DETAILED ACTION
This action is in response to the initial filing filed on May 31, 2023 Claims 1-25 have been examined in this application.
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 .
Information Disclosure Statement
The Information Disclosure Statement (IDS) filed on 5/31/2023, 9/14/2023, and 9/27/2023 have been acknowledged.
Priority
Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55.
Minor Informalities
Claim 22 recites “The method according to claim 16, c further comprising,” containing an extraneous character (“c” ) of indeterminate meaning. Examiner will interpret the claim language to read “The method according to claim 16, further comprising,” for the sake of compact prosecution. Appropriate correction is required.
Claim Rejections - 35 USC § 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 1-4, 6-17, and 19-25 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e. an abstract idea) without significantly more.
Step 1: Claims 1-15 and 25 are drawn to a device (i.e., a manufacture) and claims 16-24 are drawn to a method. As such, claims 1-25 are drawn to one of the statutory categories of invention (Step 1: YES).
Under Step 2A Prong 1, the claims are analyzed to determine whether the claims recite any judicial exceptions including certain groupings of abstract ideas (i.e., mathematical concepts, certain methods of organizing human activity such as a fundamental economic practice, or mental processes).
Claims 1, and 16 recite a system of detecting food intake events from wearable devices, the system comprising: a signal bank to store physiological digital data signals collected via a wearable device including at least one sensor to sense at least one physiological parameter of a user; a preprocessor module to process the physiological digital data signals stored on the signal bank and to create a descriptive representation of the sensed at least one physiological parameter; a feature extractor module including an automatically learned feature extractor and an analytical feature extractor to obtain features from the descriptive representation, the feature extractor module being configured to automatically select at least one of the automatically learned feature extractor or the analytical feature extractor; and a probability estimator module to determine whether a food intake event of the user occurs based on the obtain features. If claim limitations, under their broadest reasonable interpretation, include a mental process and/or certain methods of organizing human activity, the limitations fall under the abstract ideas judicial exception and therefore recite ineligible subject matter. Accordingly, claims 1, and 16 recite abstract ideas.
Representative Claim 1: A system of detecting food intake events from wearable devices, the system comprising: a signal bank to store physiological digital data signals collected via a wearable device including at least one sensor to sense at least one physiological parameter of a user; a preprocessor module to process the physiological digital data signals stored on the signal bank and to create a descriptive representation of the sensed at least one physiological parameter; a feature extractor module including an automatically learned feature extractor and an analytical feature extractor to obtain features from the descriptive representation, the feature extractor module being configured to automatically select at least one of the automatically learned feature extractor or the analytical feature extractor; and a probability estimator module to determine whether a food intake event of the user occurs based on the obtain features.
Representative Claim 11: A method of detecting food intake events from wearable devices, the method comprising: storing physiological digital data signals in a signal bank, the physiological digital data signals being collected via a wearable device including at least one sensor to sense at least one physiological parameter of a user; processing, with a preprocessor module, the physiological digital data signals stored on the signal bank, and creating a descriptive representation of the sensed at least one physiological parameter; obtaining, with a feature extractor module, features from the descriptive representation, the feature extractor module including an automatically learned feature extractor and an analytical feature extractor and being configured to automatically select at least one of the automatically learned feature extractor or the analytical feature extractor; and generating, with a probability estimator module, an estimated probability to determine whether a food intake event of the user occurs based on the obtained features.
(Examiner notes: The underlined claim terms above are interpreted as additional elements beyond the abstract idea and are further analyzed under Step 2A - Prong Two)
The additional elements are instructions for applying the judicial exceptions with a generic computing device as, under their broadest reasonable interpretation, the additional elements of a wearable device and a sensor are generic computer components for performing the above method, per MPEP 2106.05(f). Under their broadest reasonable interpretation, the additional elements are generic components of a computing device used to apply the abstract idea.
Under their broadest reasonable interpretation, the recited steps of a system of detecting food intake events from wearable devices, the system comprising: storing data signals from a sensor, processing the data signals and creating a descriptive representation of the sensed data, obtaining features from the descriptive representation of the data and identifying data patterns and a characteristic or variable, and estimating the probability of food intake based on identified features (i.e., one or more concepts performed in the human mind, such as one or more observations, evaluations, judgments, opinions), then it also falls within the “Mental Processes” subject matter grouping of abstract ideas. The recited steps are a simulation that applies an abstract idea, specifically mental processes (observation (storing data signals from a sensor, obtaining features from the descriptive representation of the data), and/or evaluation (processing the data signals and creating a descriptive representation of the sensed data, identifying data patterns and a characteristic or variable, estimating the probability of food intake based on identified features)). If claim limitations, under their broadest reasonable interpretation, include a mental process and/or certain methods of organizing human activity (CMOHA), the limitations fall under the abstract ideas judicial exception and therefore recite ineligible subject matter. Accordingly, claims 1 and 16 recite abstract ideas.
Dependent Claims 2-4, 6-15, 17, and 19-25 further narrow the abstract ideas of storing data signals from a sensor, processing the data signals and creating a descriptive representation of the sensed data, obtaining features from the descriptive representation of the data and identifying data patterns and a characteristic or variable, and estimating the probability of food intake based on identified features (i.e., one or more concepts performed in the human mind, such as one or more observations, evaluations, judgments, opinions), then it also falls within the “Mental Processes” and is an abstract idea and then it also falls within the “Organizing Human Processes” subject matter grouping of abstract ideas and then also falls within the “Organizing Human Processes” subject matter grouping of abstract ideas.
Independent claim(s) 1 and 16 recite/describe nearly identical steps (and therefore also recite limitations that fall within this subject matter grouping of abstract ideas), and this/these claim(s) is/are therefore determined to recite an abstract idea under the same analysis.
As such, the Examiner concludes that claims 1 and 16 recite an abstract idea (Step 2A – Prong One: YES).
Under Step 2A Prong 2 the claims are analyzed to determine whether the claims recite additional elements that integrate the judicial exception into a practical application.
Step 2A - Prong Two: In prong two of step 2A, an evaluation is made whether a claim recites any additional element, or combination of additional elements, that integrate the exception into a practical application of that exception. An “addition element” is an element that is recited in the claim in addition to (beyond) the judicial exception (i.e., an element/limitation that sets forth an abstract idea is not an additional element). The phrase “integration into a practical application” is defined as requiring 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 it is more than a drafting effort designed to monopolize the exception.
The requirement to execute the claimed steps/functions using “storing data signals from a sensor”, “processing the data signals and creating a descriptive representation of the sensed data”, “obtaining features from the descriptive representation of the data and identifying data patterns and a characteristic or variable”, and “estimating the probability of food intake based on identified features” etc. (Claims 1 and 16) are equivalent to adding the words “apply it” on a generic computer and/or mere instructions to implement the abstract idea on a generic computer.
Similarly, the limitations of applying “storing data signals from a sensor”, “processing the data signals and creating a descriptive representation of the sensed data”, “obtaining features from the descriptive representation of the data and identifying data patterns and a characteristic or variable”, and “estimating the probability of food intake based on identified features” etc. Independent Claim(s) 1 and 16, and dependent claims 2-4, 6-15, 17, and 19-25 are recited at a high level of generality and amount to no more than mere instructions to apply the exception using generic computer components in a vehicle. This/these limitation(s) do/does not impose any meaningful limits on practicing the abstract idea, and therefore do/does not integrate the abstract idea into a practical application (see MPEP 2106.05(f)).
Further, the additional limitations beyond the abstract idea identified above, serves merely to generally link the use of the judicial exception to a particular technological environment or field of use. Specifically, it/they serve(s) to limit the application of the abstract idea to computerized environments (e.g., storing data signals from a sensor, processing the data signals and creating a descriptive representation of the sensed data, obtaining features from the descriptive representation of the data and identifying data patterns and a characteristic or variable, and estimating the probability of food intake based on identified features etc.). This/these limitation(s) do/does not impose any meaningful limits on practicing the abstract idea, and therefore do/does not integrate the abstract idea into a practical application (see MPEP 2106.05(h)).
The recited additional element(s) of storing data signals from a sensor, processing the data signals and creating a descriptive representation of the sensed data, obtaining features from the descriptive representation of the data and identifying data patterns and a characteristic or variable, and estimating the probability of food intake based on identified features (Claim(s) 1 and 16), additionally and/or alternatively simply append insignificant extra-solution activity to the judicial exception, (e.g., mere pre-solution activity, such as data gathering, in conjunction with an abstract idea). This/these limitation(s) do/does not impose any meaningful limits on practicing the abstract idea, and therefore do/does not integrate the abstract idea into a practical application. (See MPEP 2106.05(g)).
Dependent claims 2-4, 6-15, 17, and 19-25 fail to include any additional elements. In other words, each of the limitations/elements recited in respective dependent claims is/are further part of the abstract idea as identified by the Examiner for each respective dependent claim (i.e. they are part of the abstract idea recited in each respective claim).
The Examiner has therefore determined that the additional elements, or combination of additional elements, do not integrate the abstract idea into a practical application. Accordingly, the claim(s) is/are directed to an abstract idea (Step 2A – Prong two: NO).
Step 2B: In step 2B, the claims are analyzed to determine whether any additional element, or combination of additional elements, is/are sufficient to ensure that the claims amount to significantly more than the judicial exception. This analysis is also termed a search for an "inventive concept." An "inventive concept" is furnished by an element or combination of elements that is recited in the claim in addition to (beyond) the judicial exception, and is sufficient to ensure that the claim as a whole amounts to significantly more than the judicial exception itself.
As discussed above in “Step 2A – Prong 2”, the identified additional elements in independent claim(s) 1 and 16, and dependent claims 2-4, 6-15, 17, and 19-25 are equivalent to adding the words “apply it” on a generic computer, and/or generally link the use of the judicial exception to a particular technological environment or field of use. Therefore, the claims as a whole do not amount to significantly more than the judicial exception itself.
The recited additional element(s) of storing data signals from a sensor, processing the data signals and creating a descriptive representation of the sensed data, obtaining features from the descriptive representation of the data and identifying data patterns and a characteristic or variable, and estimating the probability of food intake based on identified features (Claim(s) 1 and 16), additionally and/or alternatively simply append insignificant extra-solution activity to the judicial exception, (e.g., mere pre-solution activity, such as data gathering, in conjunction with an abstract idea) i.e. selecting users (i.e. using a user interface) is similar to “Receiving or transmitting data over a network, e.g., using the Internet to gather data”, is a well-understood, routine, and conventional function when it is claimed in a merely generic manner (as it is here) (See MPEP 2106.05(d) (II)).
This conclusion is based on a factual determination. Applicant’s own disclosure in paragraphs [00033], [00073], and [00097] acknowledges that “The proposed invention aims to detect meal intake events using different pre- processing, modeling, and post-processing configurations. These configurations can be adapted to use a variety of sensor modalities, sensor positions, and computational resources”, “wearable device (e.g., smartwatch) has the computational capacity to run components to perform all steps of the proposed invention. In this embodiment, a person 101 wearing a wearable device 102 needs only the smart device to run the whole proposed method”, and “The exemplificative embodiments described herein may be implemented using hardware, software, or any combination thereof and may be implemented in one or more computer systems or other processing systems” (i.e. conventional nature of using a computer and/or computer program). This additional element therefore does not ensure the claim amounts to significantly more than the abstract idea.
Viewing the additional limitations in combination also shows that they fail to ensure the claims amount to significantly more than the abstract idea. When considered as an ordered combination, the additional components of the claims add nothing that is not already present when considered separately, and thus simply append the abstract idea with words equivalent to “apply it” on a generic computer and/or mere instructions to implement the abstract idea on a generic computer or/and append the abstract idea with insignificant extra solution activity associated with the implementation of the judicial exception, and/or simply appending well-understood, routine, conventional activities previously known to the industry, specified at a high level of generality, to the judicial exception.
The dependent claims 2-4, 6-15, 17, and 19-25 are dependent from claims 1 and 16 and include all the limitations of the independent claims, but fail to include any additional elements. In other words, each of the limitations/elements recited in respective independent claims is/are further part of the abstract idea as identified by the Examiner for each respective dependent claim (i.e. they are part of the abstract idea recited in each respective claim). Therefore, the dependent claims recite the same abstract idea. The limitations of the dependent claims fail to amount to significantly more than the judicial exception. For example:
The limitations of claims 2, 4, 6, 8, 9, 11, 12, 17, 20, 21, 23, and 24, recite clarifications of utilizing one of a selection of sensors, filtering and normalizing signals from the sensor(s), segmenting the signals, extracting analytical features, using probability to estimate a dominant hand, selecting one of two processing methods to further analyze gathered data and data signals, and detecting changes in a transition of events. Such clarifications, under their broadest reasonable interpretation, are merely defining/selecting a type of data to be manipulated which, per MPEP 2106.05(g), is insignificant extra-solution activity. Therefore, the limitations fail to provide any teaching that integrates the judicial exceptions into a practical application or amount to significantly more than the judicial exception. For this reason, the analysis performed on the independent claims is also applicable on these claims.
The limitations of claim 3, 7, 10, 13, 14, 15, 19, 22, and 25 recite clarifications of augmenting data, extract features from digital data signals using machine learning, forming a time series and outputting an adjusted probability curve, recording (food intake) events, storing (detected food intake) events, storing data representation(s) in a buffer, and using a non-transitory computer-readable storage medium for storing computer-readable instructions to perform the claimed method. The limitations are further instructions for applying the judicial exceptions with a generic computing device/interface acting as an intermediary for performing the abstract ideas of storing data signals from a sensor, processing the data signals and creating a descriptive representation of the sensed data, obtaining features from the descriptive representation of the data and identifying data patterns and a characteristic or variable, and estimating the probability of food intake based on identified features, see MPEP 2106.05(f). Therefore, the limitations fail to provide any teaching that integrates the judicial exceptions into a practical application or amount to significantly more than the judicial exception. For this reason, the analysis performed on the independent claims is also applicable on these claims.
The Examiner has therefore determined that no additional element, or combination of additional claims elements is/are sufficient to ensure the claim(s) amount to significantly more than the abstract idea identified above (Step 2B: NO).
Therefore, claims 1-4, 6-17, and 19-25 are not eligible subject matter under 35 USC 101.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-4, 6, 7, 10, 12, 16, 17, 19, 22, 24, and 25 are rejected under 35 U.S.C. 103 as being unpatentable over Kyritsis, Konstantinos et al., “A Data Driven End-to-End Approach for In-the-Wild Monitoring of Eating Behavior Using Smartwatches”, January 2021, Vol. 25, No. 1, Pages 22-34 IEEE [online]: Journal of Biomedical and Health Informatics [retrieved on 2026-07-15]. Retrieved from: IEEE Xplore, in view of Stankoski, Simon et al., “Smartwatch-Based Eating Detection: Data Selection for Machine Learning from Imbalanced Data with Imperfect Labels”, March 2021, Vol. 21, No. 5, pages 1-25, MDPI [online]: Sensors [retrieved on 2026-07-15], Retrieved from: MDPI, DOI: 10.3390/s21051902.
Regarding Claim 1, Kyritsis discloses a system of detecting food intake events from wearable devices (Pg. 23 Lines 21-24 In this paper we propose a complete framework for automatically measuring eating (food intake event) behavior from in-the-wild collected inertial data (acceleration and orientation velocity) using a smartwatch (wearable device)), the system comprising:
a signal bank to store physiological digital data signals collected via a wearable device including at least one sensor to sense at least one physiological parameter of a user (Pg. 25 Lines 34-37 In this section we present a method for processing the raw triaxial acceleration and orientation velocity signals that originate from a commercial smartwatch (wearable device) with the aim of detecting bite events, Pg. 28 Col. 1 Lines 19-20 All datasets contain the triaxial acceleration and orientation velocity (6 DoF) signals, Pg. 28 Col. 2 Line 1 originating from a commercial smartwatch, Pg. 33 Lines 9-11 Another technical limitation regarding the recording of IMU signals (recording or “banking” signals) throughout the day is the high battery consumption when capturing the gyroscope sensor);
a preprocessor module to process the physiological digital data signals stored on the signal bank and to create a descriptive representation of the sensed at least one physiological parameter (Pg. 23 Lines 36-60 Their processing pipeline starts by preprocessing the acceleration and orientation velocity streams, continues with the extraction of 66 statistical features using a sliding window approach and a classification scheme to characterize windows as feeding or non-feeding gestures);
a feature extractor module including an automatically learned feature extractor and an analytical feature extractor to obtain features from the descriptive representation (Pg. 24 Col. 1 Lines 15-21 After a preprocessing step, the authors follow a feature extraction scheme on the data of each sensor before performing early fusion using an relatively shallow Artificial Neural Network (ANN) consisted by an input, a single hidden and an output layer. Results on their dataset of 12 subjects wearing the sensing platform for 24 hours reveal that the system is able to detect food intakes (feature) with an accuracy of 0.898, Pg. 24 Lines Col. 2 32-38 A data-driven approach for detecting food intake events (i.e. bites) (automatically learned feature extractor) during the course of a meal using an end-to-end NN with both convolutional and recurrent layers (Section III). An algorithm for the temporal localization of eating episodes (analytical feature extractor) using the distribution of bite detections produced by the end-to-end NN (Section IV)); and
a probability estimator module to determine whether a food intake event of the user occurs based on the obtain features (Pg. 27 Fig. 2 Overall pipeline of the in-meal bite detection (food intake event of the user occurs) approach… The network processes the M × 6 data, (essentially R from Section III-A), and outputs the N-dimensional bite probability vector p (probability estimator), where N = M/4 due to the two max pooling operations).
However, Kyritsis is not relied upon disclosing a feature extractor module including an automatically learned feature extractor and an analytical feature extractor to obtain features from the descriptive representation, the feature extractor module being configured to automatically select at least one of the automatically learned feature extractor or the analytical feature extractor.
Stankoski teaches a feature extractor module including an automatically learned feature extractor and an analytical feature extractor to obtain features from the descriptive representation, the feature extractor module being configured to automatically select at least one of the automatically learned feature extractor or the analytical feature extractor (Fig. 3, Pg. 2 Lines 28-31 The first step automatically cleans (automatically selects) the eating class from non-eating instances. The second step selects representative non-eating instances that are difficult to distinguish and includes them in the training set, Pg. 6 Lines 10-15 In this section, we describe the initial steps of our eating detection method (feature extractor module). The preprocessing technique includes various filtering steps from which additional streams are extracted. In addition, we describe how we extract virtual streams from predictions of DL models. Finally, we describe in detail the features extracted from each stream used in the pipeline and the procedure for selecting only the most relevant ones. The steps described in this section are shown in Figure 3, Pg. 7 Lines 24-27 Beside the raw sensor signals, we derived two virtual sensor streams (feature extractors) that are useful for activity recognition tasks. We calculated the magnitude for both the accelerometer and gyroscope signals, which provide general information about the intensity of hand movement regardless of the direction of that movement).
Kyritsis and Stankoski are both considered to be analogous to the claimed invention, because they are in the same field of eating detection/monitoring using smartwatches. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the applicant’s invention for modifying a system of detecting food intake events from wearable devices, as disclosed by Kyritsis, further including a feature extractor module including an automatically learned feature extractor and an analytical feature extractor to obtain features from the descriptive representation, the feature extractor module being configured to automatically select at least one of the automatically learned feature extractor or the analytical feature extractor, as taught by Stankoski for the purpose of identifying temporal patterns of food and nutrient intake accurately in order to suggest interventions for a healthy lifestyle (Stankoski, Pg. 2 Lines 1-6).
Regarding Claim 2, Kyritsis discloses wherein the at least one sensor is an inertial sensor, a temperature sensor, a bioelectrical impedance sensor, or a photoplethysmography sensor (Para. 5 Inertial information from wrist-mounted sensors (inertial sensors) can also be used in order to detect eating [21], [22] and liquid intake episodes [23]).
Regarding Claim 3, Kyritsis discloses wherein the preprocessor module comprises a data augmenter to apply data augmentation on the signal bank in order to create representative data to describe the food intake event (Pg. 31 Lines 47-58 Table V shows the effects of synthetically augmenting the dataset during the training process of the proposed method regarding the intake cycle detection performance).
Regarding Claim 4, Kyritsis discloses wherein the preprocessor module further comprises:
a filter and a normalizer to adjust a signal received from at least one sensor into a numerical representation, wherein the filter is configured to perform noise reduction, signal smoothing and/or to suppress unwanted signal frequencies, and the normalizer is configured to adjust a range of the signal (Pg. 26 Lines 7-22 To deal with the small fluctuations (noise) introduced by the sensors’ hardware we individually convolve each (applied to each stream, hence normalized) of the 3D accelerometer and gyroscope streams with a moving average filter (perform noise reduction)… the final preprocessing step is to attenuate (suppress) the Earth component. We achieve this by individually convolving the ax, ay and az components of R with a high-pass finite impulse response (FIR) filter).
However, Kyritsis is not relied upon disclosing the normalizer is configured to adjust a range of the signal.
Stankoski teaches the normalizer is configured to adjust a range of the signal (Pg. 7 Lines 2-5 The data collected for this study include signals from an inertial sensor with 6 degrees of freedom, i.e., three signals from an accelerometer and three signals from a gyroscope. The first step in the preprocessing pipeline involves interpolation of the signals to a fixed frequency of 100Hz, Pg. 8 Line 22 In the inception blocks, the batch normalization [66] is applied after the convolutions, Pg. 13 Lines 33-34 The second stage takes temporal information into account and smooths the predictions)
Regarding Claim 6, Kyritsis discloses wherein the preprocessor module further comprises a segmenter configured to subdivide sampled signals into segments (Pg. 23 Lines 6-9 The work presented in [24] uses a novel segmentation technique (subdivide sample signals into segments) for the detection and recognition of eating and drinking gestures from continuous IMU data recorded in free-living conditions).
However, Kyritsis is not relied upon disclosing the segmenter is further configured to subdivide the segments into windows.
Stankoski teaches the segmenter is further configured to subdivide the segments into windows (Pg. 7 Lines 28-34 The next step in the pipeline was to select the appropriate window size for the sliding window segmentation technique (subdivide segments into windows). Previous studies related to eating detection typically used window sizes of up to 2 s, as their approach was based on precisely labelled intake gestures. However, our data contain only labelled eating segments. To increase the probability that an intake gesture is captured in a window, we used a longer window size. We determined the optimal window size empirically. The signals were segmented using a window size of 15 s with a 3-s slide between consecutive windows).
Regarding Claim 7, Kyritsis discloses wherein the Automatically Learned Feature Extractor comprises a machine learning model configured to extract representative features from the physiological digital data signals (Pg. 24 Lines 2-5 This allows the convolutional part of the network to discover the optimal features representations before performing temporal modeling using an LSTM, Pg. 24 Lines 15-19 After a preprocessing step, the authors follow a feature extraction scheme on the data of each sensor before performing early fusion using an relatively shallow Artificial Neural Network (ANN) consisted by an input, a single hidden and an output layer, Pg. 24 Lines 21-32 The same dataset of 12 people was also used in [35] as a benchmark for three different ensemble techniques, boosting, bootstrap aggregation and stacking, trained with three different weak classifiers (examples of machine learning models): i) Decision Trees, ii) Linear Discriminant Analysis and iii) Logistic Regression. Following the same feature extraction scheme as in [29] the authors of [35] show that using bootstrap aggregation with Fisher LDA as the base classifier leads to an overall improvement of 0.04 (0.938 accuracy instead of 0.898). In both works however, [29] and [35], the authors do not provide a final estimate of the meal start and end points; instead they classify segments as food intake or not).
Regarding Claim 10, Kyritsis discloses further comprising a postprocessor module configured to form a time series on the estimated probability and to output an adjusted probability curve (Pg. 27 Lines 10-23 The next step is to construct the timeseries s(n) with n =1,...,N, that spans the entire duration of the recording and is equal to one only at the moments where a bite is detected and zero everywhere else… The time series s(n) is then convolved with a Gaussian filter of length and standard deviation equal to fs·240 4 and fs·45 4 samples. The latter is important as it smooths s(n) and closes the gaps between groups of bites that are close to each other, similar to a morphological closing operation, which is a phenomenon present in long meals).
Regarding Claim 12, Kyritsis discloses wherein the postprocessor module further comprises a change point detector configured to detect changes in a transition of events (Pg. 27 Lines 15-20 We perform bite detection by initially replacing with zeros the elements of p that are lower than a probability threshold λp. Next, by performing a local maxima search, with a minimum distance between two (events) consecutive peaks (change point detection) set at 2 seconds, on the thresholded series p we obtain the set of detected bites B = {b1,...,bL} (transition of events), Pg. 33 Lines 22-27 Initially, we follow a representation learning approach that includes an end-to-end NN with both convolutional and recurrent layers to detect bite events. Next we use the distribution of detected bites throughout the day to temporally localize meals by detecting their start and end points, using signal processing algorithms).
Regarding Claim 16, Kyritsis discloses a method of detecting food intake events from wearable devices, the method comprising: storing physiological digital data signals in a signal bank, the physiological digital data signals being collected via a wearable device including at least one sensor to sense at least one physiological parameter of a user;
processing, with a preprocessor module, the physiological digital data signals stored on the signal bank (Pg. 25 Lines 34-37 In this section we present a method for processing the raw triaxial acceleration and orientation velocity signals that originate from a commercial smartwatch (wearable device) with the aim of detecting bite events, Pg. 33 Lines 9-11 Another technical limitation regarding the recording of IMU signals (recording or “banking” signals) throughout the day is the high battery consumption when capturing the gyroscope sensor), and
creating a descriptive representation of the sensed at least one physiological parameter (Pg. 23 Lines 36-60 Their processing pipeline starts by preprocessing the acceleration and orientation velocity streams, continues with the extraction of 66 statistical features using a sliding window approach and a classification scheme to characterize windows as feeding or non-feeding gestures);
obtaining, with a feature extractor module, features from the descriptive representation, the feature extractor module including an automatically learned feature extractor and an analytical feature extractor (Pg. 24 Col. 1 Lines 15-21 After a preprocessing step, the authors follow a feature extraction scheme on the data of each sensor before performing early fusion using an relatively shallow Artificial Neural Network (ANN) consisted by an input, a single hidden and an output layer. Results on their dataset of 12 subjects wearing the sensing platform for 24 hours reveal that the system is able to detect food intakes (feature) with an accuracy of 0.898, Pg. 24 Lines Col. 2 32-38 A data-driven approach for detecting food intake events (i.e. bites) (automatically learned feature extractor) during the course of a meal using an end-to-end NN with both convolutional and recurrent layers (Section III). An algorithm for the temporal localization of eating episodes (analytical feature extractor) using the distribution of bite detections produced by the end-to-end NN (Section IV)); and
generating, with a probability estimator module, an estimated probability to determine whether a food intake event of the user occurs based on the obtained features (Pg. 27 Fig. 2 Overall pipeline of the in-meal bite detection (food intake event of the user occurs) approach… The network processes the M × 6 data, (essentially R from Section III-A), and outputs the N-dimensional bite probability vector p (probability estimator), where N = M/4 due to the two max pooling operations).
However, Kyritsis is not relied upon disclosing obtaining, with a feature extractor module, features from the descriptive representation, the feature extractor module including an automatically learned feature extractor and an analytical feature extractor and being configured to automatically select at least one of the automatically learned feature extractor or the analytical feature extractor.
Stankoski teaches obtaining, with a feature extractor module, features from the descriptive representation, the feature extractor module including an automatically learned feature extractor and an analytical feature extractor and being configured to automatically select at least one of the automatically learned feature extractor or the analytical feature extractor (Fig. 3, Pg. 2 Lines 28-31 The first step automatically cleans (automatically selects) the eating class from non-eating instances. The second step selects representative non-eating instances that are difficult to distinguish and includes them in the training set, Pg. 6 Lines 10-15 In this section, we describe the initial steps of our eating detection method (feature extractor module). The preprocessing technique includes various filtering steps from which additional streams are extracted. In addition, we describe how we extract virtual streams from predictions of DL models. Finally, we describe in detail the features extracted from each stream used in the pipeline and the procedure for selecting only the most relevant ones. The steps described in this section are shown in Figure 3, Pg. 7 Lines 24-27 Beside the raw sensor signals, we derived two virtual sensor streams (feature extractors) that are useful for activity recognition tasks. We calculated the magnitude for both the accelerometer and gyroscope signals, which provide general information about the intensity of hand movement regardless of the direction of that movement).
Regarding Claim 17, Kyritsis discloses wherein the preprocessor module further comprises a filter and a normalizer, and the method further comprises:
adjusting, with the filter and the normalizer, a data signal received from at least one sensor into a numerical representation; and performing noise reduction, data smoothing, and suppressing unwanted signal frequencies with the filter (Pg. 26 Lines 7-22 To deal with the small fluctuations introduced by the sensors’ hardware we individually convolve each (applied to each stream, hence normalized) of the 3D accelerometer and gyroscope streams with a moving average filter (noise reduction)… the final preprocessing step is to attenuate (suppress) the Earth component. We achieve this by individually convolving the ax, ay and az components of R with a high-pass finite impulse response (FIR) filter).
However, Kyritsis is not relied upon disclosing adjusting, with the normalizer, a range of the data signal.
Stankoski teaches adjusting, with the normalizer, a range of the data signal (Pg. 7 Lines 2-5 The data collected for this study include signals from an inertial sensor with 6 degrees of freedom, i.e., three signals from an accelerometer and three signals from a gyroscope. The first step in the preprocessing pipeline involves interpolation of the signals to a fixed frequency of 100Hz, Pg. 8 Line 22 In the inception blocks, the batch normalization [66] is applied after the convolutions, Pg. 13 Lines 33-34 The second stage takes temporal information into account and smooths the predictions).
Regarding Claim 19, Kyritsis discloses wherein the Automatically Learned Feature Extractor comprises a machine learning model configured to extract representative features from the physiological digital data signals (Pg. 24 Lines 2-5 This allows the convolutional part of the network to discover the optimal features representations before performing temporal modeling using an LSTM, Pg. 24 Lines 15-19 After a preprocessing step, the authors follow a feature extraction scheme on the data of each sensor before performing early fusion using an relatively shallow Artificial Neural Network (ANN) consisted by an input, a single hidden and an output layer, Pg. 24 Lines 21-32 The same dataset of 12 people was also used in [35] as a benchmark for three different ensemble techniques, boosting, bootstrap aggregation and stacking, trained with three different weak classifiers (examples of machine learning models): i) Decision Trees, ii) Linear Discriminant Analysis and iii) Logistic Regression. Following the same feature extraction scheme as in [29] the authors of [35] show that using bootstrap aggregation with Fisher LDA as the base classifier leads to an overall improvement of 0.04 (0.938 accuracy instead of 0.898). In both works however, [29] and [35], the authors do not provide a final estimate of the meal start and end points; instead they classify segments as food intake or not).
Regarding Claim 22, Kyritsis discloses c further comprising: forming, with a postprocessor module, a time series on the estimated probability and to output an adjusted probability curve (Pg. 27 Lines 10-23 The next step is to construct the timeseries s(n) with n =1,...,N, that spans the entire duration of the recording and is equal to one only at the moments where a bite is detected and zero everywhere else… The time series s(n) is then convolved with a Gaussian filter of length and standard deviation equal to fs·240 4 and fs·45 4 samples. The latter is important as it smooths s(n) and closes the gaps between groups of bites that are close to each other, similar to a morphological closing operation, which is a phenomenon present in long meals).
Regarding Claim 24, Kyritsis discloses further comprising detecting, with a postprocessor module which includes a change point detector, changes in a transition of events (Pg. 27 Lines 15-20 We perform bite detection by initially replacing with zeros the elements of p that are lower than a probability threshold λp. Next, by performing a local maxima search, with a minimum distance between two (events) consecutive peaks (change point detection) set at 2 seconds, on the thresholded series p we obtain the set of detected bites B = {b1,...,bL} (transition of events), Pg. 33 Lines 22-27 Initially, we follow a representation learning approach that includes an end-to-end NN with both convolutional and recurrent layers to detect bite events. Next we use the distribution of detected bites throughout the day to temporally localize meals by detecting their start and end points, using signal processing algorithms).
Regarding Claim 25, Kyritsis discloses a non-transitory computer-readable storage medium storing computer-readable instructions, when performed by a processor, cause a computer to perform the method defined in claim 19 (Pg. 23 Lines 21-24 In this paper we propose a complete framework for automatically measuring eating behavior from in-the-wild collected inertial data (acceleration and orientation velocity) using a smartwatch (non-transitory computer-readable storage medium)).
Claims 5 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Kyritsis, Konstantinos et al., “A Data Driven End-to-End Approach for In-the-Wild Monitoring of Eating Behavior Using Smartwatches”, January 2021, Vol. 25, No. 1, Pages 22-34 IEEE [online]: Journal of Biomedical and Health Informatics [retrieved on 2026-07-15]. Retrieved from: IEEE Xplore, in view of Stankoski, Simon et al., “Smartwatch-Based Eating Detection: Data Selection for Machine Learning from Imbalanced Data with Imperfect Labels”, March 2021, Vol. 21, No. 5, pages 1-25, MDPI [online]: Sensors [retrieved on 2026-07-15], Retrieved from: MDPI, DOI: 10.3390/s21051902, and in further view of Tu et al. (US 2016/0018872 A1).
Regarding Claim 5, Kyritsis is not relied upon disclosing wherein the preprocessor module further comprises: a hand-laterally detector configured to, in case the wearable device is positioned on an arm of the user, detect in which user's arm the device wearable is being used, and a transformer is configured to transform inertial physiological data based on an output of the hand-laterally detector.
Tu teaches wherein the preprocessor module further comprises:
a hand-laterally detector configured to, in case the wearable device is positioned on an arm of the user, detect in which user's arm the device wearable is being used, and a transformer is configured to transform inertial physiological data based on an output of the hand-laterally detector ([0141] Arm position estimation unit 1406 (a hand-laterally detector) can receive motion sensor data and determine from the data an estimated arm position of an arm on which the device is presumed to be worn. For instance, arm position estimation unit 1406 can implement block 1314 of process 1300 described above. Arm position estimation unit 1406 can provide information about the arm position (e.g., an estimated arm position or a change in arm position) to focus determination unit 1408)
Kyritsis and Tu are both considered to be analogous to the claimed invention, because they are in the same field of smartwatches. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the applicant’s invention for modifying a system of detecting food intake events from wearable devices, as disclosed by Kyritsis, further including a hand-laterally detector configured to, in case the wearable device is positioned on an arm of the user, detect in which user's arm the device wearable is being used, and a transformer is configured to transform inertial physiological data based on an output of the hand-laterally detector, as taught by Tu for the purpose of determining from the data an estimated arm position of an arm on which the device is presumed to be worn (Tu, [0229]).
Regarding Claim 18, Kyritsis is not relied upon disclosing wherein the preprocessor module further comprises a hand-laterally detector and a transformer, and the method further comprises: in case the wearable device is positioned on an arm of the user, detecting in which user's arm the device wearable is being used; and transforming inertial physiological data based on an output of the hand-laterally detector.
Tu teaches wherein the preprocessor module further comprises a hand-laterally detector and a transformer, and the method further comprises:
in case the wearable device is positioned on an arm of the user, detecting in which user's arm the device wearable is being used; and transforming inertial physiological data based on an output of the hand-laterally detector ([0141] Arm position estimation unit 1406 (a hand-laterally detector) can receive motion sensor data and determine from the data an estimated arm position of an arm on which the device is presumed to be worn. For instance, arm position estimation unit 1406 can implement block 1314 of process 1300 described above. Arm position estimation unit 1406 can provide information about the arm position (e.g., an estimated arm position or a change in arm position) to focus determination unit 1408).
Claims 8 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Kyritsis, Konstantinos et al., “A Data Driven End-to-End Approach for In-the-Wild Monitoring of Eating Behavior Using Smartwatches”, January 2021, Vol. 25, No. 1, Pages 22-34 IEEE [online]: Journal of Biomedical and Health Informatics [retrieved on 2026-07-15]. Retrieved from: IEEE Xplore, in view of Stankoski, Simon et al., “Smartwatch-Based Eating Detection: Data Selection for Machine Learning from Imbalanced Data with Imperfect Labels”, March 2021, Vol. 21, No. 5, pages 1-25, MDPI [online]: Sensors [retrieved on 2026-07-15], Retrieved from: MDPI, DOI: 10.3390/s21051902, and in further view of Ding, Fengqian et al., “The Entropy-Based Time Domain Feature Extraction for Online Concept Drift Detection”, November 30, 2019, Pages 1-22, MDPI [online]: Entropy [retrieved on 2026-07-15], Retrieved from: MDPI, DOI: 10:3390/e21121187.
Regarding Claim 8, Kyritsis discloses wherein the Analytical Feature Extractor is configured to extract multiple analytical features (Pg. 24 Lines 2-5 This allows the convolutional part of the network to discover the optimal features representations before performing temporal modeling using an LSTM, Pg. 24 Lines 15-19 After a preprocessing step, the authors follow a feature extraction scheme on the data of each sensor before performing early fusion using an relatively shallow Artificial Neural Network (ANN) consisted by an input, a single hidden and an output layer, Pg. 27 Lines 6-9 After inspecting the learning curves regarding the training accuracy and cross entropy loss on the training set we observed only a marginal improvement after the 5th epoch; therefore, we selected 5 as the number of epochs for training the network).
However, Kyritsis is not relied upon disclosing wherein the Analytical Feature Extractor is configured to extract multiple analytical features and is further configured to automatically select at least one of entropy, statistical or complexity features.
Ding teaches wherein the Analytical Feature Extractor is configured to extract multiple analytical features and is further configured to automatically select at least one of entropy, statistical or complexity features (Abstract In this paper, a novel method called online entropy-based time domain feature extraction (ETFE) for concept drift detection is proposed. Firstly, the empirical mode decomposition based on extrema symmetric extension is used to decompose time series, where features in various time scales can be adaptively extracted).
Kyritsis and Ding are both considered to be analogous to the claimed invention, because they are in the same field of feature extraction. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the applicant’s invention for modifying a system of detecting food intake events from wearable devices, as disclosed by Kyritsis, further including wherein the Analytical Feature Extractor is configured to extract multiple analytical features and is further configured to automatically select at least one of entropy, statistical or complexity features, as taught by Ding for the purpose of ensuring time-domain features in different time scales can be effectively extracted, have good signal-to-noise ratios, and be used to determine the concept drift (Ding, Pg. 20 Lines 6-15).
Regarding Claim 20, Kyritsis discloses wherein the Analytical Feature Extractor is configured to extract multiple analytical features (Pg. 24 Lines 2-5 This allows the convolutional part of the network to discover the optimal features representations before performing temporal modeling using an LSTM, Pg. 24 Lines 15-19 After a preprocessing step, the authors follow a feature extraction scheme on the data of each sensor before performing early fusion using an relatively shallow Artificial Neural Network (ANN) consisted by an input, a single hidden and an output layer, Pg. 27 Lines 6-9 After inspecting the learning curves regarding the training accuracy and cross entropy loss on the training set we observed only a marginal improvement after the 5th epoch; therefore, we selected 5 as the number of epochs for training the network)).
However, Kyritsis is not relied upon disclosing wherein the Analytical Feature Extractor is configured to extract multiple analytical features and is further configured to automatically select at least one of entropy, statistical or complexity features.
Ding teaches wherein the Analytical Feature Extractor is configured to extract multiple analytical features and is further configured to automatically select at least one of entropy, statistical or complexity features (Abstract In this paper, a novel method called online entropy-based time domain feature extraction (ETFE) for concept drift detection is proposed. Firstly, the empirical mode decomposition based on extrema symmetric extension is used to decompose time series, where features in various time scales can be adaptively extracted).
Claims 9 and 21 are rejected under 35 U.S.C. 103 as being unpatentable over Kyritsis, Konstantinos et al., “A Data Driven End-to-End Approach for In-the-Wild Monitoring of Eating Behavior Using Smartwatches”, January 2021, Vol. 25, No. 1, Pages 22-34 IEEE [online]: Journal of Biomedical and Health Informatics [retrieved on 2026-07-15]. Retrieved from: IEEE Xplore, in view of Stankoski, Simon et al., “Smartwatch-Based Eating Detection: Data Selection for Machine Learning from Imbalanced Data with Imperfect Labels”, March 2021, Vol. 21, No. 5, pages 1-25, MDPI [online]: Sensors [retrieved on 2026-07-15], Retrieved from: MDPI, DOI: 10.3390/s21051902, and in further view of Pavlov et al. (WO 2021/040292 A1).
Regarding Claim 9, Kyritsis is not relied upon disclosing wherein the probability estimator module further comprises a dominant hand detector configured to detect a dominant hand of the user.
Pavlov teaches wherein the probability estimator module further comprises a dominant hand detector configured to detect a dominant hand of the user ([16] At the same time, movements associated with eating are well defined if the smart watch is on the dominant hand (right-handed, left-handed) and worse defined if the smart watch is on the non-dominant hand, [149] FIG. 9 shows the results of the detection of food intakes. For a dominant hand, the boxplot is shown on the right. The F-measure is used (F1-score is a joint estimation of accuracy, precision and completeness, recall), it is equal 0.7, on average. The boxplot on the right is for a non-dominant hand; here the F-measure is lower than and equal to 0.63, [176] The personalized system comprises a smart watch including an accelerometer, a gyroscope, and a PPG sensor configured to sense and provide signals related to the user's physical activity, heart rate, blood oxygenation level, and sleep quality (where, location data which is based on the data, the user's profile, and geolocation is used to estimate food intake times, eating habits, and physical activity patterns, considering whether the user is right-handed or left-handed, and the PPG sensor is positioned on a display side of the smart watch and configured to form a PPG signal by touching with the user's finger), a unit for processing signals from the PPG sensor to estimate vascular stiffness which defines vascular age, and a unit configured to generate recommendations for the user's eating habit pattern, detecting the relevance between the user's behavior and the user's physiological changes based on signals received from the smart watch, select the most appropriate recommendations, and display the recommendations on the display of the smart watch).
Kyritsis and Pavlov are both considered to be analogous to the claimed invention, because they are in the same field of electronic devices (including smartwatches) for providing users with personalized information. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the applicant’s invention for modifying a system of detecting food intake events from wearable devices, as disclosed by Kyritsis, further including wherein the probability estimator module further comprises a dominant hand detector configured to detect a dominant hand of the user, as taught by Pavlov for the purpose of having well defined movements associated with eating (Pavlov, [16]).
Regarding Claim 21, Kyritsis is not relied upon disclosing wherein the probability estimator module further comprises a dominant hand detector configured to detect a dominant hand in a feed action of the user.
Pavlov teaches wherein the probability estimator module further comprises a dominant hand detector configured to detect a dominant hand in a feed action of the user ([16] At the same time, movements associated with eating are well defined if the smart watch is on the dominant hand (right-handed, left-handed) and worse defined if the smart watch is on the non-dominant hand, [149] FIG. 9 shows the results of the detection of food intakes. For a dominant hand, the boxplot is shown on the right. The F-measure is used (F1-score is a joint estimation of accuracy, precision and completeness, recall), it is equal 0.7, on average. The boxplot on the right is for a non-dominant hand; here the F-measure is lower than and equal to 0.63, [176] The personalized system comprises a smart watch including an accelerometer, a gyroscope, and a PPG sensor configured to sense and provide signals related to the user's physical activity, heart rate, blood oxygenation level, and sleep quality (where, location data which is based on the data, the user's profile, and geolocation is used to estimate food intake times, eating habits, and physical activity patterns, considering whether the user is right-handed or left-handed, and the PPG sensor is positioned on a display side of the smart watch and configured to form a PPG signal by touching with the user's finger), a unit for processing signals from the PPG sensor to estimate vascular stiffness which defines vascular age, and a unit configured to generate recommendations for the user's eating habit pattern, detecting the relevance between the user's behavior and the user's physiological changes based on signals received from the smart watch, select the most appropriate recommendations, and display the recommendations on the display of the smart watch).
Claims 11 and 23 are rejected under 35 U.S.C. 103 as being unpatentable over Kyritsis, Konstantinos et al., “A Data Driven End-to-End Approach for In-the-Wild Monitoring of Eating Behavior Using Smartwatches”, January 2021, Vol. 25, No. 1, Pages 22-34 IEEE [online]: Journal of Biomedical and Health Informatics [retrieved on 2026-07-15]. Retrieved from: IEEE Xplore, in view of Stankoski, Simon et al., “Smartwatch-Based Eating Detection: Data Selection for Machine Learning from Imbalanced Data with Imperfect Labels”, March 2021, Vol. 21, No. 5, pages 1-25, MDPI [online]: Sensors [retrieved on 2026-07-15], Retrieved from: MDPI, DOI: 10.3390/s21051902, and in further view of Sharma, Surya et al., “Top-Down Detection of Eating Episodes by Analyzing Large Windows of Wrist Motion Using a Convolutional Neural Network”, February 2022, Vol. 9, No. 2, pages 1-16, MDPI [online]: Bioengineering [retrieved on 2026-07-15], Retrieved from: MDPI, DOI: 10.3390/bioengineering9020070.
Regarding Claim 11, Kyritsis discloses wherein the postprocessor module is further configured to select at least one of a Thresholding-Based Postprocessor and a Heuristic-Based Postprocessor, wherein the Thresholding-Based Postprocessor is configured to select at least one of reference or hysteresis thresholding (Pg. 24 Lines 46-50 In more detail, the authors use a heuristic peak detector (heuristic-based postprocessor) based on the concept of hysteresis threshold to create potential meal segments and then follow a feature extraction scheme and a Naive Bayes classification scheme to classify the segmented sequences as eating or not, Pg. 26 Lines 20-22 The filter’s cutoff frequency fc and length of the tap delay line lhp, were set to 1 Hz and 512 samples respectively, Pg. 27 Col. 1 Lines 13-23 By processing (postprocessing) an M × 6 recording R that contains the preprocessed inertial observations, the inference end-to-end network outputs the N-length prediction series p, with N = M/4 due to the max pooling operations (Section III-B). We perform bite detection by initially replacing with zeros the elements of p that are lower than a probability threshold λp (thresholding-based postprocessor). Next, by performing a local maxima search, with a minimum distance between two consecutive peaks set at 2 seconds, on the thresholded series p we obtain the set of detected bites B = {b1,...,bL}. Each bi element is essentially the timestamp that corresponds to the i-th peak, Pg. 27 Col. 2 Lines 1-3 detecting the start and end time-points (meal time) of in-the-wild meals using the trained end-to-end network from Section III coupled with signal processing algorithms, Pg. 27 Col. 2 Lines 20-23 The latter is important as it smooths s(n) and closes the gaps (session cut-removing) between groups of bites that are close to each other, similar to a morphological closing operation, which is a phenomenon present in long meals).
However, Kyritsis is not relied upon disclosing wherein the postprocessor module is further configured to select at least one of a Thresholding-Based Postprocessor and a Heuristic-Based Postprocessor, wherein the Thresholding-Based Postprocessor is configured to select at least one of reference or hysteresis thresholding, and the heuristic-based postprocessor is configured to select at least one of a minimal meal time, an onset time compensator, an offset time compensator, and a session cut-removing module.
Sharma teaches wherein the postprocessor module is further configured to select at least one of a Thresholding-Based Postprocessor and a Heuristic-Based Postprocessor, wherein the Thresholding-Based Postprocessor is configured to select at least one of reference or hysteresis thresholding, and the heuristic-based postprocessor is configured to select at least one of a minimal meal time, an onset time compensator, an offset time compensator, and a session cut-removing module (Pg. 4 Lines 15-21 Figure 4 demonstrates the method. An episode is detected if the probability of eating is higher than threshold (Thresholding-Based Postprocessor) TS (start meal). A detected episode is ended if the probability of eating becomes lower than threshold TE (end meal). The use of two thresholds for hysteresis (Heuristic-Based postprocessor) helps to smooth episode detections in a manner similar to button debouncing [36], Pg. 6 Lines 7-10 We identify periods of eating using a hysteresis-based detector to reduce the effect of noise in the p(e) signal [36]. The pseudo-code for the detector is shown in Algorithm 1. The inputs are p(e), the probability of eating at time t output by the NN, and TS and TE, the hysteresis thresholds, Pg. 6 Lines 20-22 Finally, as all meals in CAD are longer than 1 min in length, we combine any two segments that are within 1 min of each other, and remove any segments (session cut-removing module) shorter than 1 min to help avoid oversegmentation, Pg. 7 Lines 16-20 We report the average difference between the meal start time (onset time compensator) and detected segment start time as the start boundary error, and the average difference between the meal end time (offset time compensator) and detected segment end time as the end boundary error. In the special case of self-reported meals overlapping multiple detected events (Figure 6b,c), for each self-reported meal, we use the start/end of the first/last overlapping detection).
Kyritsis and Sharma are both considered to be analogous to the claimed invention, because they are in the same field of electronic devices that detect food intake (eating). Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the applicant’s invention for modifying a system of detecting food intake events from wearable devices, as disclosed by Kyritsis, further including wherein the postprocessor module is further configured to select at least one of a Thresholding-Based Postprocessor and a Heuristic-Based Postprocessor, wherein the Thresholding-Based Postprocessor is configured to select at least one of reference or hysteresis thresholding, and the heuristic-based postprocessor is configured to select at least one of a minimal meal time, an onset time compensator, an offset time compensator, and a session cut-removing module, as taught by Sharma for the purpose reducing false positives by requiring a strong probability and improving boundary precision by allowing for a weaker probability at the end of detection. (Sharma, Pg. 6 Lines 13-16).
Regarding Claim 23, Kyritsis discloses wherein the postprocessor module is further configured to select at least one of a Thresholding-Based Postprocessor and a Heuristic-Based Postprocessor, wherein the Thresholding-Based Postprocessor is configured to select at least one reference or hysteresis thresholding (Pg. 24 Lines 46-50 In more detail, the authors use a heuristic peak detector (heuristic-based postprocessor) based on the concept of hysteresis threshold to create potential meal segments and then follow a feature extraction scheme and a Naive Bayes classification scheme to classify the segmented sequences as eating or not, Pg. 26 Lines 20-22 The filter’s cutoff frequency fc and length of the tap delay line lhp, were set to 1 Hz and 512 samples respectively, Pg. 27 Col. 1 Lines 13-23 By processing (postprocessing) an M × 6 recording R that contains the preprocessed inertial observations, the inference end-to-end network outputs the N-length prediction series p, with N = M/4 due to the max pooling operations (Section III-B). We perform bite detection by initially replacing with zeros the elements of p that are lower than a probability threshold λp (thresholding-based postprocessor). Next, by performing a local maxima search, with a minimum distance between two consecutive peaks set at 2 seconds, on the thresholded series p we obtain the set of detected bites B = {b1,...,bL}. Each bi element is essentially the timestamp that corresponds to the i-th peak, Pg. 27 Col. 2 Lines 1-3 detecting the start and end time-points (meal time) of in-the-wild meals using the trained end-to-end network from Section III coupled with signal processing algorithms, Pg. 27 Col. 2 Lines 20-23 The latter is important as it smooths s(n) and closes the gaps (session cut-removing) between groups of bites that are close to each other, similar to a morphological closing operation, which is a phenomenon present in long meals).
However, Kyritsis is not relied upon disclosing wherein the postprocessor module is further configured to select at least one of a Thresholding-Based Postprocessor and a Heuristic-Based Postprocessor, wherein the Thresholding-Based Postprocessor is configured to select at least one reference or hysteresis thresholding, and the Heuristic-Based Postprocessor is configured to select at least one of a minimal meal time, an onset time compensator, an offset time compensator, and a session cut-removing module.
Sharma teaches wherein the postprocessor module is further configured to select at least one of a Thresholding-Based Postprocessor and a Heuristic-Based Postprocessor, wherein the Thresholding-Based Postprocessor is configured to select at least one reference or hysteresis thresholding, and the Heuristic-Based Postprocessor is configured to select at least one of a minimal meal time, an onset time compensator, an offset time compensator, and a session cut-removing module (Pg. 4 Lines 15-21 Figure 4 demonstrates the method. An episode is detected if the probability of eating is higher than threshold (Thresholding-Based Postprocessor) TS (start meal). A detected episode is ended if the probability of eating becomes lower than threshold TE (end meal). The use of two thresholds for hysteresis (Heuristic-Based postprocessor) helps to smooth episode detections in a manner similar to button debouncing [36], Pg. 6 Lines 7-10 We identify periods of eating using a hysteresis-based detector to reduce the effect of noise in the p(e) signal [36]. The pseudo-code for the detector is shown in Algorithm 1. The inputs are p(e), the probability of eating at time t output by the NN, and TS and TE, the hysteresis thresholds, Pg. 6 Lines 20-22 Finally, as all meals in CAD are longer than 1 min in length, we combine any two segments that are within 1 min of each other, and remove any segments (session cut-removing module) shorter than 1 min to help avoid oversegmentation, Pg. 7 Lines 16-20 We report the average difference between the meal start time (onset time compensator) and detected segment start time as the start boundary error, and the average difference between the meal end time (offset time compensator) and detected segment end time as the end boundary error. In the special case of self-reported meals overlapping multiple detected events (Figure 6b,c), for each self-reported meal, we use the start/end of the first/last overlapping detection).
Claims 13-15 are rejected under 35 U.S.C. 103 as being unpatentable over Kyritsis, Konstantinos et al., “A Data Driven End-to-End Approach for In-the-Wild Monitoring of Eating Behavior Using Smartwatches”, January 2021, Vol. 25, No. 1, Pages 22-34 IEEE [online]: Journal of Biomedical and Health Informatics [retrieved on 2026-07-15]. Retrieved from: IEEE Xplore, and in further view of Sazonov et al. (US 2018/0242908 A1).
Regarding Claim 13, Kyritsis is not relied upon disclosing the system further comprising a logbook for recording the food intake events.
Sazonov teaches the system further comprising a logbook for recording the food intake events ([0117] The hand gesture sensor 212 on FIG. 2 may record signals indicating a gesture of bringing food to the mouth, [0150] Participants were required to keep a log of their eating episodes only).
Kyritsis and Sazonov are both considered to be analogous to the claimed invention, because they are in the same field of food intake monitoring devices. Therefore, it would have been obvious to someone of ordinary skill in the art before the effective filling date of the applicant’s invention for modifying a system of detecting food intake events from wearable devices, as disclosed by Kyritsis, further including the system further comprising a logbook for recording the food intake events, as taught by Sazonov for the purpose of identifying food intake and performing computerized statistical analysis on the data (Sazonov, [0063]).
Regarding Claim 14, Kyritsis is not relied upon disclosing the system further comprising a meal event buffer for storing the detected food intake events.
Sazonov teaches the system further comprising a meal event buffer for storing the detected food intake events ([0069] a jaw sensor configured to detect jaw motion and transmit a plurality of cycles of jaw sensor data to the data buffer (meal event buffer) for storage in the time domain).
Regarding Claim 15, Kyritsis is not relied upon disclosing the system further comprising a data buffer to store data representation created by the preprocessor, wherein the data buffer is connected to another data buffer over a network, and another data buffer stores transmitted data representation.
Sazonov teaches the system further comprising a data buffer to store data representation created by the preprocessor, wherein the data buffer is connected to another data buffer over a network, and another data buffer stores transmitted data representation ([0069] The system embodied in FIGS. 26 and 28 includes the above described physical components for monitoring food intake into a body and may be connected, via a data network or by hard wire connection, to a central processing unit connected to computer memory, a data buffer (data buffer), and an image buffer (another data buffer)).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Connor (US 2015/0168365 A1) is in the field of a caloric intake measuring system using spectroscopic and 3D imaging analysis (Abstract).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HAMID TARIQ HAFIZ whose telephone number is (571) 272-4629. The examiner can normally be reached 7:30 AM - 5:00 PM, Monday through Thursday.
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, Kang Hu can be reached at 571-270-1344. 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.
/HAMID TARIQ HAFIZ/
Examiner, Art Unit 3715
/ROBERT J UTAMA/Primary Examiner, Art Unit 3715
Prosecution Insights