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 .
Status of Application
This action is in reply to the reply received October 24, 2025 (hereinafter “Reply”).
Claims 1, 3, 5-8, and 10-14 are amended.
Claim 2 is cancelled.
Claim 16 is new.
Claims 1 and 3-16 are pending.
Claim Rejections - 35 U.S.C. § 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 16 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 pre-AIA the applicant regards as the invention.
Claim 16: The term key in claim 16’s limitations wherein the vector retains key aspects of the patient data type of the patient data record is a relative term that renders the claims indefinite. The term key is not defined by the claims, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention.
Claim Rejections - 35 U.S.C. § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1 and 3-16 are rejected under 35 U.S.C. § 101 because the claimed invention is directed to non-statutory subject matter. Claims 1-15 are directed to an abstract idea without significantly more as required by the Alice test as discussed below.
Step 1
Claims 1 and 3-16 are directed to a process, machine, manufacture, or composition of matter.
Step 2A
Claims 1 and 3-16 are directed to abstract ideas, as explained below.
Prong one of the Step 2A analysis requires identifying the specific limitation(s) in the claim under examination that the examiner believes recites an abstract idea; and determining whether the identified limitation(s) falls within at least one of the groupings of abstract ideas of mathematical concepts, mental processes, and certain methods of organizing human activity.
The claims recite the following limitations that are directed to abstract ideas. Claim 1 recites receiving a set of input data comprising a plurality of patient data records, wherein the plurality of patient data records comprise medical imaging data and at least one other patient data type; selecting an encoding algorithm from a plurality of encoding algorithms for each of the plurality of patient data records, wherein the encoding algorithm is selected based on a patient data type of a plurality of patient data types, wherein the medical imaging data and the at least one other patient data type are included in the plurality of patient data types; generating a vector for each of the plurality of patient data records by processing each patient data record using the encoding algorithm selected for each of the plurality of patient data records; analyzing the vector for each of the plurality of patient data records with a machine learning model; and providing from the machine learning model an output containing information relating to a patient’s health. Claims 3-14 and 16 further specify features of these abstract ideas or characteristics of the data used thereby.
Examiner notes that the claimed machine learning model refers to a class of algorithms are capable of being implemented by a machine, but need not be. To the extent that machine learning model (as used in the claim) refers to aspects of the algorithm(s) actually being implemented by a machine, this nuance will be addressed below when identifying and considering additional elements of the claims.
These limitations describe abstract ideas that correspond to concepts identified as abstract ideas by the courts as mathematical concepts—such as mathematical relationships, mathematical formulas or equations, and mathematical calculations—because the claimed features identified above (including the mathematical models employed by the machine learning model) are mathematical relationships, mathematical formulas or equations, and mathematical calculations.
These limitations describe abstract ideas that correspond to concepts identified as abstract ideas by the courts as mental processes—such as concepts performed in the human mind (including an observation, evaluation, judgment, or opinion)—because the claimed features identified above are concepts performed in the human mind (including an observation, evaluation, judgment, or opinion).
These limitations describe abstract ideas that correspond to concepts identified as abstract ideas by the courts as certain methods of organizing human activity—such as fundamental economic principles or practices (including hedging, insurance, mitigating risk), commercial or legal interactions (including agreements in the form of contracts; legal obligations; advertising, marketing or sales activities or behaviors; business relations), managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions)—because the claim features identified above manage personal behavior or relationships or interactions between people including following rules or instructions.
Thus, the concepts set forth in claims 1 and 3-16 recite abstract ideas.
Prong two of the Step 2A requires identifying whether there are any additional elements recited in the claim beyond the judicial exception(s), and evaluating those additional elements to determine whether they integrate the exception into a practical application of the exception. “Integration into a practical application” requires an additional element or a combination of additional elements in the claim to apply, rely on, or use the judicial exception in a manner that imposes a meaningful limit on the judicial exception, such that the claim is more than a drafting effort designed to monopolize the exception. Further, “integration into a practical application” uses the considerations laid out by the Supreme Court and the Federal Circuit to evaluate whether the judicial exception is integrated into a practical application, such as considerations discussed in M.P.E.P. § 2106.05(a)-(h).
The claims recite the following additional elements beyond those identified above as being directed to an abstract idea. Claim 1 recite s that its method is computer implemented and a machine learning model (as noted above, this feature is being considered insofar as the claim requires a machine to implement the machine learning model algorithms or calculations). Claim 15 recites a memory and one or more processors.
The identified judicial exception(s) are not integrated into a practical application for the following reasons.
First, evaluated individually, the additional elements do not integrate the identified abstract ideas into a practical application. The additional computer elements identified above—the memory and one or more processors, computer, and machine—are recited at a high level of generality. Inclusion of these elements amounts to mere instructions to implement the identified abstract ideas on a computer. See M.P.E.P. § 2106.05(f). To the extent that the claims transform data, the mere manipulation of data is not a transformation. See M.P.E.P. § 2106.05(c). Inclusion of computing system in the claims amounts to generally linking the use of the judicial exception to a particular technological environment or field of use. See M.P.E.P. § 2106.05(h). Thus, taken alone, the additional elements do not amount to significantly more than a judicial exception.
Second, evaluating the claim limitations as an ordered combination adds nothing that is not already present when looking at the elements taken individually. There is no indication that the combination of elements improves the functioning of a computer or improves any other technology. See M.P.E.P. § 2106.05(a). Their collective functions merely provide an implementation of the identified abstract ideas on a computer system in the general field of use of medical diagnosis. See M.P.E.P. § 2106.05(h).
Thus, claims 1 and 3-16 recite mathematical concepts, mental processes, or certain methods of organizing human activity without including additional elements that integrate the exception into a practical application of the exception.
Accordingly, claims 1 and 3-16 are directed to abstract ideas.
Step 2B
Claims 1 and 3-16 do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the additional elements, when considered both individually and as an ordered combination, do not amount to significantly more than the abstract idea.
The analysis above describes how the claims recite the additional elements beyond those identified above as being directed to an abstract idea, as well as why identified judicial exception(s) are not integrated into a practical application. These findings are hereby incorporated into the analysis of the additional elements when considered both individually and in combination. There are no additional features to consider below.
Evaluated individually and for the same reasons above presented in the discussion regarding Step 2A, prong two, the additional elements do not amount to significantly more than a judicial exception. Thus, taken alone, the additional elements do not amount to significantly more than a judicial exception.
Evaluating the claim limitations as an ordered combination adds nothing that is not already present when looking at the elements taken individually. In addition to the factors discussed regarding Step 2A, prong two, there is no indication that the combination of elements improves the functioning of a computer or improves any other technology. Their collective functions merely amount to mere instructions to implement the identified abstract ideas on a computer.
Thus, claims 1 and 3-16, taken individually and as an ordered combination of elements, are not directed to eligible subject matter since they are directed to an abstract idea without significantly more.
Claim Rejections - 35 U.S.C. § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. § 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 1 and 3-16 are rejected under 35 U.S.C. § 102(a)(1)-(2) as being anticipated by Arcot Desai et al. (U.S. Pub. No. 2020/0272857 A1) (hereinafter “Arcot”).
Claim 1: Arcot, as shown, discloses the following limitations:
receiving a set of input data comprising a plurality of patient data records, wherein the plurality of patient data records comprise medical imaging data and at least one other patient data type (see at least ¶ [0034]: an EEG record corresponding to electrical activity of a patient’s brain may be visualized in the form of a time series waveform image; see also at least ¶ [0035]: while the methods and systems disclosed herein are primarily described with reference to EEG records in the form of time series waveform images, other forms of EEG records may be used. For example, EEG records in the form of spectrograms may be processed by the methods and systems. Furthermore, while the methods and systems disclosed in Arcot are primarily described with reference to records comprising electrical activity of the brain, it will be appreciated that other physiological information and non-physiological information may be processed; see also at least ¶ [0036]: other types or modalities of physiological information included in a dataset besides electrical activity of the brain may be included in a dataset. For example, records or files of other modalities of physiological information in a dataset may include measurements of pH levels in neural tissue, blood oxygen levels in neural tissue, neurotransmitters concentrations in neural tissue, heart rate, blood pressure, blood glucose levels, hormones sensed in sweat, accelerometer recordings, and sleep patterns; see also at least ¶ [0037]: with respect to non-physiological information, a dataset may include records or files of patient demographics (e.g., age, gender), patient drug regimen (e.g., type of drug, dose, and time of day of dose), and patient clinical outcomes, such as the rate of electrographic seizure detection and electrographic seizure onset (e.g., as detected and recorded by the implanted neurostimulation system), the rate of clinical seizures (e.g., as reported in a seizure diary or detected based on accelerometer recordings)); and
selecting an encoding algorithm from a plurality of encoding algorithms for each of the plurality of patient data records, wherein the encoding algorithm is selected based on a patient data type of a plurality of patient data types, wherein the medical imaging data and the at least one other patient data type are included in the plurality of patient data types (see at least ¶ [0069]: in one filtering application, EEG records are processed in groups based on their associated triggering event. For example, EEG records having a “scheduled” trigger may sent through the deep learning model, the dimensionality reduction and the clustering algorithm processes of the records classification processor 104 first, followed by EEG records having a “long episode” trigger, followed by EEG records having a “saturation” trigger. In another filtering application, all EEG record trigger types are sent through the deep learning model and the dimensionality reduction processes of the records classification processor 104, with the clustering being performed separately on the feature vectors resulting from the dimensionality reduction based on their associated triggering event—i.e., a patient data type; see also at least ¶ [0080]: other types of models may be used to extract features. For example, a deep learning model may be trained from scratch on a relevant problem (for example, a deep learning model may be trained to classify different types of EEG records) and applied to this problem for feature extraction. Alternatively, handcrafted features such as spectral power, Fast Fourier Transform or wavelets, may be extracted from the EEG records; see also at least ¶¶ [0034]-[0042], [0069], and [0086].
Applicant’s specification provides examples of different types of data types including “e.g. specific to medical imaging data or a specific type of medical imaging data, or non-medical imaging data or a specific type of non-imaging medical data.”);
generating a vector for each of the plurality of patient data records by processing each patient data record using the encoding algorithm selected for each of the plurality of patient records (see at least ¶¶ [0069] and [0080] and the analysis above; see also at least ¶ [0038]: regardless of the type or modality of physiological records used by the method and system, a deep learning model is applied to each physiological record to extract features from that record and provide a multi-dimensional feature vector. While the exact nature or characteristics of the features extracted from the physiological records by the deep learning model are not entirely understood, the features are believed to include hierarchically filtered versions of the data forming the record. The deep learning model may be, for example, a pretrained convolution neural network (CNN), autoencoders, recurrent neural network (RNN), or a deep neural network configured to derive features from the physiological records; see also at least ¶ [0041]: each feature vector outcome from the deep learning model typically contains thousands of rows, where each row corresponds to a feature extracted from the record by the deep learning model. Each of these large-scale, multi-dimensional feature vectors is then reduced to a smaller dimensional feature vector that includes a plurality of different features extracted from the physiological records. For example, the multi-dimensional feature vector may be reduced to a two-dimensional feature vector using a two-step process, where principal component analysis (sklearn.decomposition.PCA) is used to reduce the number of dimensions from a large number to a more manageable number; and then t-distributed stochastic gradient descent or t-distributed stochastic gradient neighbor embedding (sklearn.manifold.TSNE) is used to further reduce the number of dimensions to two; see also at least ¶ [0042]: a similarities algorithm is then applied to the feature vectors developed from the deep learning algorithm and the dimensional reduction algorithm to identify one or more clusters of similar physiological records; see also at least ¶¶ [0039] and [0086]).
analyzing the vector for each of the plurality of patient data records with a machine learning model (see at least ¶ [0038]; see also at least ¶ [0041]: each feature vector outcome from the deep learning model typically contains thousands of rows, where each row corresponds to a feature extracted from the record by the deep learning model. Each of these large-scale, multi-dimensional feature vectors is then reduced to a smaller dimensional feature vector that includes a plurality of different features extracted from the physiological records. For example, the multi-dimensional feature vector may be reduced to a two-dimensional feature vector using a two-step process, where principal component analysis (sklearn.decomposition.PCA) is used to reduce the number of dimensions from a large number to a more manageable number; and then t-distributed stochastic gradient descent or t-distributed stochastic gradient neighbor embedding (sklearn.manifold.TSNE) is used to further reduce the number of dimensions to two; see also at least ¶ [0042]: a similarities algorithm is then applied to the feature vectors developed from the deep learning algorithm and the dimensional reduction algorithm to identify one or more clusters of similar physiological records; see also at least ¶¶ [0039]); and
providing from the machine learning model an output containing information relating to a patient’s health (see at least ¶ [0067]: the records classification processor 104 also interfaces with a display 210 to enable the display of EEG records, the display of clusters of such records, and the display of classification labels. The records classification processor 104 also interfaces with a user interface 212 to receive inputs from expert users; see also at least ¶ [0091]: the records classification processor 104 provides an output comprising information that enables a display of one or more of the plurality of clusters. To this end, the similarities module 204 provides clustering results 218 to the labeling module 206. The labeling module 206 includes a cluster image module 220 that provides an output to the display 210 that enables a visual display based on the clustering results. For example, the output may enable the display 210 to display all of the clusters formed by the similarities module 204, such as shown in FIG. 8. In other configurations, the output may enable the display 210 to display a subset of the clusters; see also at least ¶¶ [0068] and [0086]).
Claim 3: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
combining the vectors prior to analyzing the vectors (see at least ¶ [0041]: each feature vector outcome from the deep learning model typically contains thousands of rows, where each row corresponds to a feature extracted from the record by the deep learning model. Each of these large-scale, multi-dimensional feature vectors is then reduced to a smaller dimensional feature vector that includes a plurality of different features extracted from the physiological records. For example, the multi-dimensional feature vector may be reduced to a two-dimensional feature vector using a two-step process, where principal component analysis (sklearn.decomposition.PCA) is used to reduce the number of dimensions from a large number to a more manageable number; and then t-distributed stochastic gradient descent or t-distributed stochastic gradient neighbor embedding (sklearn.manifold.TSNE) is used to further reduce the number of dimensions to two; see also at least ¶¶ [0080]: the deep learning model 404 used for feature extraction may be, for example, a pretrained convolution neural network (CNN), autoencoders, recurrent neural network (RNN), or a deep neural network configured to derive features from the records; see also at least ¶ [0086]: the foregoing process may be repeated multiple times, once for all EEG records collected from a given patient, to provide one feature vector or data point for each EEG record in the reduced two-dimensional feature space).
Claim 4: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
wherein combining the vectors comprises combining the vectors to form a one-dimensional or multidimensional vector (see at least ¶ [0041]: each feature vector outcome from the deep learning model typically contains thousands of rows, where each row corresponds to a feature extracted from the record by the deep learning model. Each of these large-scale, multi-dimensional feature vectors is then reduced to a smaller dimensional feature vector that includes a plurality of different features extracted from the physiological records. For example, the multi-dimensional feature vector may be reduced to a two-dimensional feature vector using a two-step process, where principal component analysis (sklearn.decomposition.PCA) is used to reduce the number of dimensions from a large number to a more manageable number; and then t-distributed stochastic gradient descent or t-distributed stochastic gradient neighbor embedding (sklearn.manifold.TSNE) is used to further reduce the number of dimensions to two; see also at least ¶¶ [0080]: the deep learning model 404 used for feature extraction may be, for example, a pretrained convolution neural network (CNN), autoencoders, recurrent neural network (RNN), or a deep neural network configured to derive features from the records; see also at least ¶ [0086]: the foregoing process may be repeated multiple times, once for all EEG records collected from a given patient, to provide one feature vector or data point for each EEG record in the reduced two-dimensional feature space).
Claim 5: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
wherein the output is a prediction pertaining to a diagnosis, an onset and/or a progression of a disease for the patient (see at least ¶ [0042]: a similarities algorithm is then applied to the feature vectors developed from the deep learning algorithm and the dimensional reduction algorithm to identify one or more clusters of similar physiological records; see also at least ¶¶ [0039] and [0086]; see also at least ¶ [0079]: a deep learning model 404 is applied to the EEG record 402 to extract features. Assuming, the aim of the records labeling process is to assign a label to the EEG record as a whole (as opposed to each individual channel record), each sub-record of the EEG record 402 is run through the deep learning model 404, and the features extracted for each sub-record are concatenated; see also at least ¶ [0096]: the records classification processor 104 automatically assigns the expert label to the EEG records corresponding to the remaining feature vectors in the selected cluster. In cases where only one feature vector is selected in block 308, the label assignment module 222 is configured to automatically associate or link the expert label with the EEG record of each feature vector in the cluster based only on the expert labeling of the one selected EEG record; see also at least ¶ [0097]).
Claim 6: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
wherein the machine learning model is a recursive model and/or recurrent model (see at least ¶ [0038]: the deep learning model may be, for example, a pretrained convolution neural network (CNN), autoencoders, recurrent neural network (RNN), or a deep neural network configured to derive features from the physiological records).
Claim 7: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
wherein the plurality of patient data records comprise patient data records obtained at a plurality of points in time (see at least ¶ [0034]: an EEG record corresponding to electrical activity of a patient's brain may be visualized in the form of a time series waveform image. For example, with reference to FIG. 1A, an EEG record 122 may consist of four channels of EEG data, each visualized as a separate time series waveform 124 a, 124 b, 124 c, 124 d. These four separate time series waveforms 124 a, 124 b, 124 c, 124 d (and their corresponding EEG data) collectively represent the EEG record 122 and may be individually referred to as sub-records of the EEG record. Depending on the granularity of expert labeling desired, the systems and methods disclosed herein may process the EEG record 122 as a whole (if the aim is to assign a label to the whole EEG record) or may process at the individual sub-record level (if the aim is to assign a label to each sub-record). In the latter case, each of the sub-records may be considered an individual EEG record; see also at least ¶ [0056]: physiological records may have a tag that indicates the basis, e.g., seizure detection, magnet swipe, scheduled time of day, for preserving the record. These tags allow a set of physiological records to be selected for processing based on a single criterion or a combination of criteria. Other tags may include day of capture, area of the brain at which the electrical activity was captured, basis for record creation (e.g., seizure detection, scheduled, patient initiated), characteristic of the record (e.g., power spectral density of EEG signal prior to stimulation; see also at least ¶ [0086]: the foregoing process may be repeated multiple times, once for all EEG records collected from a given patient, to provide one feature vector or data point for each EEG record in the reduced two-dimensional feature space; see also at least ¶¶ [0054] and [0117]), and
wherein analyzing the vectors using the machine learning model comprises applying a weighting to each patient data record based on a point in time that the record was created (see at least ¶ [0038]: regardless of the type or modality of physiological records used by the method and system, a deep learning model is applied to each physiological record to extract features from that record and provide a multi-dimensional feature vector. While the exact nature or characteristics of the features extracted from the physiological records by the deep learning model are not entirely understood, the features are believed to include hierarchically filtered versions of the data forming the record. The deep learning model may be, for example, a pretrained convolution neural network (CNN), autoencoders, recurrent neural network (RNN), or a deep neural network configured to derive features from the physiological records; see also at least ¶ [0041]: each feature vector outcome from the deep learning model typically contains thousands of rows, where each row corresponds to a feature extracted from the record by the deep learning model. Each of these large-scale, multi-dimensional feature vectors is then reduced to a smaller dimensional feature vector that includes a plurality of different features extracted from the physiological records. For example, the multi-dimensional feature vector may be reduced to a two-dimensional feature vector using a two-step process, where principal component analysis (sklearn.decomposition.PCA) is used to reduce the number of dimensions from a large number to a more manageable number; and then t-distributed stochastic gradient descent or t-distributed stochastic gradient neighbor embedding (sklearn.manifold.TSNE) is used to further reduce the number of dimensions to two; see also at least ¶ [0042]: a similarities algorithm is then applied to the feature vectors developed from the deep learning algorithm and the dimensional reduction algorithm to identify one or more clusters of similar physiological records; see also at least ¶ [0086]. The different processing of these values weights the inputs differently to arrive at the outputs).
Claim 8: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
wherein the machine learning model is trained using medical data duplets, each medical data duplet comprising:
a plurality of patient data records of a respective patient (see at least ¶ [0034]: an EEG record corresponding to electrical activity of a patient’s brain may be visualized in the form of a time series waveform image; see also at least ¶ [0035]: while the methods and systems disclosed herein are primarily described with reference to EEG records in the form of time series waveform images, other forms of EEG records may be used. For example, EEG records in the form of spectrograms may be processed by the methods and systems. Furthermore, while the methods and systems disclosed in Arcot are primarily described with reference to records comprising electrical activity of the brain, it will be appreciated that other physiological information and non-physiological information may be processed; see also at least ¶ [0036]: other types or modalities of physiological information included in a dataset besides electrical activity of the brain may be included in a dataset. For example, records or files of other modalities of physiological information in a dataset may include measurements of pH levels in neural tissue, blood oxygen levels in neural tissue, neurotransmitters concentrations in neural tissue, heart rate, blood pressure, blood glucose levels, hormones sensed in sweat, accelerometer recordings, and sleep patterns; see also at least ¶ [0037]: with respect to non-physiological information, a dataset may include records or files of patient demographics (e.g., age, gender), patient drug regimen (e.g., type of drug, dose, and time of day of dose), and patient clinical outcomes, such as the rate of electrographic seizure detection and electrographic seizure onset (e.g., as detected and recorded by the implanted neurostimulation system), the rate of clinical seizures (e.g., as reported in a seizure diary or detected based on accelerometer recordings)); and
disease diagnosis and/or progression data for the patient (see at least ¶ [0041]: Each feature vector outcome from the deep learning model typically contains thousands of rows, where each row corresponds to a feature extracted from the record by the deep learning model. Each of these large-scale, multi-dimensional feature vectors is then reduced to a smaller dimensional feature vector that includes a plurality of different features extracted from the physiological records. For example, the multi-dimensional feature vector may be reduced to a two-dimensional feature vector using a two-step process, where principal component analysis (sklearn.decomposition.PCA) is used to reduce the number of dimensions from a large number to a more manageable number; and then t-distributed stochastic gradient descent or t-distributed stochastic gradient neighbor embedding (sklearn.manifold.TSNE) is used to further reduce the number of dimensions to two. An example of a dimensionality reduction algorithm is disclosed in Dermatologist-level classification of skin cancer with deep neural networks, by Andre Esteva et al., Nature, published Feb. 2, 2017, Volume 542, pp 115-118, which is herein incorporated by reference. See also, S. A. Desai, T. Tcheng, and M. Morrell, “Transfer-learning for differentiating epileptic patients who respond to treatment based on chronic ambulatory ECoG data,” in 2019 9th International IEEE/EMBS Conference on Neural Engineering (NER), 2019: IEEE, pp. 1-4; see also at least ¶ [0105]: the user will be able to quickly verify whether the labels automatically assigned to the top nine EEG records in the sorted list are accurate and move on to the next display EEG records. In the last validation page illustrated in FIG. 9C, it can be seen that EEG record #99 and EEG record #100 do not look very much like the rest of the EEG records in this cluster. Accordingly, the user will be able to select these EEG records and reclassify/relabel them as something else. For example, in the case of FIG. 9C, the last two EEG records may be labeled as interictal epileptiform or baseline EEG records).
Claim 9: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
wherein the at least one other patient data type is non-medical imaging data relating to a patient (see also at least ¶ [0036]: other types or modalities of physiological information included in a dataset besides electrical activity of the brain may be included in a dataset. For example, records or files of other modalities of physiological information in a dataset may include measurements of pH levels in neural tissue, blood oxygen levels in neural tissue, neurotransmitters concentrations in neural tissue, heart rate, blood pressure, blood glucose levels, hormones sensed in sweat, accelerometer recordings, and sleep patterns; see also at least ¶ [0037]: with respect to non-physiological information, a dataset may include records or files of patient demographics (e.g., age, gender), patient drug regimen (e.g., type of drug, dose, and time of day of dose), and patient clinical outcomes, such as the rate of electrographic seizure detection and electrographic seizure onset (e.g., as detected and recorded by the implanted neurostimulation system), the rate of clinical seizures (e.g., as reported in a seizure diary or detected based on accelerometer recordings).
Claim 10: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
wherein the non-image medical data comprises physiological data (see also at least ¶ [0036]: other types or modalities of physiological information included in a dataset besides electrical activity of the brain may be included in a dataset. For example, records or files of other modalities of physiological information in a dataset may include measurements of pH levels in neural tissue, blood oxygen levels in neural tissue, neurotransmitters concentrations in neural tissue, heart rate, blood pressure, blood glucose levels, hormones sensed in sweat, accelerometer recordings, and sleep patterns; see also at least ¶ [0037]: with respect to non-physiological information, a dataset may include records or files of patient demographics (e.g., age, gender), patient drug regimen (e.g., type of drug, dose, and time of day of dose), and patient clinical outcomes, such as the rate of electrographic seizure detection and electrographic seizure onset (e.g., as detected and recorded by the implanted neurostimulation system), the rate of clinical seizures (e.g., as reported in a seizure diary or detected based on accelerometer recordings).
Claim 11: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
wherein non-medical imaging data is in the form of at least one of text-based data, signal data, and tabular data (see also at least ¶ [0036]: other types or modalities of physiological information included in a dataset besides electrical activity of the brain may be included in a dataset. For example, records or files of other modalities of physiological information in a dataset may include measurements of pH levels in neural tissue, blood oxygen levels in neural tissue, neurotransmitters concentrations in neural tissue, heart rate, blood pressure, blood glucose levels, hormones sensed in sweat, accelerometer recordings, and sleep patterns; see also at least ¶ [0037]: with respect to non-physiological information, a dataset may include records or files of patient demographics (e.g., age, gender), patient drug regimen (e.g., type of drug, dose, and time of day of dose), and patient clinical outcomes, such as the rate of electrographic seizure detection and electrographic seizure onset (e.g., as detected and recorded by the implanted neurostimulation system), the rate of clinical seizures (e.g., as reported in a seizure diary or detected based on accelerometer recordings; see also at least ¶ [0095]).
Claim 12: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
wherein generating the vector for each of the plurality of patient data records by processing each patient data record using the corresponding encoding algorithm is carried out using a neural network model, a traditional machine learning model, signal processing and/or a statistical model (see at least ¶ [0038]: the deep learning model may be, for example, a pretrained convolution neural network (CNN), autoencoders, recurrent neural network (RNN), or a deep neural network configured to derive features from the physiological records; see also at least ¶ [0041]: each feature vector outcome from the deep learning model typically contains thousands of rows, where each row corresponds to a feature extracted from the record by the deep learning model. Each of these large-scale, multi-dimensional feature vectors is then reduced to a smaller dimensional feature vector that includes a plurality of different features extracted from the physiological records. For example, the multi-dimensional feature vector may be reduced to a two-dimensional feature vector using a two-step process, where principal component analysis (sklearn.decomposition.PCA) is used to reduce the number of dimensions from a large number to a more manageable number; and then t-distributed stochastic gradient descent or t-distributed stochastic gradient neighbor embedding (sklearn.manifold.TSNE) is used to further reduce the number of dimensions to two; see also at least ¶ [0042]: a similarities algorithm is then applied to the feature vectors developed from the deep learning algorithm and the dimensional reduction algorithm to identify one or more clusters of similar physiological records; see also at least ¶¶ [0039], [0069], [0080], and [0086]).
Claim 13: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
wherein the neural network is a convolutional neural network, a transformer neural network, or a fully connected neural network (see at least ¶ [0038]: regardless of the type or modality of physiological records used by the method and system, a deep learning model is applied to each physiological record to extract features from that record and provide a multi-dimensional feature vector. While the exact nature or characteristics of the features extracted from the physiological records by the deep learning model are not entirely understood, the features are believed to include hierarchically filtered versions of the data forming the record. The deep learning model may be, for example, a pretrained convolution neural network (CNN), autoencoders, recurrent neural network (RNN), or a deep neural network configured to derive features from the physiological records; see also at least ¶ [0041]: each feature vector outcome from the deep learning model typically contains thousands of rows, where each row corresponds to a feature extracted from the record by the deep learning model. Each of these large-scale, multi-dimensional feature vectors is then reduced to a smaller dimensional feature vector that includes a plurality of different features extracted from the physiological records. For example, the multi-dimensional feature vector may be reduced to a two-dimensional feature vector using a two-step process, where principal component analysis (sklearn.decomposition.PCA) is used to reduce the number of dimensions from a large number to a more manageable number; and then t-distributed stochastic gradient descent or t-distributed stochastic gradient neighbor embedding (sklearn.manifold.TSNE) is used to further reduce the number of dimensions to two; see also at least ¶ [0042]: a similarities algorithm is then applied to the feature vectors developed from the deep learning algorithm and the dimensional reduction algorithm to identify one or more clusters of similar physiological records; see also at least ¶¶ [0039] and [0086]).
Claim 14: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
wherein the plurality of patient data records comprise patient data records obtained at a plurality of points in time (see at least ¶ [0034]: an EEG record corresponding to electrical activity of a patient's brain may be visualized in the form of a time series waveform image. For example, with reference to FIG. 1A, an EEG record 122 may consist of four channels of EEG data, each visualized as a separate time series waveform 124 a, 124 b, 124 c, 124 d. These four separate time series waveforms 124 a, 124 b, 124 c, 124 d (and their corresponding EEG data) collectively represent the EEG record 122 and may be individually referred to as sub-records of the EEG record. Depending on the granularity of expert labeling desired, the systems and methods disclosed herein may process the EEG record 122 as a whole (if the aim is to assign a label to the whole EEG record) or may process at the individual sub-record level (if the aim is to assign a label to each sub-record). In the latter case, each of the sub-records may be considered an individual EEG record; see also at least ¶ [0056]: physiological records may have a tag that indicates the basis, e.g., seizure detection, magnet swipe, scheduled time of day, for preserving the record. These tags allow a set of physiological records to be selected for processing based on a single criterion or a combination of criteria. Other tags may include day of capture, area of the brain at which the electrical activity was captured, basis for record creation (e.g., seizure detection, scheduled, patient initiated), characteristic of the record (e.g., power spectral density of EEG signal prior to stimulation; see also at least ¶ [0086]: the foregoing process may be repeated multiple times, once for all EEG records collected from a given patient, to provide one feature vector or data point for each EEG record in the reduced two-dimensional feature space; see also at least ¶¶ [0054] and [0117]), and
wherein generating the vector for each of the plurality of patient data records comprises generating time-resolved feature vectors (see at least ¶ [0056]: physiological records may have a time stamp that allows a set of physiological records at a given point in time to be located for processing; see also at least ¶ [0074]: the plurality of records 214 may or may not have a common parameter or tag, e.g., time stamp, day of capture, area of the brain at which the electrical activity was captured, basis for record creation (e.g., seizure detection, scheduled, patient initiated), or characteristic of the record (e.g., power spectral density of EEG signal prior to stimulation; see also at least ¶¶ [0034], [0056], [0117]).
Claim 15: Arcot discloses the limitations as shown in the rejection above. Further, Arcot, as shown, discloses the following limitations:
a memory comprising instruction data representing a set of instructions (see at least ¶¶ [0131]-[0137]); and
one or more processors configured to communicate with the memory (see at least ¶¶ [0131]-[0137]) and to execute the set of instructions, wherein the set of instructions, when executed by the processor cause the processor to carry out the computer implemented method given in claim 1 (see supra the rejection of claim 1).
Claim 16: Arcot discloses the limitations as shown in the rejections above. Further, Arcot, as shown, discloses the following limitations:
wherein the vector retains key aspects of the patient data type of the patient data record (see at least ¶¶ [0069] and [0080] and the analysis above; see also at least ¶ [0038]: regardless of the type or modality of physiological records used by the method and system, a deep learning model is applied to each physiological record to extract features from that record and provide a multi-dimensional feature vector. While the exact nature or characteristics of the features extracted from the physiological records by the deep learning model are not entirely understood, the features are believed to include hierarchically filtered versions of the data forming the record—i.e., key aspects of the patient data type. The deep learning model may be, for example, a pretrained convolution neural network (CNN), autoencoders, recurrent neural network (RNN), or a deep neural network configured to derive features from the physiological records; see also at least ¶ [0041]: each feature vector outcome from the deep learning model typically contains thousands of rows, where each row corresponds to a feature extracted from the record by the deep learning model—i.e., key aspects of the patient data type. Each of these large-scale, multi-dimensional feature vectors is then reduced to a smaller dimensional feature vector that includes a plurality of different features extracted from the physiological records—i.e., key aspects of the patient data type. For example, the multi-dimensional feature vector may be reduced to a two-dimensional feature vector using a two-step process, where principal component analysis (sklearn.decomposition.PCA) is used to reduce the number of dimensions from a large number to a more manageable number; and then t-distributed stochastic gradient descent or t-distributed stochastic gradient neighbor embedding (sklearn.manifold.TSNE) is used to further reduce the number of dimensions to two; see also at least ¶ [0042]: a similarities algorithm is then applied to the feature vectors developed from the deep learning algorithm and the dimensional reduction algorithm to identify one or more clusters of similar physiological records; see also at least ¶¶ [0039] and [0086].
Response to Arguments
The arguments submitted with the Reply have been fully considered but are not persuasive.
Arguments regarding the Rejections Under 35 U.S.C. § 101
Applicant argues that the claims do not recite a mathematical concept because they are similar to the decision in XY, LLC, claim 1. Reply, p. 7. Examiner disagrees that the present claims can be analyzed in the same manner, because the examples cited in XY, LLC are not the same type of operations recited in the present claims.
Applicant argues that the “processing” and “analyzing” steps of the claims “cannot practically be performed in the human mind, nor has the Office provided any support for such an assertion.” Reply, p. 8. Examiner disagrees. As noted in the rejections, the claimed machine learning model refers to a class of algorithms are capable of being implemented by a machine, but need not be. For example, the machine learning model can be a simple decision tree that adds a rule for every instance seen, similar to how a person mentally keep track of instances. To employ such a model, one would simply refer to the rules: for example, “If the vector has a value corresponding to a tumor, then the patient is sick,” and the vector has a value corresponding to a tumor, then the output is that the patient is sick. The claims do not limit the generically-recited machine learning model to anything more complicated than this example, which can be performed by a human (e.g., by a doctor).
Applicant argues that “claim 1 does not recite any feature that could be interpreted managing personal behavior of any patient or user, nor indicate any interactions between people. Further, none of the features of claim 1 describe people following rule or instructions.” Reply, p. 9. Examiner disagrees. As demonstrated in the previous paragraph, the claims include activities such as reading lab results and following rules to provide a diagnosis, similar to how a doctor would when interacting with a patient.
Applicant argues that the claims include an improvement to “computer implemented methods for analyzing heterogeneous data” because allegedly, these “data are usually very complex and heterogenous.” Reply, p. 11. Applicant presents similar arguments regarding Step 2B of the Alice test. Reply, p. 13. Examiner disagrees, because formulating an allegedly improved way of analyzing data (e.g., by providing a new type of statistical analysis), would be an alleged improvement to medical knowledge or data sciences—not to a particular or technological field. In other words, the alleged improvement would be to the abstract idea, not anything relating to the technical aspects of the claimed invention. See SAP Am., Inc. v. InvestPic, LLC, No. 2017-2081, slip op. at 14 (Fed. Cir. Aug. 2, 2018) (“What is needed is an inventive concept in the non-abstract application realm. … [L]imitation of the claims to a particular field of information … does not move the claims out of the realm of abstract ideas.”). Moreover, “[A] claim for a new abstract idea is still an abstract idea.” Synopsys, Inc. v. Mentor Graphics Corp., 839 F.3d 1138, 1151 (Fed. Cir. 2016) (emphasis added). “[U]nder the Mayo/Alice framework, a claim directed to a newly discovered law of nature (or natural phenomenon or abstract idea) cannot rely on the novelty of that discovery for the inventive concept necessary for patent eligibility ….” Genetic Techs. Ltd. v. Merial L.L.C., 818 F.3d 1369, 1376 (Fed. Cir. 2016) (citations omitted). Further, the claims do not appear to reflect the alleged improvement; the claims to not appear to operate on any overly “complex and heterogeneous” information. Instead, refer to the information as a plurality of patient records with medical imaging data and at least one other patient data type—i.e., two pieces of information. And unlike in Ex Parte Desjardins, machine learning or artificial intelligence does not appear to be improved.
Arguments regarding the Rejections Under 35 U.S.C. § 102
Applicant argues that Arcot does not disclose the features presented via the amendments in the independent claims. Reply, p. 14. Examiner disagrees for the reasons presented in the rejections above. More particularly, ¶ [0069] of Arcot describes how EEG records are processed in groups based on their associated triggering event, as well as at ¶ [0080] that other types of models may be used to extract features. For example, a deep learning model may be trained from scratch on a relevant problem (for example, a deep learning model may be trained to classify different types of EEG records) and applied to this problem for feature extraction. Both of these portions describe how patient data types influence which deep learning model—i.e., encoding algorithm—is selected for and based on the type of input data—i.e., based on a patient data type.
Conclusion
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.
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure. The following references have been cited to further show the state of the art with respect to analyzing medical imaging.
Zhang et al. (U.S. Pub. No. 2022/0147818 A1) (auxiliary model for predicting new model parameters);
Parmar et al. (“Data analysis strategies in medical imaging.” Clinical cancer research 24.15 (2018): 3492-3499).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Christopher Tokarczyk, whose telephone number is 571-272-9594. The examiner can normally be reached Monday-Thursday between 6:00 AM and 4:00 PM Eastern.
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, Mamon Obeid, can be reached at 571-270-1813. 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.
/CHRISTOPHER B TOKARCZYK/ Primary Examiner, Art Unit 3687