an attempt to monopolize the judicial exceptions of signal classification within the field of brain scans.
Conclusion: The examiner finds arguments convincing in light of the amendments and the rejection is withdrawn by the examiner.
Response to 102 and 103
Issue: Applicant argues that prior art fails to disclose newly added limitations in the claims. Specifically the newly added limitations of “extracting … through a temporal convolution layer … time feature … generating … through a network embedding layer … using matrix multiplication processing .. time feature” Applicant argues that while Zhang_2020 does include matrix multiplication, Zhang_2020 is silent regarding the generation of a convolution based time feature at the relevant stage.
Rule: Claims are given their broadest reasonable interpretation in light of the specifications and what is commonly understood in the art. MPEP 2144.01 states "[I]n considering the disclosure of a reference, it is proper to take into account not only specific teachings of the reference but also the inferences which one skilled in the art would reasonably be expected to draw therefrom."Analysis: Examiner disagrees with applicant argument that Zhang_2020 does not teach amended claim limitations. Zhang_2020 teaches the usage of Convolutional Neural Networks with spatial temporal modeling, see figure 2: “CNN layer to extract spatio-temporal features of each slice” One ordinarily in the art would infer that this encompasses a convolutional spatial layer. Zhang_2020 also discusses the use of matrix multiplications to generate the embedded features. However, examiner agrees that Zhang_2020 does not expressly recite all limitations as claimed. As scope of claim has been amended, please see new grounds of rejection introduced which are mapped to the amended claim limitations. Conclusion: Examiner considered applicant arguments persuasive and previous 102 rejection is withdrawn. New grounds of rejection will be introduced as necessitated by applicant amendment.
End Response to Arguments
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claim 9 is rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 9 recites the limitation “for each one of at least one temporal convolution kernel of the temporal convolution layer,”. There is insufficient antecedent basis for this limitation in the claim.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-3, and 9-17 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang_2020 (“Motor Imagery Classification via Temporal Attention Cues of Graph Embedded EEG Signals.” By Zhang et al.) and Xu_2020 (“Multivariate Time Series Classification with Hierarchical Variational Graph Pooling”)
Claim 1:
Zhang_2020 makes obvious A method for classification processing of an electrophysiological signal, the method comprising: (page 2570 col 1 par 2: Motor imagery classification is the basic to a BCI, which supports motor rehabilitation of post-stroke patients [1]. The EEG signals, (Examiner note: an electrophysiological signal) which are captured from a human’s scalp and thus reflect the electrical activities of human the cortex, is one of the most active physiological cues to build a BCI system. … page 2572 col 1 par 3: “The beep and cue are used to notice and indicate the subject to perform the motor imagery task. The duration of motor imagery is of research interest. Formally, the duration of interest is T-second long. Each of the n EEG nodes has a sensor recording sequence ri∈[1,n] = [si 1,si 2,...,si k] ∈ Rk through k = T × f time points, where f is the sampling frequency and si t is the measurement of the ith EEG sensor at the time point t. Thus the raw EEG features of the trial T is a two-dimensional(2D) tensor XT =[r1;r2;...;rn] ∈ Rn×k with one dimension representing EEG node and the other representing time series. Our goal is to make motor imagery classification of the EEG trials XT.”
acquiring a multi-channel electrophysiological signal collected by an acquisition device; ((page 2570 col 1 par 2: “Researchers have widely explored the EEG-based BCI due to its zero clinical risks as well as portable and cost-effective acquisition devices.” … page 2572 col 1 par 3: “The beep and cue are used to notice and indicate the subject to perform the motor imagery task. The duration of motor imagery is of research interest. Formally, the duration of interest is T-second long. Each of the n EEG nodes has a sensor recording sequence) … page 2570 col 2 par 2: “In the meantime, not all EEG nodes can provide distinguishable information in terms of PSD features. An EEG channel selection approach is usually preferred to choose the most discriminative EEG nodes [6]. C3, C4, and Cz are three commonly reported channels that are most useful for motor imagery classification.”)
acquiring a channel association feature corresponding to the acquisition device, the channel association feature indicating spatial locations of multiple acquisition channels of the acquisition device, each of the multiple acquisition channels collecting the multi-channel electrophysiological signal at a respective spatial location; (page 2572 col 2 par 2: “In the node dimension of XT, one EEG node at most has two neighbors. Such a representation is limited to reflect the real-world situation where an EEG node usually has multiple neighboring nodes acquiring EEG signals of a certain brain area. Thus representing the relations of different EEG nodes is essential to successful EEG analysis. In our work, we leverage the EEG node positioning to form graph representations of EEG nodes, which include spatial information of the natural EEG node. In particular ,we construct an undirected spatial graph G =(V,E) on the EEG node positioning. The node set V = {si|i ∈ [1,n]} includes all the EEG nodes in an experiment. Depending on the structure of the adjacency matrix of EEG nodes, we design three EEG representation graphs: N-Graph (NG), D-Graph (DG), and S-Graph (SG). The graph definition enhances the brain area representation ability of EEG signals but decreases the effect of noise on each EEG node by combining neighboring nodes to represent the central one. This design also empowers the EEG representations to be robust to missing value issues by embedding each EEG node with the assist of its neighboring nodes instead of only relying on the measurement of itself. 1) N-Graph: Fig. 3 shows an example positioning of 64-channel EEG nodes.”)
extracting, by processing circuitry through a [spatial-] temporal convolution layer of an electrophysiological signal classification model, a time feature based on the multi-channel electrophysiological signal; (fig 2: Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification. We first represent the raw EEG measurement by a spatial graph drawn from EEG node positions; then we apply a sliding window technique to crop continuous EEG sequences into temporal slices and utilize a CNN layer to extract spatio-temporal features of each slice; a recurrent attention layer is used to extract the attentive temporal dynamic features; lastly the extracted features are classified to the target using a dense layer and a standard softmax classifier.)
generating, by the processing circuitry through a network embedding layer of the electrophysiological signal classification model using matrix multiplication processing, an embedded feature based on the channel association feature and the time feature; (page 2572 section B: Pipeline Overview: “Fig. 2 shows an overview of our proposed approach. The EEG signals are first embedded by a graph representation. Three different graph embedding schemes are developed based on different considerations.” One example is given page 2572 col 2 1) N-graph “Then the N-Graph representation Zv of raw EEG signals is the matrix product of the normalized N-Graph adjacency matrix Aˆv and the raw EEG trial XT :”)
extracting, by the processing circuitry through a spatial convolution layer of the electrophysiological signal classification model, a spatial feature based on the embedded feature; (fig 2: Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification. We first represent the raw EEG measurement by a spatial graph drawn from EEG node positions; (Examiner note: the embedded feature) then we apply a sliding window technique to crop continuous EEG sequences into temporal slices and utilize a CNN layer to extract spatio-temporal features of each slice; a recurrent attention layer is used to extract the attentive temporal dynamic features; lastly the extracted features are classified to the target using a dense layer and a standard softmax classifier.)
and obtaining, by the processing circuitry through at least a non-linear layer and a fully connected layer of the electrophysiological signal classification model, a classification result corresponding to the multi-channel electrophysiological signal based on the spatial feature, (fig 2: Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification. We first represent the raw EEG measurement by a spatial graph drawn from EEG node positions; then we apply a sliding window technique to crop continuous EEG sequences into temporal slices and utilize a CNN layer to extract spatio-temporal features of each slice; a recurrent attention layer is used to extract the attentive temporal dynamic features; lastly the extracted features are classified to the target using a dense layer (examiner note: fully connected layer) and a standard softmax classifier.) … Fig 4. “Fig. 4. Illustration of the self-attention module. A nonlinear encoding layer first transforms the encoded EEG temporal slices and the results are scaled and normalized to get the attention weight of each temporal slice. Lastly, the attention weight is multiplied with its corresponding encoded features”) (Examiner note: where this process in figure 4 occurs after the spatial encoding and is used for classification)
wherein the electrophysiological signal classification model is a machine learning model that is trained based on a training multi-channel electrophysiological signal collected by the acquisition device. (page 2575 col 1 par 5: “2) BCICIV2a Dataset: The BCICIV2a dataset contains EEG signals of 22 nodes recorded with nine healthy subjects and two sessions on two different days. Each session consists of 288 four-second trials of motor imagery per subject (imagining the movement of the left hand, the right hand, the feet, and the tongue). The signals were sampled with 250 Hz and bandpass-filtered between 0.5 Hz and 100 Hz by the dataset provider before release. The original dataset uses the 288 trials of the first session as training and the 288 trials of the second session as a test. However, in the subject-independent scenario, the original dataset needs to be re-split by subject with the leave-one-subject-out manner. Consequently, nine evaluation datasets (A01-A09) are achieved, each of which has 576 trials (288 trials × 2 sessions) of one subject as a test and 4608 trials (288 trials × 2 sessions × 8 subjects) of the remaining eight subjects as training.”)
Zhang_2020 Does not expressly recite [a separate temporal layer]
Xu_2020 however makes obvious [a separate temporal layer]
(page 3 col 2 par 5 under section Temporal convolution: “Therefore, it is reasonable and necessary to extract the features of the time series in units of multiple specific periods. To simulate this situation, we use multiple CNN filters with different receptive fields, namely kernel sizes, to extract features at multiple time scales. For the i-th CNN filters, given the input time se1ies X, the feature vector h; are extracted as follows h;. = cr(1Y, * X +b), where * denotes the convolution operation, er is a nonlinear activation function, such as RELU (:r) = max(O, x),W; represents the i.-th CNN kernel and bis the bias.” Examiner note: See figure 1 which also shows temporal convolution occurring separately before spatial temporal convolution
Zhang_2020 and Xu_2020 are analogous art to the claimed invention because they are from the same field of endeavor called EEG electrode signal classification. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Zhang_2020 and Xu_2020. The rationale for doing so would have been the use of a known technique to improve similar devices in the same way.
Zhang_2020 teaches the use of a CNN window to extract time feature temporal slices page 2573 fig. 2. “Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification. We first represent the raw EEG measurement by a spatial graph drawn from EEG node positions; then we apply a sliding window technique to crop continuous EEG sequences into temporal slices and utilize a CNN layer to extract spatio-temporal features of each slice;” See also page 2574 section “spatio-Temporal Encoding.” Zhang_2020 does this after first generating the embedded features.
However, Xu_2020 first extracts the time features then generates the embeddings for spatial temporal modeling. Figure 1: “Adjacency matrix and feature matrix are constructed through graph structure learning and temporal convolution respectively. Then GNNs aggregate and fuse spatial-temporal features.” One ordinarily in the art could have applied the known improvement ,generating temporal convolutions before spatial temporal modeling” for the predictable result of extracting features along multiple time scales before embedding. This result would have been predictable to one ordinarily skilled in the art, as Zhang_2020 does this exact process, (cropping the signal through windows) , and both methodologies lead to the same result of allowing spatial-temporal modeling to occur after.
Therefore, it would have been obvious to combine the CNN signal classification workflow of Zhang_2020 with the use of a explicitly separate temporal convolution of Xu_2020 for the benefit and predictable result of extracting the temporal features of the EEG signal for spatial temporal modeling to obtain the invention as specified in the claims.
Claim 2:The method according to claim 1, further comprising:
Zhang_2020 makes obvious Acquiring the electrophysiological signal classification model corresponding to the acquisition device, page 2573 fig 2: “Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification.” see also table II on page 2575 which compares different models, implies that a model is acquired, where all the above models use/correspond to an acquisition device.
wherein the electrophysiological signal classification model is based on the channel association feature corresponding to the acquisition device(page 2578 col 1 par 2: “given an EEG headset, the coordinates (locations) of its EEG nodes would be fixed, and consequently, the graph representation could be achieved. Therefore, the proposed graph representation approach is adaptive to different amounts of EEG nodes” Examiner note: this also implies that the channel association feature corresponds to the device (headset) as well. )
Claim 3:The method according to claim 2, wherein the acquiring the electrophysiological signal classification model comprises:
Zhang_2020 makes obvious determining a classification task; (page 2578 col 1 par 6: “This paper targets the EEG motor imagery classification task and proposes a novel deep learning approach.”)
acquiring multiple candidate electrophysiological signal classification models corresponding to the classification task, the candidate electrophysiological signal classification models having corresponding candidate acquisition devices; (page 2574 col 2 par 2: “The PhysioNet dataset and BCI CIV2a dataset we used are roughly balanced. Thus we evaluate the proposed model with classification accuracy and the Area Under ROC Curve (ROC-AUC). Table II presents the overall comparison results and the detailed results can be found in the supporting documents. Because deep learning is an advanced technique that relies on proper structure design ,we compare with several deep learning approaches with various model structures and feature embedding strategies. To make a fair comparison and show the superior structure of the proposed approach, the most recent state-of-the-art approaches whose implementation code is available online are selected for comparison.” Examiner note: Where the comparison of different models in table II for a same task makes implies acquiring those models for the same task. Where all models correspond to standard acquisition device, as implied by page 2578 col 1 par 2: “given an EEG headset, the coordinates (locations) of its EEG nodes would be fixed, and consequently, the graph representation could be achieved. Therefore, the proposed graph representation approach is adaptive to different amounts of EEG nodes”
and selecting the electrophysiological signal classification model corresponding to the acquisition device from the candidate electrophysiological signal classification models. (page 2573 fig2 description: “Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification (Examiner note: Where the use of this model implies the selection of that model for the task amongst a plurality of options.). We first represent the raw EEG measurement by a spatial graph drawn from EEG node positions (Examiner note: corresponds to the acquisition device) ; then we apply a sliding window technique to crop continuous EEG sequences into temporal slices and utilize a CNN layer to extract spatio-temporal features of each slice; a recurrent attention layer is used to extract the attentive temporal dynamic features; lastly the extracted features are classified to the target using a dense layer and a standard softmax classifier.” )
Claim 9:The method according to claim 1, wherein
Zhang_2020 makes obvious the multi-channel electrophysiological signal includes electrophysiological signals corresponding to each of the multiple acquisition channels respectively, and(page 2572 col 1 par 4: “Each of the n EEG nodes has a sensor recording sequence ri∈[1,n] = [si 1,si 2,...,si k] ∈ Rk through k = T × f time points, where f is the sampling frequency and si t is the measurement of the ith EEG sensor at the time point t.”)
Zhang_2020 does not expressly recite the extracting the time feature includes:
for each one of at least one temporal convolution kernel of the temporal convolution layer, separately extracting time sub-features of each of the electrophysiological signals based on the corresponding temporal convolution kernel to obtain a respective set of time sub-features corresponding to each of the at least one temporal convolution kernel;
for each one of at least one temporal convolution kernel of the temporal convolution layer, generating an intermediate time feature based on the respective set of time sub-features to obtain at least one intermediate time feature corresponding to the at least one temporal convolution kernel respectively;
and obtaining the time feature based on the at least one intermediate time feature.
Xu_2020 however makes obvious the extracting the time feature includes:
for each one of at least one temporal convolution kernel of the temporal convolution layer, (page 3 col 2 par 5 under section Temporal Convolution: “Therefore, it is reasonable and necessary to extract the features of the time series in units of multiple specific periods. To simulate this situation, we use multiple CNN filters with different receptive fields, namely kernel sizes, to extract features at multiple time scales.”
separately extracting time sub-features of each of the electrophysiological signals based on the corresponding temporal convolution kernel to obtain a respective set of time sub-features corresponding to each of the at least one temporal convolution kernel; (page 3 col 2 par 5 under section Temporal convolution: “Therefore, it is reasonable and necessary to extract the features of the time series in units of multiple specific periods. To simulate this situation, we use multiple CNN filters with different receptive fields, namely kernel sizes, to extract features at multiple time scales. For the i-th CNN filters, given the input time se1ies X, the feature vector h; are extracted as follows h;. = cr(1Y, * X +b), where * denotes the convolution operation, er is a nonlinear activation function, such as RELU (:r) = max(O, x),W; represents the i.-th CNN kernel and bis the bias.”
for each one of at least one temporal convolution kernel of the temporal convolution layer, generating an intermediate time feature based on the respective set of time sub-features to obtain at least one intermediate time feature corresponding to the at least one temporal convolution kernel respectively; (page 3 col 2 par 5 under section Temporal convolution: “Therefore, it is reasonable and necessary to extract the features of the time series in units of multiple specific periods. To simulate this situation, we use multiple CNN filters with different receptive fields, namely kernel sizes, to extract features at multiple time scales. For the i-th CNN filters, given the input time se1ies X, the feature vector h; are extracted as follows h;. = cr(1Y, * X +b), where * denotes the convolution operation, er is a nonlinear activation function, such as RELU (:r) = max(O, x),W; represents the i.-th CNN kernel and bis the bias.”
and obtaining the time feature based on the at least one intermediate time feature. (page 3 col 2 par 6: “And the final feature vector can be expressed ash= [h 1 , h2 , ... , h1J where p is the CNN filters number and [*] means concatenate operation. In this way, features under different period
are extracted, which provides effective information for time series classification.” Examiner note: the mathematic formulas are hard to see, please refer to the prior art source for more clarity. Where the examiner understands that extracting a feature vector is the generation of an intermediate time feature, where the final feature vector is a concatenation of the intermediate time features.
Zhang_2020 and Xu_2020 are analogous art to the claimed invention because they are from the same field of endeavor called EEG electrode signal classification. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Zhang_2020 and Xu_2020. The rationale for doing so would have been the use of a known technique to improve a similar device in the same way.
Zhang_2020 teaches the use of a CNN window to extract time feature temporal slices page 2573 fig. 2. “Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification. We first represent the raw EEG measurement by a spatial graph drawn from EEG node positions; then we apply a sliding window technique to crop continuous EEG sequences into temporal slices and utilize a CNN layer to extract spatio-temporal features of each slice;” See also page 2574 section “spatio-Temporal Encoding.” Zhang_2020 does not explicitly mention that this is done only on the time dimension and separately to the embedded layer. In the context of Zhang_2020, the kernels act on the embedded layer which contains both the time and space dimensions. See page 2574 Table 1 which explicitly mentions the use of Kernels in the layers.
However, Xu_2020 implies that a kernel is a way to perform the function as outlined by Zhang_2020 separately for the temporal layer. Xu_2020 states page 3 col 2 par 5: “Therefore, it is reasonable and necessary to extract the features of the time series in units of multiple specific periods. To simulate this situation, we use multiple CNN filters with different receptive fields, namely kernel sizes, to extract features at multiple time scales.” Therefore, using temporal kernel exclusively is a known technique to applying CNN filters for temporal fields. Zhang_2020 outlines a base device that uses CNN filters on an embedded layer to classify the temporal aspects of a signal. Xu_2020 is a comparable device, which specifically outlines that the filters are temporal Kernels only. One ordinarily skilled in the art would appreciate and understand that the CNN windows of Zhang_2020 are also kernels that apply temporally, and function in a similar way to the claimed invention, with the only difference being the ordering where Zhang_2020 embeds the time features first before applying the kernel.
Therefore, it would have been obvious to combine the CNN windows of Zhang_2020 with the use of a temporal kernel of Xu_2020 for the benefit of extracting the temporal features of the EEG signal to obtain the invention as specified in the claims.
Claim 10:method according to claim 1,
Zhang_2020 makes obvious wherein the time feature includes multiple intermediate time features, (page 2573 fig.2.: “we apply a sliding window technique to crop continuous EEG sequences into temporal slices” Examiner note: intermediate time features)
and the generating the embedded feature includes:
separately embedding the channel association feature into the intermediate time features to obtain initial embedded features corresponding to the intermediate time features; page 2574 col 1 par 2: “Let the interval between two neighbouring slices be p, then m = int((k − w)/p) slices are obtained from one EEG trial. We specifically design a CNN to encode the spatio-temporal information within a temporal slice.” Examiner note: Where the information encoded within a temporal splice implies that the information is embedded into each intermediate time feature (ie: slice) separately. See annotated figure below.
and obtaining the embedded feature based on the initial embedded features.
See annotated figure 2. Page 2573 below
PNG
media_image1.png
377
1154
media_image1.png
Greyscale
Claim 11:Zhang_2020 makes obvious The method according to claim 10, wherein
the channel association feature includes association sub-features corresponding to the acquisition channels respectively, (page 2572 col 1 par 4L “Formally, the duration of interest is T-second long. Each of the n EEG nodes has a sensor recording sequence ri∈[1,n] = [si 1,si 2,...,si k] ∈ Rk through k = T × f time points, where f is the sampling frequency and si t is the measurement of the ith EEG sensor at the time point t.”)
the intermediate time features include time sub-features corresponding to the acquisition channels respectively, ; (page 2572 col 1 par 4: “.Thus the raw EEG features of the trial T is a two-dimensional(2D) tensor XT =[r1;r2;...;rn] ∈ Rn×k with one dimension representing EEG node and the other representing time series.”) Examiner note: Where this shows that the time feature includes a sub feature that corresponds to each EEG node.
and the separately embedding the channel association feature into the intermediate time features includes, for a current intermediate time feature of the intermediate time features:
embedding an association sub-feature corresponding to a same acquisition channel into a corresponding time sub-feature, to obtain embedded sub-features corresponding to the acquisition channels respectively; page 2574 col 1 par 2: “Although deep networks have strong learning abilities, deeper is not always better for EEG analysis [43]. Table I gives the detailed configuration of the proposed spatio-temporal encoding network. We use one CNN layer and one pooling layer. The height of the CNN kernel is set to n, same to the amount of EEG nodes, for considering all EEG nodes at once. The width of the kernel is extended to 45 for exploring long temporal dynamics. The output amount of CNN filters is empirically set to 40. The convolutional filtering thus can uncover the spatio-temporal information across different EEG nodes.” Examiner note: Where as the examiner understands, time sub-features corresponding to acquisition channels is the retrieval of the temporal information across the different EEG nodes. Where this process is occurring as part of the embedding step, see figure 2 below as well for visualization.
PNG
media_image2.png
368
1154
media_image2.png
Greyscale
and obtaining an initial embedded feature corresponding to the current intermediate time feature based on the embedded sub-features.
See annotated figure 2. Page 2573 below
PNG
media_image1.png
377
1154
media_image1.png
Greyscale
Claim 12:
The method according to claim 1, wherein the extracting the spatial feature comprises:
Zhang_2020 makes obvious performing spatial feature extraction on the embedded feature based on at least one spatial convolution kernel of the spatial [temporal] convolution layer to obtain at least one intermediate spatial feature corresponding to the at least one spatial convolution kernel respectively; (page 2574 col 1 par 3: “We use one CNN layer and one pooling layer. The height of the CNN kernel is set to n, same to the amount of EEG nodes, for considering all EEG nodes at once. The width of the kernel is extended to 45 for exploring long temporal dynamics. The output amount of CNN filters is empirically set to 40. The convolutional filtering thus can uncover the spatio-temporal information across different EEG nodes. Each temporal slice is encoded to higher-level representations {Ui ∈ Rwc|Ui = Conv(Qi),i∈ [1,m]}. The activation function used in the convolutional operations is the Exponential Linear Unit (ELU) function. We use the valid padding option. Thus the output of the CNN layer has the height of 1. A maxpooling layer is then applied to reduce the number of parameters and extract important information.” Examiner information: Where this important information is understood to be intermediate spatial features.)
and obtaining the spatial feature based on the at least one intermediate spatial feature. Page 2574 col 1 par 4: “Following the spatio-temporal feature extraction within single EEG temporal slices, are current attention network is introduced to discover the attentive temporal dependencies across different EEG temporal slices. In traditional recurrent networks, the features that are accumulated from the previous time step are usually adopted for further analysis.” Examiner note: Where the intermediate spatial features based on a time are used to determine the spatial feature.
Zhang_2020 Does not expressly recite [a separate temporal layer]
Xu_2020 however makes obvious [a separate temporal layer]
(page 3 col 2 par 5 under section Temporal convolution: “Therefore, it is reasonable and necessary to extract the features of the time series in units of multiple specific periods. To simulate this situation, we use multiple CNN filters with different receptive fields, namely kernel sizes, to extract features at multiple time scales. For the i-th CNN filters, given the input time se1ies X, the feature vector h; are extracted as follows h;. = cr(1Y, * X +b), where * denotes the convolution operation, er is a nonlinear activation function, such as RELU (:r) = max(O, x),W; represents the i.-th CNN kernel and bis the bias.” Examiner note: See figure 1 which also shows temporal convolution occurring separately before spatial temporal convolution
Zhang_2020 and Xu_2020 are analogous art to the claimed invention because they are from the same field of endeavor called EEG electrode signal classification. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Zhang_2020 and Xu_2020. The rationale for doing so would have been the use of a known technique to improve similar devices in the same way.
Zhang_2020 teaches the use of a CNN window to extract time feature temporal slices page 2573 fig. 2. “Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification. We first represent the raw EEG measurement by a spatial graph drawn from EEG node positions; then we apply a sliding window technique to crop continuous EEG sequences into temporal slices and utilize a CNN layer to extract spatio-temporal features of each slice;” See also page 2574 section “spatio-Temporal Encoding.” Zhang_2020 does this after first generating the embedded features.
However, Xu_2020 first extracts the time features then generates the embeddings for spatial temporal modeling. Figure 1: “Adjacency matrix and feature matrix are constructed through graph structure learning and temporal convolution respectively. Then GNNs aggregate and fuse spatial-temporal features.” One ordinarily in the art could have applied the known improvement ,generating temporal convolutions before spatial temporal modeling” for the predictable result of extracting features along multiple time scales before embedding. This result would have been predictable to one ordinarily skilled in the art, as Zhang_2020 does this exact process, (cropping the signal through windows) , and both methodologies lead to the same result of allowing spatial-temporal modeling to occur after.
Therefore, it would have been obvious to combine the CNN signal classification workflow of Zhang_2020 with the use of a explicitly separate temporal convolution of Xu_2020 for the benefit and predictable result of extracting the temporal features of the EEG signal for spatial temporal modeling to obtain the invention as specified in the claims.
Claim 13:The method according to claim 1, wherein the obtaining the classification result comprises:
Zhang_2020 makes obvious performing nonlinear processing on the spatial feature through the non-linear layer to obtain a target fitting feature; (page 2574 fig 4: “Illustration of the self-attention module. A nonlinear encoding layer first transforms the encoded EEG temporal slices and the results are scaled and normalized to get the attention weight of each temporal slice. Lastly, the attention weight is multiplied with its corresponding encoded features.” … “Each slice representation hi is first non-linearly transformed into a latent space: Hi =tanh(Wihi +bi),Hi ∈ Rha whereWi ∈ Rl× ha andbi ∈ Rha are the input-to-hidden weight matrix and bias for a hidden layer of size ha. The softmax activation function, defined as softmax(xi) = 1 Zexp(xi) with Z = iexp(xi), is applied to the nonlinear latent representation Hi to obtain the weight of importance for each slice:” Examiner note: Where the attentive layers are interpreted as target fitting features.
and performing classification processing on the target fitting feature through the fully connected layer to obtain the classification result. (fig 2: Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification. We first represent the raw EEG measurement by a spatial graph drawn from EEG node positions; then we apply a sliding window technique to crop continuous EEG sequences into temporal slices and utilize a CNN layer to extract spatio-temporal features of each slice; a recurrent attention layer is used to extract the attentive temporal dynamic features; lastly the extracted features are classified to the target using a dense layer (examiner note: fully connected layer) and a standard softmax classifier.) … Page 2574 col 2 par 4: “Lastly, in the interest of computational efficiency, a weighted sum of all EEG temporal slices is computed to a slice-focused representation: A= Vihi,A∈ Rl The attentive temporal dynamic representation A is fed into a standard softmax classifier: P =softmax(WA+b), where W and b are weight and bias matrices respectively of the motor imagery classification layers. Then the cross-entropy error over all labeled samples is evaluated: L =− ˆ Yclog(Pc), c where ˆYc and Pc is the label and the classification probability of motor imagery strategy c respectively. The network weights and biases are trained with batch gradient descent. The final classification result is defined as the motor imagery strategy with max classification probability.)
Claim 14:Zhang_2020 makes obvious A method for classification processing the method comprising: (page 2570 col 1 par 2: Motor imagery classification is the basic to a BCI, which supports motor rehabilitation of post-stroke patients [1]. The EEG signals, (Examiner note: an electrophysiological signal) which are captured from a human’s scalp and thus reflect the electrical activities of human the cortex, is one of the most active physiological cues to build a BCI system. … page 2572 col 1 par 3: “The beep and cue are used to notice and indicate the subject to perform the motor imagery task. The duration of motor imagery is of research interest. Formally, the duration of interest is T-second long. Each of the n EEG nodes has a sensor recording sequence ri∈[1,n] = [si 1,si 2,...,si k] ∈ Rk through k = T × f time points, where f is the sampling frequency and si t is the measurement of the ith EEG sensor at the time point t. Thus the raw EEG features of the trial T is a two-dimensional(2D) tensor XT =[r1;r2;...;rn] ∈ Rn×k with one dimension representing EEG node and the other representing time series. Our goal is to make motor imagery classification of the EEG trials XT.”
acquiring a training multi-channel electrophysiological signal collected by an acquisition device and a training label corresponding to the training multi-channel electrophysiological signal; page 2575 col 1 par 5: “2) BCICIV2a Dataset: The BCICIV2a dataset contains EEG signals of 22 nodes recorded with nine healthy subjects and two sessions on two different days. Each session consists of 288 four-second trials of motor imagery per subject (imagining the movement of the left hand, the right hand, the feet, and the tongue). The signals were sampled with 250 Hz and bandpass-filtered between 0.5 Hz and 100 Hz by the dataset provider before release. The original dataset uses the 288 trials of the first session as training and the 288 trials of the second session as a test. However, in the subject-independent scenario, the original dataset needs to be re-split by subject with the leave-one-subject-out manner. Consequently, nine evaluation datasets (A01-A09) are achieved, each of which has 576 trials (288 trials × 2 sessions) of one subject as a test and 4608 trials (288 trials × 2 sessions × 8 subjects) of the remaining eight subjects as training.”
inputting the training multi-channel electrophysiological signal into an initial electrophysiological signal classification model corresponding to the acquisition device, (page 2575 col 1 par 5: “. The original dataset uses the 288 trials of the first session as training”)
the initial electrophysiological signal classification model including a channel association feature corresponding to the acquisition device, the channel association feature indicating spatial locations of multiple acquisition channels of the acquisition device; (page 2576 col 1 par 2: “Compared with the pure deep learning models which do not have particular data representations, like EEGNet and CTCNN, our proposed graph representation embeds the spatial relationship of EEG nodes, which facilitates the following neural network to analyze EEG signals.” … “our graph scheme introduces an adjacency matrix to optimize the raw data to a more effective embedding.”)
extracting, by processing circuitry through a [spatial] temporal convolution layer of the initial electrophysiological signal classification model, a time feature based on the training multi-channel electrophysiological signal; (fig 2: Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification. We first represent the raw EEG measurement by a spatial graph drawn from EEG node positions; (Examiner note: the embedded feature) then we apply a sliding window technique to crop continuous EEG sequences into temporal slices and utilize a CNN layer to extract spatio-temporal features of each slice; a recurrent attention layer is used to extract the attentive temporal dynamic features; lastly the extracted features are classified to the target using a dense layer and a standard softmax classifier.)
generating, by the processing circuitry through a network embedding layer of the initial electrophysiological signal classification model using matrix multiplication processing, an embedded feature based on the channel association feature and the time feature; (page 2572 section B: Pipeline Overview: “Fig. 2 shows an overview of our proposed approach. The EEG signals are first embedded by a graph representation. Three different graph embedding schemes are developed based on different considerations.” One example is given page 2572 col 2 1) N-graph “Then the N-Graph representation Zv of raw EEG signals is the matrix product of the normalized N-Graph adjacency matrix Aˆv and the raw EEG trial XT :”)
extracting, by the processing circuitry through a spatial convolution layer of the initial electrophysiological signal classification model, a spatial feature based on the embedded feature (fig 2: Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification. We first represent the raw EEG measurement by a spatial graph drawn from EEG node positions; (Examiner note: the embedded feature) then we apply a sliding window technique to crop continuous EEG sequences into temporal slices and utilize a CNN layer to extract spatio-temporal features of each slice; a recurrent attention layer is used to extract the attentive temporal dynamic features; lastly the extracted features are classified to the target using a dense layer and a standard softmax classifier.)
obtaining, by the processing circuitry through at least a non-linear layer and a fully connected layer of the initial electrophysiological signal classification model, a predicted label corresponding to the training multi-channel electrophysiological signal based on the spatial feature; (fig 2: Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification. We first represent the raw EEG measurement by a spatial graph drawn from EEG node positions; then we apply a sliding window technique to crop continuous EEG sequences into temporal slices and utilize a CNN layer to extract spatio-temporal features of each slice; a recurrent attention layer is used to extract the attentive temporal dynamic features; lastly the extracted features are classified to the target using a dense layer (examiner note: fully connected layer) and a standard softmax classifier.) … Fig 4. “Fig. 4. Illustration of the self-attention module. A nonlinear encoding layer first transforms the encoded EEG temporal slices and the results are scaled and normalized to get the attention weight of each temporal slice. Lastly, the attention weight is multiplied with its corresponding encoded features”) (Examiner note: where this process in figure 4 occurs after the spatial encoding and is used for classification)
and adjusting a model parameter of the initial electrophysiological signal classification model based on a difference between the training label and the predicted label, until a convergence condition is met, to obtain a trained electrophysiological signal classification model. Page 2575 col 2 par 1: “We make use of the TensorFlow framework for a GPU-based implementation using matrix multiplications. The stochastic gradient descent with Adam update rule is used to minimize the cross-entropy loss function. The network parameters are optimized with a learning rate of 10−5. Dropout regularization is applied after the CNN layer and the recurrent network layer with the dropout probability of 0.5. The hidden state size of theLSTMcelllis64.The non-linear transformation size of the self-attention is 512. The proposed model has 16 hyper-parameters and 420,356 trainable parameters” Examiner note: See also the abstract on page 2570, which implies that it is known in the art that an EEG model is trained. And that this model is a trained model, “Thus the research of directly extending a pre-trained model to new users is particularly desired and indispensable.”
Zhang_2020 Does not expressly recite [a separate temporal layer]
Xu_2020 however makes obvious [a separate temporal layer]
(page 3 col 2 par 5 under section Temporal convolution: “Therefore, it is reasonable and necessary to extract the features of the time series in units of multiple specific periods. To simulate this situation, we use multiple CNN filters with different receptive fields, namely kernel sizes, to extract features at multiple time scales. For the i-th CNN filters, given the input time se1ies X, the feature vector h; are extracted as follows h;. = cr(1Y, * X +b), where * denotes the convolution operation, er is a nonlinear activation function, such as RELU (:r) = max(O, x),W; represents the i.-th CNN kernel and bis the bias.” Examiner note: See figure 1 which also shows temporal convolution occurring separately before spatial temporal convolution
Zhang_2020 and Xu_2020 are analogous art to the claimed invention because they are from the same field of endeavor called EEG electrode signal classification. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Zhang_2020 and Xu_2020. The rationale for doing so would have been the use of a known technique to improve similar devices in the same way.
Zhang_2020 teaches the use of a CNN window to extract time feature temporal slices page 2573 fig. 2. “Overview of the graph Convolutional Recurrent Attention Model (G-CRAM) on EEG motor imagery classification. We first represent the raw EEG measurement by a spatial graph drawn from EEG node positions; then we apply a sliding window technique to crop continuous EEG sequences into temporal slices and utilize a CNN layer to extract spatio-temporal features of each slice;” See also page 2574 section “spatio-Temporal Encoding.” Zhang_2020 does this after first generating the embedded features.
However, Xu_2020 first extracts the time features then generates the embeddings for spatial temporal modeling. Figure 1: “Adjacency matrix and feature matrix are constructed through graph structure learning and temporal convolution respectively. Then GNNs aggregate and fuse spatial-temporal features.” One ordinarily in the art could have applied the known improvement ,generating temporal convolutions before spatial temporal modeling” for the predictable result of extracting features along multiple time scales before embedding. This result would have been predictable to one ordinarily skilled in the art, as Zhang_2020 does this exact process, (cropping the signal through windows) , and both methodologies lead to the same result of allowing spatial-temporal modeling to occur after.
Therefore, it would have been obvious to combine the CNN signal classification workflow of Zhang_2020 with the use of a explicitly separate temporal convolution of Xu_2020 for the benefit and predictable result of extracting the temporal features of the EEG signal for spatial temporal modeling to obtain the invention as specified in the claims.
Claim 15:
The limitations of claim 15 are substantially the same as those of claim 1 and are therefore rejected due to the same reasons as outlined above for claim 1. Additionally, Zhang_2020 teaches the additional limitations of “An apparatus for classification processing of an electrophysiological signal, comprising:
processing circuitry configured to (par 2575 col 1 par 1: “We make use of the TensorFlow framework for a GPU-based implementation using matrix multiplications. The stochastic gradient descent with Adam update rule is used to minimize the cross-entropy loss function.”) Examiner note: Where a GPU based implementation implies processing circuitry, which is performing the function of the experiment as outlined by Zhang_2020.
Claim 16:
The limitations of claim 16 are substantially the same as those of claim 2 except that it depends from claim 15. Therefore this claim is rejected due to the same reasons as outlined above for claims 2 and 15.
Claim 17:
The limitations of claim 17 are substantially the same as those of claim 3 except that it depends from claim 15. Therefore this claim is rejected due to the same reasons as outlined above for claims 3 and 15.
Claims 4, 6-8, 18, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang_2020 , Xu_2020, and Wang_2017 (“Topological Network Analysis of Electroencephalographic Power Maps”) as evidenced by Lucas_2016 ( “An introduction to the Delaunay triangulation”)
Claim 4:Zhang_2020 makes obvious The method according to claim 1, wherein the acquiring the channel association feature comprises:
mapping spatial locations of the multiple acquisition channels to a same plane to obtain plane locations of the multiple acquisition channels; page 2572 col 2 par 3: “1) N-Graph: Fig. 3 shows an example positioning of 64-channel EEG nodes. In the 2D position projection (Fig. 3(b)), each node has several naturally neighbors (up, down, left, right, up-left, up-right, down-left, and down right); for example, the node s11 has eight neighboring nodes (s3,s4,s5,s12,s19,s18,s17,s10). Based on this observation, we build a connection between two naturally neighboring EEG nodes.”
Zhang_2020 does not expressly recite defining a channel region based on the plane locations of at least three of the multiple acquisition channels;
associating the of the multiple acquisition channels when a region shape feature of the channel region is a preset shape feature and when there is no other acquisition channel in the channel region;
and generating the channel association feature based on an association relationship among the at least three of the multiple acquisition channels.
Wang_2017 as evidenced by Lucas_2016 however makes obvious defining a channel region based on the plane locations of at least three of the multiple acquisition channels;
associating the of the multiple acquisition channels when a region shape feature of the channel region is a preset shape feature and when there is no other acquisition channel in the channel region;
and generating the channel association feature based on an association relationship among the at least three of the multiple acquisition channels.
PNG
media_image3.png
720
641
media_image3.png
Greyscale
(page 2-3)
Examiner note: Where this passage states that the adjacency matrix between vertex’s (EEG channels) is based on the edge to edge connections in Delaunay triangulation neighbors. Where a Delaunay triangulation methodology to form the connections inherently implies the above limitations, as will be evidenced by Lucas_2016
Lucas_2016 page 1 par 2 :”
Given a set of arbitrarily distributed data points, there are many ways to organize them in a triangular mesh. But not all possible triangle meshes provide a favorable representation of the spatial relationships between points. How do we decide which vertices to connect in such a way as to obtain an optimal structure? A Delaunay mesh-building process makes that decision based on a principle known as "the Delaunay criterion". The Delaunay criterion specifies a rule for using geometric considerations to determine whether a pair of neighboring triangles represents an optimal choice of connections. When a pair fails to meet the criterion, the connections are reorganized so that they are "Delaunay conformant". The process continues until all triangles in the mesh meet the Delaunay criterion. In the end, the process creates a mesh that has a number of attractive and useful features and is optimal in many regards.”
PNG
media_image4.png
491
797
media_image4.png
Greyscale
(page 2)
Examiner note: Where as illustrated above. Dulaney triangulation to form adjacencies inherently defines the channel region based on at least three acquisition channels (the three vertexes of the triangle), where the preset shape feature is a triangle, and where there necessarily is no other acquisition channels in the channel region.
Wang_2017 is analogous art to the claimed invention because it is from the same field of endeavor called EEG signal analysis. Lucas_2016 is analogous art to the claimed invention because it explains an application of graph theory.
Before the effective filing date, it would have been obvious to a person ordinarily skilled in the art to combine Zhang_2020, And Wang_2017 as evidenced by Lucas_2016.
The rational for doing so would have been for the use of a known method to improve similar devices in the same way. Both Zhang_2020 and Wang_2017 analyze EEG signals in the brain by matching connections between the EEG channels. Zhang_2020 does this through distance based clustering methods, Wang_2017 does this through Delauney triangle topology methods. The result of both of these processes is to create an adjacency matrix. Lucas_2016 page 1 introduction states “ The Delaunay Triangulation is an important topic in graph theory and widely documented in published literature and on the Web.” One ordinarily skilled in the art then would be able to apply the widely known technique of Delauney triangulation to create the adjacency matrix for the predictable result of creating an adjacency matrix which would then be used in the invention of Zhang_2020 to accomplish the other limitations of the claim.
Therefore it would have been obvious to combine the classification workflow which creates adjacency matrix of Zhang_2020 with the process of Delauney triangulation for adjacency by Wang_2017 as evidenced by Lucas_2016 for the benefit of ensuring optimal connections between EEG signals to obtain the invention as specified in the claims.
Claim 6:The method according to claim 4, wherein the generating the channel association feature comprises:
Zhang_2020 makes further obvious generating an initial channel association feature based on the association relationship among the at least three of the multiple acquisition channels; (page 2572 col 2 par 3: “1) N-Graph: Fig. 3 shows an example positioning of 64-channel EEG nodes. In the 2D position projection (Fig. 3(b)), each node has several naturally neighbors (up, down, left, right, up-left, up-right, down-left, and down right); for example, the node s11 has eight neighboring nodes (s3,s4,s5,s12,s19,s18,s17,s10). Based on this observation, we build a connection between two naturally neighboring EEG nodes. Formally, the edge set can be denoted as Ev = {sisj|(i, j) ∈ H}, where H is the set of naturally neighboring EEG nodes. We also regard each node as connecting to itself. We can define the adjacency matrix of the N-Graph as a square matrix |V|×|V| with its binary element representing whether two EEG nodes are neighboring to each other:” Examiner note: Where this is an initial channel association feature pre-normalization.
and performing normalization processing on the initial channel association feature to obtain the channel association feature. (page 2572 par 4: “We then follow the spectral graph theory [31] to normalize the adjacency matrix: … Then the N-Graph representation Zv of raw EEG signals is the matrix product of the normalized N-Graph adjacency matrix ˆAv andthe raw EEG trial XT”)
Claim 7:The method according to claim 6, wherein the performing the normalization processing comprises:
generating an initial channel association matrix based on the initial channel association feature; (page 2572 col 2 par 3: “1) N-Graph: Fig. 3 shows an example positioning of 64-channel EEG nodes. In the 2D position projection (Fig. 3(b)), each node has several naturally neighbors (up, down, left, right, up-left, up-right, down-left, and down right); for example, the node s11 has eight neighboring nodes (s3,s4,s5,s12,s19,s18,s17,s10). Based on this observation, we build a connection between two naturally neighboring EEG nodes. Formally, the edge set can be denoted as Ev = {sisj|(i, j) ∈ H}, where H is the set of naturally neighboring EEG nodes. We also regard each node as connecting to itself. We can define the adjacency matrix of the N-Graph as a square matrix |V|×|V| with its binary element representing whether two EEG nodes are neighboring to each other:)
acquiring a cell matrix, and fusing the cell matrix and the initial channel association matrix to obtain an intermediate channel association matrix;
2572 col 2 par 3: “We also regard each node as connecting to itself. We can define the adjacency matrix of the N-Graph as a square matrix |V|×|V| with its binary element representing whether two EEG nodes are neighboring to each other:” Examiner note: please see claim interpretation. Based on the examiners understanding of the specifications, each node connecting to itself in the prior art is short hand for acquiring a cell matrix of 1’s in the diagonals and fusing it with the initial association matrix. See also the equation shown below which outlines this process, where A~ = A + I (where I is interpreted as an identity matrix which is the same as a cell matrix as defined in this applications specifications). See also claim interpretation which outlines this process with respect to this applications specifications.
PNG
media_image5.png
260
456
media_image5.png
Greyscale
acquiring a diagonal matrix corresponding to the initial channel association matrix, and fusing the diagonal matrix and the intermediate channel association matrix to obtain a target channel association matrix;
See math equations below from page 2572 which match this process as described. Where the diagonal matrix is given as D, and the fusion is given as A^. See also claim interpretation which outlines this process with respect to this applications specifications.
PNG
media_image6.png
260
445
media_image6.png
Greyscale
and obtaining the channel association feature based on the target channel association matrix.
PNG
media_image7.png
260
445
media_image7.png
Greyscale
Examiner note: Where the examiner under broadest reasonable interpretation understands that the target channel association matrix is an example of a target channel association feature.
Claim 8:The method according to claim 7, wherein the generating the initial channel association matrix comprises any one of:
determining matrix dimensionality of the initial channel association matrix based on a channel quantity of the acquisition channels on the acquisition device, setting matrix values corresponding to the associated acquisition channels to a first preset threshold, and setting matrix values corresponding to other acquisition channels to a second preset threshold, to obtain the initial channel association matrix;
or determining matrix dimensionality of the initial channel association matrix based on a channel quantity of the acquisition channels on the acquisition device, (page 2572 col 2 par 1-2: “The node set V = {si|i ∈ [1,n]} includes all the EEG nodes in an experiment. … We also regard each node as connecting to itself. Wecan define the adjacency matrix of the N-Graph as a square matrix |V|×|V| with its binary element representing whether two EEG nodes are neighboring to each other”) determining matrix values corresponding to the associated acquisition channels based on spatial location distances between the associated acquisition channels (page 2573 col 1 par 2: “Considering the above disadvantages, we define a distance based EEG graph called D-Graph, which uses the real-world 3D distance between EEG nodes rather than the binary connections between naturally neighboring nodes. The adjacency matrix of D-Graph has the distance between two neighboring EEG nodes as its element instead of binary elements indicating neighboring or not.”), and setting matrix values corresponding to other acquisition channels to the second preset threshold, (page 2573 col 1 par 2: “In practice, two issues should be addressed before constructing the adjacency matrix: 1) how to define neighboring nodes; 2) how to define the distance between a node and itself. For the first problem, we regard the two EEG nodes are neighboring if the distance between two nodes is smaller than the average value of the distance set L. For the second problem, the distance between a node and itself is defined as the average distance of other neighboring nodes to this node. Therefore, we define the elements of the adjacency matrix Ad as: “ Examiner note: See attached figure which depicts adjacent nodes having a value according to distance and non-neighboring/adjacent nodes having a preset value of zero.
PNG
media_image8.png
130
768
media_image8.png
Greyscale
to obtain the initial channel association matrix. Page 2573 col 2 par 1: “Therefore, we define the elements of the adjacency matrix Ad as: “
Claim 18:
The limitations of claim 18 are substantially the same as those of claim 4 except that it depends from claim 15. Therefore this claim is rejected due to the same reasons as outlined above for claims 4 and 15.
Claim 20:
The limitations of claim 20 are substantially the same as those of claim 6 except that it depends from claim 18. Therefore this claim is rejected due to the same reasons as outlined above for claims 6 and 18.
Potentially Allowable Subject Matter
Claims 5 and 19 are objected to as being dependent upon a rejected base claim, but may be potentially allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to AHMAD HUSSAM SHALABY whose telephone number is (571)272-7414. The examiner can normally be reached Mon-Fri 7:30am - 5pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Emerson Puente can be reached at 5712723652. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/A.H.S./Examiner, Art Unit 2187
/BRIAN S COOK/Primary Examiner, Art Unit 2187
Prosecution Insights