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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 10 March 2026 has been entered.
Claim Rejections - 35 USC § 103
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claim(s) 1, 6-10, 13, and 15-23 is/are rejected under 35 U.S.C. 103 as being unpatentable over Wang et al. (US Patent No. 10,398,319, published Sept. 2019, hereinafter “Wang”) in view of Kabrams et al. (US Pub. No. 2020/0194120, published June 2020, hereinafter “Kabrams”), further in view of Chan et al. (US Pub. No. 2021/0282701, filed March 2021, hereinafter “Chan”).
Regarding claim 1, Wang teaches a method for detecting or forecasting a seizure in measurement data recorded with a wearable device worn [[by]] on a limb of a subject. the method comprising:
(a) recording measurement data with the wearable device worn on the limb of the subject, wherein the measurement data comprise at least one of motion data, blood volume pulse data. electrodermal activity data, temperature data, heart rate data, or time of day (Wang, Page 13, Col. 3, Lines 54-61 – “FIG. 1 is a diagram of an implementation of a wearable system 100 which can include a wearable apparatus 110 worn by an individual and a device core 120. The wearable apparatus 110 can include a housing that is in the form of a ring, bracelet, wristband, pendant, armband, anklet, headband, belt, necklace, glove, a chest patch, or other mechanism for securing or attaching the wearable apparatus 110 to a human body.” and in Page 18, Col. 13 Line 63 - Col. 14 Line 11 – “At operation 810, one or more feature vectors is generated including at least one of: an EEG feature vector based on data based on EEG signals such as the aggregate EEG data; an ECG feature vector based on data including ECG signals such as the aggregate ECG data; and a motion sensor feature vector based on data including motion sensor data such as the aggregate motion sensor data. In an implementation, the generation of the one or more feature vectors can be performed: on a local device, such as the wearable apparatus 200 or the computing device 300; or in an offline learning network. The one or more feature vectors can include attributes and corresponding values. For example, an ECG feature vector can include heart rate values and other features of ECG signals. Further, adverse physiological event features such as epilepsy related features (e.g., high frequency power) can be extracted from the EEG signals.” – teaches recording measurement data with the wearable device worn on the limb of the subject (wearable apparatus 110 worn by individual, can include housing in form of ring, anklet, bracelet, armband, etc., all of which are worn on the limb of the subject), wherein the data comprises motion data (motion sensor data) and heart rate data (heart rate values).);
(b) accessing a trained machine learning algorithm with a computer system, wherein the trained machine learning algorithm has been trained on training data in order to monitor a likelihood of a seizure event occurring within signals contained in the measurement data (Wang, Page 16, Col. 10 Lines 29-37 – “the training of the previously trained classification model can include receiving input of EEG signals, ECG signals, and motion sensor signals that are indicative of an adverse physiological event including at least one of: a cardiovascular event such as a myocardial infarction; a neurological event such as an epileptic seizure; an ischemic event such as an ischemic stroke; a central nervous system event; and a hemorrhagic event (also known as a haemorrhagic event) such as a hemorrhagic stroke.” – teaches accessing the trained machine learning algorithm (previously trained classification model), wherein the trained machine learning algorithm has been trained on training data to monitor a likelihood of a seizure event occurring within signals contained in the measurement data (training includes receiving input of EEG signals, ECG signals, and motion sensor signals indicative of a physiological event including epileptic seizure));
(c) transmitting the measurement data from the wearable device to the computer system (Wang, Page 13, Col. 4, Lines 15-20 – “Further, the wearable apparatus 110 can exchange (send or receive) data from a remote data source. For example, a health profile of a user, including measurements (e.g., ECG measurements) from the wearable apparatus 110, can be sent to a remote cloud server where the measurements can be stored for later retrieval and use.” and in Page 20, Col. 17 Lines 7-11 - “ECG data generated at a wearable device can be added to locally stored aggregate ECG data or sent to a remote computing device for addition to aggregate ECG data stored on the remote computing device” – teaches transmitting the measurement data from the wearable device to the computer system (wearable apparatus can exchange data from remote data source, such as remote server, thus transmitting data from the wearable device to a computer system)); and
wherein the first training data comprise electroencephalography (EEG) data acquired from subjects (Wang, Fig. 8B and Pg. 16, Col. 10, Lines 19-28 – “the adverse physiological event classifier can be based on a previously trained classification model that was trained using one or more machine learning techniques and data including the aggregate EEG data, the aggregate ECG data, or the aggregate motion sensor data. Further, the training of the classification model can be based on data stored on remote computing devices including cloud computing devices which can extract features from data gathered from a plurality of sources such as EEG data, EEG data, and motion sensor data from large population groups.” – teaches wherein the first training data comprises EEG data acquired from subjects and second training data acquired from wearable device data (previously trained model using EEG data, model can be further trained using stored data that was acquired from wearable device as in Wang at Page 13, Col. 4, Lines 15-20)) and the second training data comprise wearable device data comprising non-EEG physiological signals acquired from limbs of subjects (Wang, Page 13, Col. 3, Lines 54-61 – “FIG. 1 is a diagram of an implementation of a wearable system 100 which can include a wearable apparatus 110 worn by an individual and a device core 120. The wearable apparatus 110 can include a housing that is in the form of a ring, bracelet, wristband, pendant, armband, anklet, headband, belt, necklace, glove, a chest patch, or other mechanism for securing or attaching the wearable apparatus 110 to a human body.”, Pg. 13, Col. 3, Lines 37-47 – “The disclosed technology can leverage big data and deep learning networks to analyze the relationship between abnormal electrical activity in the brain (such as can be detected by an EEG) and other physical indicators such as electrical cardiac activity that can be monitored by an ECG. Further, the disclosed technology can monitor ECG signals, motion sensor signals, and other signals on a continuous basis. In this way a comprehensive snapshot of a user's physiological state can be generated and an imminent adverse physiological event, such as an epileptic seizure, can be more effectively detected.” , and in Page 16, Col. 10 Lines 29-37 – “the training of the previously trained classification model can include receiving input of EEG signals, ECG signals, and motion sensor signals that are indicative of an adverse physiological event including at least one of: a cardiovascular event such as a myocardial infarction; a neurological event such as an epileptic seizure; an ischemic event such as an ischemic stroke; a central nervous system event; and a hemorrhagic event (also known as a haemorrhagic event) such as a hemorrhagic stroke.” – teaches second training data (previously trained classification model retrained on second training data) comprising wearable device data comprising non-EEG physiological signals acquired from limbs of subjects (wearable apparatus may be worn on limbs of subjects (housing of wearable apparatus may be a ring, armlet, anklet, bracelet, armband, etc.), and acquires non-EEG physiological data such as ECG signals and motion sensor signals to measure a snapshot of patient’s physiological state, thus teaching second training data comprising wearable device data comprising non-EEG physiological data acquired from limbs of subjects))
Wang fails to explicitly teach applying the measurement data to the trained machine learning algorithm with the computer system, generating an output as an indication of at least one of detecting or forecasting a seizure event in the measurement data.
However, analogous to the field of seizure detection and forecasting, Kabrams teaches:
(d) applying the measurement data to the trained machine learning algorithm with the computer system, generating an output as an indication of at least one of detecting or forecasting a seizure event in the measurement data (Kabrams, [0150] – “one or more alerts may be generated by a machine learning algorithm trained to detect and/or predict seizures. For example, the machine learning algorithm may include a deep learning network, e.g., as described with respect to FIGS. 11B and 11C. When the algorithm detects that a seizure is present, or predicts that a seizure is likely to develop in the near future (e.g., within an hour), an alert may be sent to a mobile application.” – teaches applying the measurement data to the trained machine learning algorithm to generate an output as an indication of at least one of detecting or forecasting a seizure event in the measurement data).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the output generation of Kabrams to the recording measurement data and training of Wang in order to detect or forecast a seizure event in measurement data. Doing so would provide benefit from the device being portable and able to measure a variety of physiological signals while also leveraging the advantages of previously generated data (Wang, Pg. 13, Col. 3) and in response to determining that the brain is exhibiting the symptom of the neurological disorder, initiate a response attempting to suppress the symptom of the neurological disorder, e.g., a seizure (Kabrams, [0129]).
The combination of Wang and Kabrams fails to explicitly teach wherein the trained machine learning algorithm is trained on the training data using a multi-stage training process; wherein the multi-stage training process includes training an initial machine learning algorithm on first training data and retraining the initial machine learning algorithm on second training data, generating an output as the trained machine learning algorithm; wherein the first training data comprise non-ambulatory electroencephalography (EEG) data acquired from non-ambulatory subjects; wherein the initial machine learning algorithm is retrained using transfer learning on the second training data.
However, analogous to the field of the claimed invention, Chan teaches:
wherein the trained machine learning algorithm is trained on the training data using a multi-stage training process (Chan, [0032] – “The models in model set 161 may be trained or retrained in the same set up on GPU farm 163, using training and testing data from storage unit 170. When retraining happens as designed or scheduled (as discussed in further detail below), or when desired, GPU farm 163 retrains the early detection models based on the collected scalp EEG data, or other suitable data, to update model set 161.” – teaches wherein the trained machine learning algorithm is trained on the training data using a multi-stage training process).
wherein the multi-stage training process includes training an initial machine learning algorithm on first training data and retraining the initial machine learning algorithm on second training data, generating an output as the trained machine learning algorithm (Chan, [0039] – “Retrained model set 226, which replaces models selected from the previously trained model library, has higher accuracy with respect to detecting and predicting patient 100's seizures, as model set 226 has appropriately incorporated user-specific scalp EEG data in the medical records. At this point, as the data that is used to create model set 226 is primarily collected under special environments (e.g., hospital and medicated conditions), the data inherently includes biases, as already explained in with respect to the prior art data discussed above. To eliminate the biases, the seizure early detection system of the present invention continuously fine-tunes model set 226 based on patient 100's own day-to-day EEG data.” – teaches training an initial machine learning algorithm on first training data (data used to create initial model set 226 is special environment data collected under special environment) and retrained on second training data (patient 100’s own day-to-day EEG data), generating an output as the trained machine learning algorithm (models set 226));
wherein the first training data comprise non-ambulatory electroencephalography (EEG) data acquired from non-ambulatory subjects and the second training data comprise wearable device data acquired from subjects (Chan, [0039] – “At this point, as the data that is used to create model set 226 is primarily collected under special environments (e.g., hospital and medicated conditions), the data inherently includes biases, as already explained in with respect to the prior art data discussed above. To eliminate the biases, the seizure early detection system of the present invention continuously fine-tunes model set 226 based on patient 100's own day-to-day EEG data.” – teaches the first training data comprising non-ambulatory EEG data (collected under special environments, e.g., hospital and medicated conditions) and the second training data comprising wearable device data acquired from subjects (own day-to-day data), as in [0025] – “(i) scalp EEG data acquisition device 110 (“acquisition device 110”), which is worn by patient or user 100 on the head and which continuously collects the patient's scalp EEG data; (ii) mobile device 140 (e.g., a cellular “smartphone”), which patient 100 carries when wearing acquisition device 110, and which acts as a user interface for patient 100 to interact with acquisition device 110 and backend system 150. In one embodiment, mobile device 140 also streams the scalp EEG data collected by acquisition device 110 in real time to backend system 150 for processing;” – teaches the wearable device recording data);
wherein the initial machine learning algorithm is retrained using transfer learning on the second training data (Chan, [0120] – “Transfer learning may be used during training of the neural network model to compensate for the difficulty in model convergence due to the reduced set of channels.” – teaches uses transfer learning during the training of the initial machine learning algorithm, supported by [0101] – “Model set 161 is extracted from model zoo 173 and retrained at step 773 using data 761, 771 or 772, depending on which of the initial phase, the converging phase or the performance phase the seizure early detection system is operating under.” – teaches retraining using a second training data).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the multi-stage training process of Chan to the method of Wang and Kabrams in order to train the machine learning algorithm to better detect or forecast a seizure event occurring in measurement data. Doing so would provide benefit from the device being portable and able to measure a variety of physiological signals while also leveraging the advantages of previously generated data (Wang, Pg. 13, Col. 3), prevent algorithms from developing biases on one modality of data (Chan, [0006]), and improve model performance (Chan, [0033]).
Regarding claim 6, the combination of Wang, Kabrams, and Chan teaches the method of claim [[5]]1, wherein the initial machine learning algorithm is trained using a multi-layer long short-term memory (LSTM) network (Chan, [0110] – “As shown in FIG. 9, neural network 902 is formed by combining LSTM neural network 902a with full-connect neural network 902b. LSTM neural network 902a and full-connect neural network 902b may each have a recurrent neural network (RNN) architecture.” – teaches the initial machine learning algorithm trained using a multi-layer long short-term memory (LSTM) network).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the multi-layer long short-term memory network of Chan to further modify the method of Wang, Kabrams, and Chan in order to train the machine learning algorithm to better detect or forecast a seizure event occurring in measurement data using an LSTM. Doing so would significantly improve detection of ongoing and incoming seizures, as scalp EEG data is highly temporally correlated, so that inclusion of data from recent and distant time points adds useful information to seizure detection (Chan, [0065]).
Regarding claim 7, the combination of Wang, Kabrams, and Chan teaches the method of claim 6.
wherein the multi-layer LSTM network comprises at least three LSTM network layers (Kabrams, [0196] – “The output from the DCNN encoder may then be used as an input layer to an RNN for final detection. In some embodiments, the RNN may include a bidirectional-LSTM followed by two fully connected neural network layers.” – teaches a bidirectional-LSTM layer, and in [0199] – “any other suitable type of recurrent neural network architecture may be used instead of or in addition to an LSTM architecture.” – teaches the multi-layer LSTM network comprising at least three network layers).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the three LSTM layers of Kabrams to the method of Wang, Kabrams, and Chan in order to create a LSTM network with at least three layers. Doing so would enable for better depiction of the transitions of patient brains going from seizure state to non-seizure state, as recurrent neural networks and LSTMs may automatically give output that better reflect this “smoothness” (Kabrams, [0178]).
Regarding claim 8, the combination of Wang, Kabrams, and Chan teaches the method of claim 6.
wherein the multi-layer LSTM network comprises at least one non-trainable layer and at least one trainable layer (Kabrams, [0196] – “In some embodiments, the DCNN encoder may include a 13-layer 2-D convolutional neural network with fractional max-pooling (FMP). After training the DCNN encoder, the weights of this network may be fixed. The output from the DCNN encoder may then be used as an input layer to an RNN for final detection. In some embodiments, the RNN may include a bidirectional-LSTM followed by two fully connected neural network layers.”- teaches an LSTM network comprising at least one non-trainable layer (the DCNN encoder, with frozen weights, whose output is used as the input layer for an RNN that may include a bidirectional-LSTM), and at least one trainable layer (the bidirectional LSTM that inputs are fed into)).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the non-trainable and trainable layers of Kabrams to the method of Wang, Kabrams, and Chan in order to create a LSTM network with non-trainable and trainable layers. Doing so would enable for better depiction of the transitions of patient brains going from seizure state to non-seizure state, as recurrent neural networks and LSTMs may automatically give output that better reflect this “smoothness” (Kabrams, [0178]) and permit for feeding encoder output as input to a LSTM for final stages of detection (Kabrams, [0196]).
Regarding claim 9, the combination of Wang, Kabrams, and Chan teaches the method of claim 8, wherein the at least one non-trainable layer comprises a first layer of the multi-layer LSTM network (Kabrams, [0196] – “In some embodiments, the DCNN encoder may include a 13-layer 2-D convolutional neural network with fractional max-pooling (FMP). After training the DCNN encoder, the weights of this network may be fixed. The output from the DCNN encoder may then be used as an input layer to an RNN for final detection. In some embodiments, the RNN may include a bidirectional-LSTM followed by two fully connected neural network layers.”- teaches an LSTM network comprising at least one non-trainable layer (the DCNN encoder, with frozen weights, whose output is used as the input layer for an RNN that may include a bidirectional-LSTM), wherein the non-trainable layer is a first layer of the multi-layer LSTM network (the non-trainable layer is the input layer)).
Regarding claim 10, the combination of Wang, Kabrams, and Chan teaches the method of claim 8, wherein the at least one non-trainable layer comprises two non-trainable layers and the two non-trainable layers comprise a first layer and second layer of the multi-layer LSTM network (Kabrams, [0196] – “In some embodiments, the DCNN encoder may include a 13-layer 2-D convolutional neural network with fractional max-pooling (FMP). After training the DCNN encoder, the weights of this network may be fixed. The output from the DCNN encoder may then be used as an input layer to an RNN for final detection. In some embodiments, the RNN may include a bidirectional-LSTM followed by two fully connected neural network layers.”- teaches an LSTM network comprising at least one non-trainable layer (the DCNN encoder, with frozen weights, whose output is used as the input layer for an RNN that may include a bidirectional-LSTM) and in [0198] – “It should be appreciated that the described deep learning network is only one example implementation and that other implementations may be employed. For example, in some embodiments, one or more other types of neural network layers may be included in the deep learning network instead of or in addition to one or more of the layers in the described architecture.” – teaches wherein the first and second layers of the multi-layer LSTM network are non-trainable layers (the DCNN encoder with frozen weights used as input layers)).
Regarding claim 13, the combination of Wang, Kabrams, and Chan teaches the method of claim 1, wherein the measurement data comprise at least two of the motion data, the blood volume pulse data, the electrodermal activity data, the temperature data, the time of day, and the heart rate data (Wang, Page 18, Col. 13 Line 63 - Col. 14 Line 11 – “At operation 810, one or more feature vectors is generated including at least one of: an EEG feature vector based on data based on EEG signals such as the aggregate EEG data; an ECG feature vector based on data including ECG signals such as the aggregate ECG data; and a motion sensor feature vector based on data including motion sensor data such as the aggregate motion sensor data. In an implementation, the generation of the one or more feature vectors can be performed: on a local device, such as the wearable apparatus 200 or the computing device 300; or in an offline learning network. The one or more feature vectors can include attributes and corresponding values. For example, an ECG feature vector can include heart rate values and other features of ECG signals. Further, adverse physiological event features such as epilepsy related features (e.g., high frequency power) can be extracted from the EEG signals.” – teaches recording measurement data based on EEG signals, wherein the data comprises at least two of motion data (motion sensor data), heart rate data (heart rate values), and in Page 14, Col. 5 Lines 38-47 - “The sensors 206 can also comprise one or more bioimpedance sensors, microphones, temperature sensors, touch screens, finger readers, iris scanners, a combination of the above, or the like. Implementations of the sensors 206 can include a single sensor, one of each of the foregoing sensors, or any combination of the foregoing sensors. In an implementation, the signal data can be identified, detected, determined, or otherwise generated based on any single sensor or combination of sensors included in the wearable apparatus 200.” – teaches recording temperature data).
Regarding claim 15, the combination of Wang, Kabrams, and Chan teaches the method of claim 1, wherein the computer system is contained within the wearable device (Wang, Page 14 Col. 6 Lines 24-26 – “In some implementations, the computing device 300 and the wearable apparatus 200 (or any device having measurement capabilities) can be the same device.” – teaches wherein the computer system is contained within the wearable device).
Regarding claim 16, the combination of Wang, Kabrams, and Chan teaches the method of claim 1, wherein the computer system is physically separate from the wearable device (Wang, Page 15 Col. 7 Lines 43-50 – “In an implementation, the wearable event detection apparatus 410 is coupled, via the network 420, to: computing device 430, which in an implementation can include the features of the computing device 300 illustrated in FIG. 3; and server device 440 which in an implementation can include the features of a data server computing device configured to exchange (send and receive) data with requesting client devices such as the computing device 430.” – teaches wherein the computer system is physically separate from the wearable device).
Regarding claim 17, the combination of Wang, Kabrams, and Chan teaches the method of claim 1, further comprising generating an alarm to a user using the wearable device when a seizure event is at least one of detected or predicted in the measurement data (Wang, Page 18 Col. 13 Lines 41-48 – “At operation 720, an alert indication including the identity of the imminent adverse physiological event is generated. The alert indication can include at least one of a haptic alert, an audible alert, a visual alert, and an electronic message. For example, when an adverse physiological event is determined to be imminent an audible indication such as a chime can alert the wearer of a device such as the wearable apparatus 200 or the computing device 300.” – teaches generating an alarm to a user wearing the wearable device (wearable apparatus 200) when a seizure event is at least one of detected or predicted in the measurement data (adverse physiological event determined to be imminent)).
Regarding claim 18, the combination of Wang, Kabrams, and Chan teaches the method of claim 18, wherein the alarm comprises an auditory alarm (Wang, Page 18 Col. 13 Lines 41-48 – “At operation 720, an alert indication including the identity of the imminent adverse physiological event is generated. The alert indication can include at least one of a haptic alert, an audible alert, a visual alert, and an electronic message.” – teaches the alarm comprising an auditory alarm).
Regarding claim 19, the combination of Wang, Kabrams, and Chan teaches the method of claim 18, wherein the alarm comprises a visual alarm (Wang, Page 18 Col. 13 Lines 41-48 – “At operation 720, an alert indication including the identity of the imminent adverse physiological event is generated. The alert indication can include at least one of a haptic alert, an audible alert, a visual alert, and an electronic message.” – teaches the alarm comprising a visual alarm).
Regarding claim 20, the combination of Wang, Kabrams, and Chan teaches the method of claim 1, wherein the output indicates that the seizure event is presently occurring within the measurement data (Kabrams, [0147] – “the mobile device 250 or 252 may display real-time seizure risk for the person suffering from the neurological disorder. In the event of a seizure, the mobile device 250 or 252 may alert the person, a caregiver, or another suitable entity. For example, the mobile device 250 or 252 may inform a caretaker that a seizure is predicted in the next 30 minutes, next hour, or another suitable time period.” – teaches wherein the output indicates that the seizure event is presently occurring within the measurement data (real-time)).
Regarding claim 21, the combination of Wang, Kabrams, and Chan teaches the method of claim 1, wherein the output indicates that the seizure event is likely to occur within a duration of time (Kabrams, [0147] – “the mobile device 250 or 252 may inform a caretaker that a seizure is predicted in the next 30 minutes, next hour, or another suitable time period.” – teaches wherein the output indicates the seizure event is likely to occur within a duration of time).
Regarding claim 22, the combination of Wang, Kabrams, and Chan teaches the method of claim 21, wherein the duration of time is within 90 minutes (Kabrams, [0147] – “the mobile device 250 or 252 may inform a caretaker that a seizure is predicted in the next 30 minutes, next hour, or another suitable time period.” – teaches wherein the output indicates the seizure event is likely to occur within a duration of 90 minutes).
Regarding claim 23, the combination of Wang, Kabrams, and Chan teaches the method of claim 22, wherein the duration of time is within 60 to 90 minutes (Kabrams, [0147] – “the mobile device 250 or 252 may inform a caretaker that a seizure is predicted in the next 30 minutes, next hour, or another suitable time period.” – teaches wherein the output indicates the seizure event is likely to occur within a duration 60 minutes to 90 minutes).
Claim(s) 11-12, and 14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Wang, Kabrams, and Chan as applied to claim 1 above, and further in view of Karoly et al. (US Pub. No. 2022/0095993, effective filing date of Sept. 2020, hereinafter “Karoly”).
Regarding claim 11, the combination of Wang, Kabrams, and Chan teaches the method of claim [[4]]1,
before being retrained on the second training data (Chan, [0039] – “At this point, as the data that is used to create model set 226 is primarily collected under special environments (e.g., hospital and medicated conditions), the data inherently includes biases, as already explained in with respect to the prior art data discussed above. To eliminate the biases, the seizure early detection system of the present invention continuously fine-tunes model set 226 based on patient 100's own day-to-day EEG data.” – teaches retraining using the second training data (fine-tunes model based on patient 100’s own day-to-day EEG data)).
The combination of Wang, Kabrams, and Chan fails to explicitly teach wherein the initial machine learning algorithm is first retrained on third training data comprising ambulatory EEG data acquired from ambulatory subjects.
However, analogous to the field of seizure prediction, Karoly teaches:
wherein the initial machine learning algorithm is first retrained on third training data comprising ambulatory EEG data acquired from ambulatory subjects (Karoly, [0102] – “Retrospective analysis of EEG/ECG used a database of people undergoing at-home, ambulatory video-EEG/ECG diagnostic testing for epilepsy." – teaches training data comprising ambulatory EEG data acquired from ambulatory subjects, and in [0103] – “This study used 30 records of at least 8 days duration (range 8-14 days). Only continuous ECG data were used, combined with event labels derived from video-EEG. Events were labelled using computer-assisted review, whereby event detection was first performed by a machine learning algorithm.” – teaches training the machine learning algorithm on third training data comprising ambulatory EEG training data acquired from ambulatory subjects (analysis of EEG of people undergoing at-home, ambulatory video-EEG testing, thus teaching training the machine learning algorithm on ambulatory EEG data acquired from ambulatory subjects))
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the training on ambulatory data of Karoly before retraining the machine learning algorithm on the second training data of Wang, Kabrams, and Chan. Doing so would allow for balanced training data between low resolution wearable device data over long periods of time and for short-term, extremely accurate ambulatory EEG recordings (Karoly, [0086]).
Regarding claim 12, the combination of Wang, Kabrams, and Chan teaches the method of claim [[2]]1,
wherein the training data comprise wearable device data acquired from limbs of subjects (Wang, Page 13, Col. 3, Lines 54-61 – “FIG. 1 is a diagram of an implementation of a wearable system 100 which can include a wearable apparatus 110 worn by an individual and a device core 120. The wearable apparatus 110 can include a housing that is in the form of a ring, bracelet, wristband, pendant, armband, anklet, headband, belt, necklace, glove, a chest patch, or other mechanism for securing or attaching the wearable apparatus 110 to a human body.”, Pg. 13, Col. 3, Lines 37-47 – “The disclosed technology can leverage big data and deep learning networks to analyze the relationship between abnormal electrical activity in the brain (such as can be detected by an EEG) and other physical indicators such as electrical cardiac activity that can be monitored by an ECG. Further, the disclosed technology can monitor ECG signals, motion sensor signals, and other signals on a continuous basis. In this way a comprehensive snapshot of a user's physiological state can be generated and an imminent adverse physiological event, such as an epileptic seizure, can be more effectively detected.” – teaches wherein the training data comprise wearable device data acquired from limbs of subjects (wearable apparatus can include housing in form of ring, anklet, bracelet, etc. to be worn on limb of subject, and provides data of the patient for training data))
Wang fails to explicitly teach wherein the training data comprise non-ambulatory electroencephalography (EEG) data acquired from non-ambulatory subjects, [[and]] ambulatory EEG data acquired from ambulatory subjects.
However, analogous to the field of the claimed invention, Chan teaches:
wherein the training data comprise non-ambulatory electroencephalography (EEG) data acquired from non-ambulatory subjects, and wearable device data acquired from subjects (Chan, [0039] – “At this point, as the data that is used to create model set 226 is primarily collected under special environments (e.g., hospital and medicated conditions), the data inherently includes biases, as already explained in with respect to the prior art data discussed above. To eliminate the biases, the seizure early detection system of the present invention continuously fine-tunes model set 226 based on patient 100's own day-to-day EEG data.” – teaches the first training data comprising non-ambulatory EEG data (collected under special environments, e.g., hospital and medicated conditions) and the second training data comprising wearable device data acquired from subjects (own day-to-day data)).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the non-ambulatory EEG data of Chan to the method of Wang and Kabrams in order to train the machine learning algorithm to better detect or forecast a seizure event occurring in measurement data. Doing so would provide benefit from the device being portable and able to measure a variety of physiological signals while also leveraging the advantages of previously generated data (Wang, Pg. 13, Col. 3), prevent algorithms from developing biases on one modality of data (Chan, [0006]), and improve model performance (Chan, [0033]).
The combination of Wang, Kabrams, and Chan fails to explicitly teach ambulatory EEG data acquired from ambulatory subjects.
However, analogous to the field of seizure prediction, Karoly teaches:
ambulatory EEG data acquired from ambulatory subjects (Karoly, [0102] – “Retrospective analysis of EEG/ECG used a database of people undergoing at-home, ambulatory video-EEG/ECG diagnostic testing for epilepsy." – teaches training data comprising ambulatory EEG data acquired from ambulatory subjects),
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the ambulatory training data to the non-ambulatory and wearable device data of Wang, Kabrams, and Chan. Doing so would allow for balanced training data between low resolution wearable device data over long periods of time and for short-term, extremely accurate ambulatory EEG recordings (Karoly, [0086]) and in-hospital data recorded from patients that are medicated or non-ambulatory (Chan, [0039]).
Regarding claim 14, the combination of Wang, Kabrams, and Chan teaches the method of claim 1,
wherein the measurement data comprise the motion data, the temperature data, and the heart rate data (Wang, Page 18, Col. 13 Line 63 - Col. 14 Line 2 – “At operation 810, one or more feature vectors is generated including at least one of: an EEG feature vector based on data based on EEG signals such as the aggregate EEG data; an ECG feature vector based on data including ECG signals such as the aggregate ECG data; and a motion sensor feature vector based on data including motion sensor data such as the aggregate motion sensor data.” – teaches recording measurement data based on EEG signals, wherein the data comprises motion data (motion sensor data), heart rate data (heart rate values), and in Page 14, Col. 5 Lines 38-47 - “The sensors 206 can also comprise one or more bioimpedance sensors, microphones, temperature sensors, touch screens, finger readers, iris scanners, a combination of the above, or the like. Implementations of the sensors 206 can include a single sensor, one of each of the foregoing sensors, or any combination of the foregoing sensors. In an implementation, the signal data can be identified, detected, determined, or otherwise generated based on any single sensor or combination of sensors included in the wearable apparatus 200.” – teaches recording temperature data).
The combination of Wang, Kabrams, and Chan fails to explicitly teach wherein the measurement data comprises blood volume pulse data, electrodermal activity data, and the time of day.
However, analogous to the field of seizure prediction, Karoly teaches:
wherein the measurement data comprises blood volume pulse data, electrodermal activity data, and the time of day (Karoly, [0044] – “Measurement devices which may be coupled to the measurement unit 204 may comprise (but are not limited to) an EEG monitoring device 216, a heart monitor (photo-plethysmograph or ECG) 218, a sweat or electro dermal sensor 220 an accelerometer 222 (or similar motion detector), a temperature sensor 223, and oxygen saturation measurement device 224, such as a pulse oximeter, and a respiratory monitor 225. Other examples of measurement devices which may be coupled to the measurement unit 204 include blood pressure monitors, glucose monitors, cortisol sensor, and gyroscopes etc.” – teaches measuring blood volume pulse data (blood pressure monitors, photo-plethysmograph), electrodermal data (swear or electro dermal sensor), and in [0037] – “The seizure activity data 106 comprises a time (that is a time of day and date) at which each seizure or seizures occurred during the time period.” – teaches recording time of day data).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the measurement data of Karoly to the measurement data of Wang and Kabrams. Doing so would permit training machine learning algorithms with more non-EEG physiological data that is correlated with seizure occurrence (Karoly, [0034]).
Claim(s) 24-26, and 30-31 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chan in view of Karoly.
Regarding claim 24, Chan teaches a method for training a machine learning classifier algorithm for detecting or forecasting seizure events in measurement data collected with a wearable device being worn [[by]] on a limb of a subject, the method comprising:
(a) accessing training data with a computer system having a processor and a memory, the training data comprising: non-ambulatory electroencephalography (EEG) data acquired from non-ambulatory subjects, and wearable device data acquired from subjects wearing a wearable device (Chan, [0039] – “At this point, as the data that is used to create model set 226 is primarily collected under special environments (e.g., hospital and medicated conditions), the data inherently includes biases, as already explained in with respect to the prior art data discussed above. To eliminate the biases, the seizure early detection system of the present invention continuously fine-tunes model set 226 based on patient 100's own day-to-day EEG data.” – teaches the first training data comprising non-ambulatory EEG data (collected under special environments, e.g., hospital and medicated conditions) and the second training data comprising wearable device data acquired from subjects (own day-to-day data));
(b) training an initial classifier on the non-ambulatory EEG data using the computer system, generating output as a trained initial classifier (Chan, [0039] – “At this point, as the data that is used to create model set 226 is primarily collected under special environments (e.g., hospital and medicated conditions)” – teaches training an initial classifier on the non-ambulatory EEG data using the computer system, generating output as a trained classifier (model set 226));
(d) retraining the retrained classifier on the wearable device data with transfer learning using the computer system, generating output as a trained classifier (Chan, [0039] – “To eliminate the biases, the seizure early detection system of the present invention continuously fine-tunes model set 226 based on patient 100's own day-to-day EEG data.” – teaches retraining the classifier (model set 226) on the wearable device data (day-to-day EEG data) with transfer learning, as in [0120] – “Transfer learning may be used during training of the neural network model to compensate for the difficulty in model convergence due to the reduced set of channels”); and,
(e) storing the trained classifier in the memory of the computer system for later use (Chan, [0037] – “The collection of selected trained models forms patient 100's “Basic Early Detection Model” or model set 226,” – teaches storing the trained classifier in the memory of the computer system for later use).
Chan fails to explicitly teach the training data comprising: ambulatory EEG data acquired from ambulatory subjects, wearable device data comprising non-EEG physiological signals acquired from subjects wearing a wearable device on a limb, and (c) retraining the trained initial classifier on the ambulatory EEG data using the computer system, generating output as a retrained classifier;
However, analogous to the field of seizure prediction, Karoly teaches:
the training data comprising: ambulatory EEG data acquired from ambulatory subjects (Karoly, [0102] – “Retrospective analysis of EEG/ECG used a database of people undergoing at-home, ambulatory video-EEG/ECG diagnostic testing for epilepsy." – teaches training data comprising ambulatory EEG data acquired from ambulatory subjects), and wearable device data comprising non-EEG physiological signals acquired from subjects wearing a wearable device on a limb (Karoly, [0044] – “Measurement devices which may be coupled to the measurement unit 204 may comprise (but are not limited to) an EEG monitoring device 216, a heart monitor (photo-plethysmograph or ECG) 218, a sweat or electro dermal sensor 220 an accelerometer 222 (or similar motion detector), a temperature sensor 223, and oxygen saturation measurement device 224, such as a pulse oximeter, and a respiratory monitor 225. Other examples of measurement devices which may be coupled to the measurement unit 204 include blood pressure monitors, glucose monitors, cortisol sensor, and gyroscopes etc.” and in [0088] – “Continuous data were collected via mobile and wearable devices for at least 2 months and up to 2 years. Participants wore a smartwatch and manually reported seizure times in a freely available mobile diary app. The smartwatch continuously measured participants' heart rates (via photoplethysmograph) at 5 s resolution. The smartwatch also monitored estimated sleep stage (wake, REM, light and deep sleep), and step count per hour.” – teaches training data comprising wearable device data comprising non-EEG physiological signals acquired from subjects wearing a wearable device on a limb (participants wore smart watches, thus a wearable device on a limb of a patient, that acquired non-EEG physiological data (device acquired heart rate via photoplethysmograph, which is defined as non-EEG physiological data in Karoly at [0012])))
(c) retraining the trained initial classifier on the ambulatory EEG data using the computer system, generating output as a retrained classifier (Karoly, [0103] – “This study used 30 records of at least 8 days duration (range 8-14 days). Only continuous ECG data were used, combined with event labels derived from video-EEG. Events were labelled using computer-assisted review, whereby event detection was first performed by a machine learning algorithm.” – teaches training the machine learning algorithm to detect events using the ambulatory training data);
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the training on ambulatory EEG data and non-EEG physiological data acquired by a wearable device worn on a limb of a patient of Karoly to the non-ambulatory and wearable device data, multi-stage training, and storage of a classifier of Chan. Doing so would so would permit training machine learning algorithms with more non-EEG physiological data that is correlated with seizure occurrence (Karoly, [0034]), allow for balanced training data between low resolution wearable device data over long periods of time and for short-term, extremely accurate ambulatory EEG recordings (Karoly, [0086]) and retrain a model to more accurately detect the arrivals of seizure in a patient’s day to day life (Chan, [0006])
Regarding claim 25, the combination of Chan and Karoly teaches the method of claim 24, wherein the subjects wearing the wearable device comprise at least one of the non-ambulatory subjects or the ambulatory subjects (Chan, [0104] – “After acquisition device 110 is determined to be operating satisfactorily at decision step 850, at steps 860 and 861, patient 100 performs under instruction a set of typical body movements to uncover muscle artifacts in scalp EEG recording. These movements include blinking of the eyes, shaking and turning the head left and right, nodding and moving the head up and down, opening and closing the mouth, smiling and other every day human motions. Concurrently, at step 880, scalp EEG data is collected and labeled during these movements is recorded and labeled.” – teaches wherein the subjects wearing the wearable device (acquisition device 110) comprise at least one of the non-ambulatory subjects or ambulatory subjects (patient 100 performing body movements such as blinking, shaking and turning head, nodding, opening and closing mouth, smiling, etc.)).
Regarding claim 26, the combination of Chan and Karoly teaches the method of claim 24, wherein the initial classifier is trained using a multi-layer long short-term memory (LSTM) network (Chan, [0032] – “The models in model set 161 may be trained or retrained in the same set up on GPU farm 163, using training and testing data from storage unit 170. When retraining happens as designed or scheduled (as discussed in further detail below), or when desired, GPU farm 163 retrains the early detection models based on the collected scalp EEG data, or other suitable data, to update model set 161.” – teaches wherein the trained machine learning algorithm is trained on the training data using a multi-stage training process. In addition to the previously cited passages, Chan further teaches in [0110] – “As shown in FIG. 9, neural network 902 is formed by combining LSTM neural network 902a with full-connect neural network 902b. LSTM neural network 902a and full-connect neural network 902b may each have a recurrent neural network (RNN) architecture.” – teaches the initial machine learning algorithm trained using a multi-layer long short-term memory network).
Regarding claim 30, the combination of Chan and Karoly teaches the method of claim 24, wherein the initial classifier is first retrained on third training data comprising ambulatory EEG data acquired from ambulatory subjects (Karoly, [0102] – “Retrospective analysis of EEG/ECG used a database of people undergoing at-home, ambulatory video-EEG/ECG diagnostic testing for epilepsy." – teaches training data comprising ambulatory EEG data acquired from ambulatory subjects, and in [0103] – “This study used 30 records of at least 8 days duration (range 8-14 days). Only continuous ECG data were used, combined with event labels derived from video-EEG. Events were labelled using computer-assisted review, whereby event detection was first performed by a machine learning algorithm.” – teaches training the machine learning algorithm to detect events using the ambulatory training data) before being retrained on the wearable device data (Chan, [0039] – “At this point, as the data that is used to create model set 226 is primarily collected under special environments (e.g., hospital and medicated conditions), the data inherently includes biases, as already explained in with respect to the prior art data discussed above. To eliminate the biases, the seizure early detection system of the present invention continuously fine-tunes model set 226 based on patient 100's own day-to-day EEG data.” – teaches retraining using the wearable device data (fine-tunes model based on patient 100’s own day-to-day EEG data)).
Regarding claim 31, the combination of Chan and Karoly teach the method of claim 24, wherein the wearable device data comprise at least two of subject motion data, subject blood volume pulse data, subject electrodermal activity data, subject temperature data, time of day, and subject heart rate data (Karoly, [0044] – “Measurement devices which may be coupled to the measurement unit 204 may comprise (but are not limited to) an EEG monitoring device 216, a heart monitor (photo-plethysmograph or ECG) 218, a sweat or electro dermal sensor 220 an accelerometer 222 (or similar motion detector), a temperature sensor 223, and oxygen saturation measurement device 224, such as a pulse oximeter, and a respiratory monitor 225. Other examples of measurement devices which may be coupled to the measurement unit 204 include blood pressure monitors, glucose monitors, cortisol sensor, and gyroscopes etc.” – teaches measuring at least two of blood volume pulse data (blood pressure monitors, photo-plethysmograph), electrodermal data (swear or electro dermal sensor), or in [0037] – “The seizure activity data 106 comprises a time (that is a time of day and date) at which each seizure or seizures occurred during the time period.” – teaches recording time of day data).
Claim(s) 27-29 is/are rejected under 35 U.S.C. 103 as being unpatentable over Chan and Karoly as applied to claim 24 above, and further in view of Kabrams.
Regarding claim 27 the combination of Chan and Karoly teaches the method of claim 26.
The combination of Chan and Karoly fails to explicitly teach wherein the multi-layer LSTM network comprises at least one non-trainable layer and at least one trainable layer.
However, analogous to the field of seizure prediction, Kabrams teaches:
wherein the multi-layer LSTM network comprises at least one non-trainable layer and at least one trainable layer (Kabrams, [0196] – “In some embodiments, the DCNN encoder may include a 13-layer 2-D convolutional neural network with fractional max-pooling (FMP). After training the DCNN encoder, the weights of this network may be fixed. The output from the DCNN encoder may then be used as an input layer to an RNN for final detection. In some embodiments, the RNN may include a bidirectional-LSTM followed by two fully connected neural network layers.”- teaches an LSTM network comprising at least one non-trainable layer (the DCNN encoder, with frozen weights, whose output is used as the input layer for an RNN that may include a bidirectional-LSTM), and at least one trainable layer (the bidirectional LSTM that inputs are fed into)).
Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the multi-layer LSTM network to the LSTM network of Chan and Karoly. Doing so would enable pre-training other machine learning algorithms for use as input layers into an LSTM network (Kabrams, [0196]).
Regarding claim 28, the combination of Chan, Karoly, and Kabrams teaches the method of claim 27, wherein the at least one non-trainable layer comprises a first layer of the multi-layer LSTM network (Kabrams, [0196] – “In some embodiments, the DCNN encoder may include a 13-layer 2-D convolutional neural network with fractional max-pooling (FMP). After training the DCNN encoder, the weights of this network may be fixed. The output from the DCNN encoder may then be used as an input layer to an RNN for final detection. In some embodiments, the RNN may include a bidirectional-LSTM followed by two fully connected neural network layers.”- teaches an LSTM network comprising at least one non-trainable layer (the DCNN encoder, with frozen weights, whose output is used as the input layer for an RNN that may include a bidirectional-LSTM), wherein the non-trainable layer is a first layer of the multi-layer LSTM network (the non-trainable layer is the input layer)).
Regarding claim 29, the combination of Chan, Karoly, and Kabrams teaches the method of claim 27, wherein the at least one non-trainable layer comprises two non-trainable layers and the two non-trainable layers comprise a first layer and second layer of the multi-layer LSTM network (Kabrams, [0196] – “In some embodiments, the DCNN encoder may include a 13-layer 2-D convolutional neural network with fractional max-pooling (FMP). After training the DCNN encoder, the weights of this network may be fixed. The output from the DCNN encoder may then be used as an input layer to an RNN for final detection. In some embodiments, the RNN may include a bidirectional-LSTM followed by two fully connected neural network layers.”- teaches an LSTM network comprising at least one non-trainable layer (the DCNN encoder, with frozen weights, whose output is used as the input layer for an RNN that may include a bidirectional-LSTM) and in [0198] – “It should be appreciated that the described deep learning network is only one example implementation and that other implementations may be employed. For example, in some embodiments, one or more other types of neural network layers may be included in the deep learning network instead of or in addition to one or more of the layers in the described architecture.” – teaches wherein the first and second layers of the multi-layer LSTM network are non-trainable layers (the DCNN encoder with frozen weights used as input layers)).
Response to Arguments
Applicant's arguments filed 10 March 2026 have been fully considered but they are not persuasive. Wang teaches the amended limitation of claim 1 regarding “the second training data comprise wearable device data comprising non-EEG physiological signals acquired from limbs of subjects.” Karoly teaches the amended limitation of claim 24 regarding the training data comprising: wearable device data comprising non-EEG physiological signals acquired from subjects wearing a wearable device on a limb.”
In response to applicant’s argument that there is no teaching, suggestion, or motivation to combine the references, the examiner recognizes that obviousness may be established by combining or modifying the teachings of the prior art to produce the claimed invention where there is some teaching, suggestion, or motivation to do so found either in the references themselves or in the knowledge generally available to one of ordinary skill in the art. See In re Fine, 837 F.2d 1071, 5 USPQ2d 1596 (Fed. Cir. 1988), In re Jones, 958 F.2d 347, 21 USPQ2d 1941 (Fed. Cir. 1992), and KSR International Co. v. Teleflex, Inc., 550 U.S. 398, 82 USPQ2d 1385 (2007). In this case:
On pp. 3 of Remarks, Applicant argues that the combination lacks proper motivation and states that “the purpose of claim 1 as amended, which involves adapting an algorithm initially trained on abundant EEG data to work with limited wearable device biosignal data (i.e., non-EEG physiological signals) through transfer learning.” Examiner respectfully disagrees and points to Wang at Pg. 13, Col. 3, Lines 30-36 – “Moreover, the user of an adverse physiological event apparatus could benefit from the device being portable and able to measure a variety of physiological signals such as electrocardiogram (ECG) signals and motion sensor data, while also leveraging the advantages of previously generated data that could be used to reconstruct an EEG signal based on the ECG signals.” – which explicitly states the benefit of a portable device able to measure a variety of non-EEG physiological signals while also leveraging the advantages of previously generated data. Chan teaches at [0006] – “Furthermore, for medical diagnosis and evaluation purposes, the available scalp EEG data of epilepsy patients kept in the medical records most likely are biased, as the scalp EEG data are often taken under medication and medication adjustments aimed at, for example, initiating a seizure for a pre-surgical evaluation. Thus, such EEG data are not representative of the user's day-to-day life. In the previous work, any seizure prediction or early detection model trained using such EEG data would also be biased and could not predict the arrivals of seizure accurately in a patient's day-to-day life.” – which explicitly states that “early detection models” are initially trained using EEG data taken from in-hospital, medicated, and/or pre-surgical settings, and such EEG data was biased and could not predict arrivals of seizure accurately in a patient’s day to day life. Chan further teaches at [0120] that “Transfer learning may be used during training of the neural network model to compensate for the difficulty in model convergence due to the reduced set of channels.” – which states that transfer learning may be used during training to assist with model convergence. Thus, the combination of Wang, Kabrams, and Chan teaches adapting an algorithm initially trained on abundant EEG data to work with limited wearable device biosignal data (i.e., non-EEG physiological signals) through transfer learning. Doing so would provide a portable device that measures non-EEG physiological data, while leveraging the advantages of previously generated data (Wang, Pg. 13, Col. 3), and retrain a model to more accurately detect the arrivals of seizure in a patient’s day to day life (Chan, [0006]).
Regarding claim 24, Chan teaches at [0006] – “Furthermore, for medical diagnosis and evaluation purposes, the available scalp EEG data of epilepsy patients kept in the medical records most likely are biased, as the scalp EEG data are often taken under medication and medication adjustments aimed at, for example, initiating a seizure for a pre-surgical evaluation. Thus, such EEG data are not representative of the user's day-to-day life. In the previous work, any seizure prediction or early detection model trained using such EEG data would also be biased and could not predict the arrivals of seizure accurately in a patient's day-to-day life.” – which explicitly states that “early detection models” are initially trained using EEG data taken from in-hospital, medicated, and/or pre-surgical settings, and such EEG data was biased and could not predict arrivals of seizure accurately in a patient’s day to day life. Karoly teaches at [0088] – “Continuous data were collected via mobile and wearable devices for at least 2 months and up to 2 years. Participants wore a smartwatch and manually reported seizure times in a freely available mobile diary app. The smartwatch continuously measured participants' heart rates (via photoplethysmograph) at 5 s resolution. The smartwatch also monitored estimated sleep stage (wake, REM, light and deep sleep), and step count per hour.” – which states collecting training data acquired from a wearable device worn on a limb of a patient (smartwatch), the training data comprising non-EEG physiological data (smartwatch continuously measured participants' heart rates (via photoplethysmograph), which Karoly states is non-EEG physiological data at [0012]). Therefore, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to incorporate the training on ambulatory EEG data and non-EEG physiological data acquired by a wearable device worn on a limb of a patient of Karoly to the non-ambulatory and wearable device data, multi-stage training, and storage of a classifier of Chan. Doing so would so would permit training machine learning algorithms with more non-EEG physiological data that is correlated with seizure occurrence (Karoly, [0034]), allow for balanced training data between low resolution wearable device data over long periods of time and for short-term, extremely accurate ambulatory EEG recordings (Karoly, [0086]) and retrain a model to more accurately detect the arrivals of seizure in a patient’s day to day life (Chan, [0006]).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Beniczky et al. (NPL: Machine learning and wearable devices of the future, published March 2020) teaches seizure detection and prediction using wearable devices. Teaches recording EEG data of patients in a non-ambulatory setting, and that application for monitoring in the ambulatory settings are desired. Teaches a wearable device that collections non-EEG physiological data. Teaches training using both EEG data and wearable device data.
Nasseri et al. (NPL: Non-invasive wearable seizure detection using long-short-term memory networks with transfer learning, published March 2021) teaches an adaptively trained LSTM deep neural network using seizure datasets from wrist-worn devices. Transfer learning was used to adapt a classifier that was initially trained on intracranial EEG signals to facilitate classification of non-EEG physiological datasets comprising data such as accelerometry, blood volume pulse, skin electrodermal activity, heart rate, etc. Thus teaching adaptive training a classifier that was initially trained on iEEG and retrains, through transfer learning, using non-EEG physiological data. The classifier is evaluated using motor and non-motor seizures during in-hospital and ambulatory use.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LOUIS C NYE whose telephone number is 571-272-0636. The examiner can normally be reached Monday - Friday 9:00AM - 5:00PM.
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, MATT ELL can be reached at 571-270-3264. 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.
/LOUIS CHRISTOPHER NYE/Examiner, Art Unit 2141
/MATTHEW ELL/Supervisory Patent Examiner, Art Unit 2141
Prosecution Insights