System and method for artificial intelligence-enabled detection

Abstract

Monitors having housings, gas-specific sensors, environmental sensors, volatile organic compound sensors, indicators, processors, and memory configured to store event data and trained time-series classification models. Methods of monitoring comprising training a time-series classification model, directing environmental air towards sensors, receiving sensor data, determining event activity using the time-series classification model and providing the determination to an indicator.

Description

SYSTEM AND METHOD FOR ARTIFICIAL INTELLIGENCE-ENABLED DETECTION

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 625,253, filed January 25, 2024, the contents of which is hereby incorporated by reference.FIELD OF THE INVENTION

[0002] The present invention relates generally to systems, methods, apparatuses, programmed products and software for artificial intelligence- enabled detection, in particular of fecal matter.BACKGROUND

[0003] Fecal incontinence (Fl) refers to the inability to control one’s own bowel movements, resulting in involuntary loss of stool. Fl impacts roughly 7% of the population, corresponding to 500 million people worldwide. Fl prevalence rises to 22% for adults in their seventh decade; 25-35% for nursing home residents; and 10-25% for hospitalized patients. Fl is one of the primary reasons geriatrics are placed in nursing homes; it is also a leading factor in hospital-acquired infection and extended length of stay. Other populations, such as infants less than two years old, tend to suffer from Fl as well.

[0004] Overall rates of fecal incontinence are on the rise, given that the aging population is growing more rapidly than the general population. This growth will lead to an increased pressure on incontinence care.

[0005] There are high costs associated with managing Fl, including those associated with development of ailments, including skin breakdown,bed sores, ulcers, dermatitis, and tissue damage, particularly where stool is not cleaned from a person within about 15 minutes. Failure to quickly clean can also lead to spread of bacterial infections, such as C. diff. Moreover, failure to quickly clean can lead to poor satisfaction ratings by those who are not cleaned and their families, as well as permanent soiling of bedding materials.

[0006] The current standard of care in connection with detecting fecal output in hospital and nursing home patients is either (1) manual monitoring, which requires significant hours of work by staff that are often not actually performed, resulting in extending soiling, or (2) use of stool diaries, which also often fail to timely detect soiling.

[0007] In addition, so-called “smart diapers” exist where a sensor is physically placed into the diaper and can report soilage. However, the inventor of the present invention has realized that such approach is invasive and, given the cost of many such “smart diapers” for a given subject, wasteful and expensive.

[0008] Accordingly, the inventor of the instant application has realized that there exists a need for new and improved systems, methods, and apparatuses that employ artificial intelligence in order to efficiently and timely detect soiling (that is, “fecal events”), in a manner that overcomes these and other problems.SUMMARY OF INVENTIONAl-enabled fecal detection

[0009] In view of the above, it is the object of the present disclosure to provide improved systems, methods, apparatuses, programmed products and software to overcome, through novel combinations of sensors and artificial intelligence and / or machine learning techniques, the technological challenges faced in conventional approaches for fecal event detection.

[0010] It is a further object of the present disclosure to provide improved systems, methods, apparatuses, programmed products and software that, while improving an associated model in the process of detecting fecal events in a particular environment, makes use of, in accordance with exemplary embodiments of the invention, a fecal monitor including a housing, a plurality of sensors, including a plurality of gas-specific sensors contained within the housing, wherein each gas-specific sensor is configured to detect at least one specific gas associated with at least one of feces and urine, a plurality of environmental sensors contained within the housing, and a plurality of volatile organic compound (VOC) sensors contained within the housing, configured to detect at least one volatile organic compound, an airflow fan operatively connected to the housing and configured to direct environmental air towards the plurality of sensors, an indicator configured to indicate to a user that a fecal event has occurred, a processor contained within the housing, operatively connected to the indicator and to memory configured to store fecal event data and a time-series classification model trained based on training data comprising a time series of the following data tagged with fecal event activity: a plurality of gas-specific sensor data, a plurality of environmental sensor data, and a plurality of volatile organic compoundsensor data, where processor executable instructions are stored in memory, which, when executed by the processor: receive from each of the plurality of sensors, at a respective time period, respective sensor data reflecting the output of each respective sensor, store the respective sensor data in association with the respective time period in memory, determine fecal event activity using the time-series classification model, where the input is the respective sensor data, the outputted result is a determination of fecal event activity, and at least one of a neural network algorithm, an anomaly detection algorithm, an outlier detection algorithm, and a one-class learning algorithm, is employed, upon determination of fecal event activity, provide the determination to the indicator, and a power supply configured to power the airflow fan, the indicator, and the processor.

[0011] It is a further object of the present disclosure to provide improved systems, methods, apparatuses, programmed products and software that, while improving an associated model in the process of detecting fecal events in a particular environment, operates according to a method of fecal monitoring involving training a time-series classification model based on training data made up of a time series of the following data tagged with fecal event activity: a plurality of gas-specific sensor data, a plurality of environmental sensor data, and a plurality of volatile organic compound sensor data, directing, via an airflow fan, environmental air towards a plurality of sensors, receiving from each of a plurality of sensors, at a respective time period, respective sensor data reflecting the output of each respective sensor, storing the respective sensor data in association with the respective time period in memory, determining fecal event activity using the time-series classification model, where the input is the respective sensor data, theoutputted result is a determination of fecal event activity, and at least one of a neural network algorithm, an anomaly detection algorithm, an outlier detection algorithm, a one-class learning algorithm, and a supervised learning algorithm, is employed, and, upon determination of fecal event activity, providing the determination to an indicator.Al-enabled detection

[0012] The improved systems, methods, apparatuses, programmed products and software that overcome the technological challenges faced in conventional approaches for fecal event detection may find use in the detection of not only fecal events, but any event (which may be an object) associated with emission and / or presence of gasses, such as, without limitation, detection of molding fruit, gas leaks, or drugs.

[0013] Accordingly, it is a further object of the present disclosure to provide improved systems, methods, apparatuses, programmed products and software that, while improving an associated model in the process of detecting events associated with emission and / or presence of gasses in a particular environment, makes use of, in accordance with exemplary embodiments of the invention, a monitor including a housing, a plurality of sensors, including a plurality of gas-specific sensors contained within the housing, wherein each gas-specific sensor is configured to detect at least one specific gas, a plurality of environmental sensors contained within the housing, and a plurality of volatile organic compound (VOC) sensors contained within the housing, configured to detect at least one volatile organic compound, an airflow fan operatively connected to the housing and configured to direct environmental air towards the plurality of sensors, an indicator configured to indicate to a user that an event has occurred, aprocessor contained within the housing, operatively connected to the indicator and to memory configured to store event data and a time-series classification model trained based on training data comprising a time series of the following data tagged with event activity: a plurality of gas-specific sensor data, a plurality of environmental sensor data, and a plurality of volatile organic compound sensor data, where processor executable instructions are stored in memory, which, when executed by the processor: receive from each of the plurality of sensors, at a respective time period, respective sensor data reflecting the output of each respective sensor, store the respective sensor data in association with the respective time period in memory, determine event activity using the time-series classification model, where the input is the respective sensor data, the outputted result is a determination of event activity, and at least one of a neural network algorithm, an anomaly detection algorithm, an outlier detection algorithm, and a one-class learning algorithm, is employed, upon determination of event activity, provide the determination to the indicator, and a power supply configured to power the airflow fan, the indicator, and the processor.

[0014] It is a further object of the present disclosure to provide improved systems, methods, apparatuses, programmed products and software that, while improving an associated model in the process of detecting specified events associated with emission and / or presence of gasses in a particular environment, operates according to a method of monitoring involving training a time-series classification model based on training data made up of a time series of the following data tagged with event activity: a plurality of gasspecific sensor data, a plurality of environmental sensor data, and a plurality of volatile organic compound sensor data, directing, via an airflow fan,environmental air towards a plurality of sensors, receiving from each of a plurality of sensors, at a respective time period, respective sensor data reflecting the output of each respective sensor, storing the respective sensor data in association with the respective time period in memory, determining event activity using the time-series classification model, where the input is the respective sensor data, the outputted result is a determination of event activity, and at least one of a neural network algorithm, an anomaly detection algorithm, an outlier detection algorithm, a one-class learning algorithm, and a supervised learning algorithm, is employed, and, upon determination of event activity, providing the determination to an indicator.BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Fig. 1 shows an exemplary design of an artificial-intelligence- enabled system configured to monitor for fecal events in an environment, according to embodiments of the invention;

[0016] Fig. 2 shows a flowchart of an exemplary method for applying a model for detection of fecal events, and for updating the model based on feedback data along with calibration data and training data, according to embodiments of the invention.DETAILED DESCRIPTION OF THE EXEMPLARY EMBODIMENTSAl-enabled fecal detection

[0017] According to embodiments of the present invention, a device may be provided and used for detection of fecal events, comprising gas sensors and on onboard or offboard processor or processors employing machine learning or artificial intelligence to detect the presence of nearby feces. According to embodiments, the device may be small, may bereusable, for example across multiple fecal observation subjects such as nursing home or hospital patients. In embodiments, the device may be noninvasive, for example, placed in proximity to, but not touching, the subject. In embodiments, it may be placed a distance, which may be a predetermined or user-selectable distance, from the subject. By way of example, it may be placed on a stand next to a hospital bed, intravenous (“IV”) infusion pole, or bedside table, and may be a known or approximately known distance from the patient.

[0018] In embodiments, the device may, upon a determination of a fecal event, trigger an alert. In embodiments, this alert may be a local alert, such as an alarm or a light such as a flashing light. In embodiments, this alert may alternately or additionally be transmitted to a system or device associated with a person, such as a caregiver, such that the alert may be delivered remotely. In embodiments, the alert may relate solely to an indication of the presence of feces. In additional embodiments, the alert may relate to an indication of the presence of other substances such as urine. In yet additional embodiments, the alert may present information regarding detection or suspicion of a physiological decision, such as may be detected from the sampling performed by the gas sensors. Such physiological conditions may include one or more of a C. diff infection, a metabolic dysfunction, gastrointestinal and colon issues, and neurological conditions.

[0019] In embodiments, the sensors, including the gas sensors, of the device, may take measurements of gases in the air. These measurements may be taken continuously, or at preset or variable intervals. The measurements may be used to train one or more classification models, which, in preferred embodiments, may be or include a time-seriesclassification model. In embodiments, the classification model or time-series classification model may be a neural network model, an anomaly detection algorithm, an outlier detection algorithm, a one-class learning algorithm, or a supervised learning algorithm.

[0020] In an embodiment, a measurement window of the sensors may be used as an input. In embodiments, the measurement window may be a small measurement window. In further embodiments, an output may be generated from the input. In yet further embodiments, the output may be in the form of a binary output. The output may be indicative of whether or not fecal output is detected in the input. In further embodiments, the output, for example selectively upon determination that the output is a positive signal or positive binary signal, may be fed into a secondary model or secondary classifier. The secondary model or secondary classifier may be a multi-class classifier. The multi-class classifier may be used to ascertain a further feature of the fecal output. The further feature of the fecal output may pertain to a healthy or diseased condition of the output. In embodiments, the further feature may be a health score pertaining to the fecal output. The health score may be on a scale from healthy to dangerous. The health score may pertain to a particular disease condition, for example, pertaining to a probability or existence of the disease condition.

[0021] The disease condition may be a C. diff infection. In embodiments, a C. diff infection may be detected in accordance with detection of one or both of associated furans, such as 2-furancarboxaldehyde, and 3- Methylindole. The disease condition may be a rotavirus and associated Campylobacter stool. The disease condition may be inflammatory bowel disease (I BD), which may be detected based on the presence of ester. Inembodiments, a C. diff infection may be detected in accordance with detection of one or both of associated furans, such as 2- furancarboxaldehyde, and 3-Methylindole. The disease condition may be colon cancer.

[0022] According to one embodiment, training data may be used to train a one-class learning algorithm, which algorithm may function as a one-class classifier. The one-class learning algorithm may be a one-class neural network, or a support vector machine (SVM). In embodiments, the one-class learning algorithm may alternately or additionally incorporate one or more or all of distance-based methods, anomaly detection algorithms, and outlier detection algorithms. In embodiments, exemplars from only one class may be provided. For example, the exemplars may be gaseous or other measurements pertaining to feces in different contexts. In alternate embodiments, exemplars for two classes may be provided (e.g. discriminating feces from other background items such as urine, passing of gas, or food). In yet another embodiment, exemplars for additional classes may be provided (e.g., cleaning solutions, garbage cans) so algorithms can distinguish these confounding classes from true fecal events.

[0023] In embodiments, the training data may be represented as a matrix. The matrix may indicate sensor measurements over time, for example by providing sensor measurements as columns and time as rows, or vice versa. Assuming the former, the matrix may include an additional column indicating whether a datapoint represents a urine or a fecal event. Further, if so, such column may further indicate additional items of interest, such as the presence of an adverse physical condition (e.g. a C. diff infection) or metabolic condition. In embodiments, training data may becollected at various distances from the source, under various background conditions and background smells, or from individuals of various genders, races, ethnicities, dietary habits, or lifestyles. Such items of relevance may be further specified in the training data or column.

[0024] In an exemplary embodiment, sample training data may be according to the following tables 1 A-1 F:Table 1ATable 1BTable 1CUrine Event FeedbackTable 1DTable 1ETable 1F

[0025] In embodiments, the one-class learning algorithm may transform the data to lie in a geometric space, which may be a high-dimensional space. The one-class learning algorithm may then proceed to calculate an envelope hypersphere, which may be a compact envelope or compact hypersphere, surrounding examples (e.g. exemplars within the fecal class) in the geometric space. In further embodiments, the one-class learning algorithm may, for further gas samplings, compute its associated location in the geometric space with respect to the envelope or hypersphere. Adetermination may be made by the learning algorithm, based on whether the associated location is within the envelope or hypersphere, as to whether a fecal event has occurred in connection with the further gas samplings. A probabilistic determination may also be made based on the distance between the associated location and the center of the envelope or hypersphere. In embodiments, a sensitivity threshold may be customized in order to balance between false positive and false negative rates, for example, with adjustments being made based on user feedback as to whether a fecal event has in fact occurred. In embodiments, the user feedback may adjust the envelope or hypersphere, so as to capture fecal events with sensitivity depending on user selected preferences.

[0026] In embodiments, an adaptive threshold system is employed in the algorithm. For example, in applications where the target smell may emerge slowly and immediate triggers, within seconds to minutes, are not required, then algorithm may employ higher thresholds to boost accuracy over speed. In other applications where immediate triggers are required, the algorithm may optimize for a user-defined trade-off between speed and accuracy.

[0027] In embodiments, to adjust the learning algorithm to a particular location or context, a user interface may be provided, for example with one or more buttons, allowing a user to provide an indication of one or two or more of a false positive, a true positive and a false negative as to whether a fecal event has occurred. Such feedback may be used to refine classification boundaries.

[0028] In further embodiments, adjustments may be made to account for background gas levels, stated differently, background “noise.”

[0029] The inventors of the present invention have noticed that environments may vary, inter alia, in temperature, humidity, and pressure, and such variability can create difficulty in consistent and accurate detection of target odorants, such as those associated with a fecal event. In order to address this problem, in embodiments, the device may, as a part of calibration phase, sample environmental factors and calculate and develop statistics based thereon, and may also sample neutral background odors present in the environment. Further, as part of the calibration phase, model parameters may be adjusted. Such sampling may be used to set a baseline level of presence of a particular target odorant in order to result in a determination that a fecal event has occurred. In embodiments, such calibration may occur once a day, for example at a pre-defined time, or at another interval such as twice daily, four times daily, hourly, every two days, or weekly. Advantageously, such calibration may account for changing conditions and sensor drift over time.

[0030] In embodiments, as a further way to account for background gas levels, background subtraction may be employed. The inventors of the present invention have noticed that robust detection of fecal smells requires that background odors, which may, for example, take the form of presence of foods, an open diaper pail, or the presence of cleaning solutions, be subtracted from sampling measurements, so that they do not obfuscate target foreground odors and interfere with the reliability of fecal event detection. Accordingly, a background subtraction algorithm may be employed. Such background subtraction algorithm may employ one or more or all of temporal average filters, Gaussian mixture models, and a neuralnetwork trained to learn neutral background odors that are consistently present, so that their contributions may be subtracted from the samplings.

[0031] One or both of short-term habituation algorithms and long-term habituation algorithms may be employed. The short-term habituation algorithm may involve habituation occurring on the order of five seconds, ten seconds, twenty seconds, thirty seconds, one minute, two minutes, fives minutes, ten minutes, fifteen minutes, or 30 minutes, or a time value within the range of ten to 120 seconds. The long-term habituation algorithm may involve habituation occurring on the order of one hour, two hours, four hours, eight hours, twelve hours, one day, two days, four days, or a week, or a time value chosen from the range of twelve to twenty-four hours. Advantageously, such background subtraction enables otherwise subtle changes in the environment to become more readily detectable.

[0032] In embodiments, as a further way to account for background gas levels, sensor normalization may be employed. The inventors of the present invention have realized that the signal with respect to target gaseous odors being sensed can very over ten orders of magnitude in concentration, while sensor responses are often limited to only one or two orders of magnitude. In embodiments of the present invention, for example to address such issue, sensor data may be normalized. In embodiments, such normalization may be carried out by a normalization algorithm, for example divisive normalization, batch normalization, z-score, min-max normalization, or unit vector normalization. Advantageously, normalization may be employed to prevent sensor saturation and enable dynamic range adjustment, better representing the environment with respect to detectability of target gaseous odors.

[0033] In embodiments, as a further way to account for background gas levels, transient gaseous odors may be detected. For example, in embodiments, such transient gaseous odors may be representative of passed gas as opposed to fecal events requiring changing of sheets or other intervention of a caretaker. Thus, in embodiments, when a measurement may be indicative of a potential fecal output, the learning algorithm may monitor the duration of positive signals. In embodiments, a decay in signal strength, for example within a period of less than 15 seconds, or 20 seconds, or 30 seconds, or 1 minute, or an occurrence of positive signals in intermittent bursts, the learning algorithm may avoid triggering an alert despite the intermittent signal strength.

[0034] In an embodiment, the device may be configured to avoid triggering based on fecal events or gaseous odorants that are located more than a preferred radius of detectability away from the device. In a further embodiment, the radius of detectability may be selected by a user, for example through a user interface. For example, a user may select 3 feet or 6 feet, often associated with a device next to or in proximity to a hospital bed, 12 feet, or 30 feet, or a particular inputted preferred distance, based for example on distance from the subject. Advantageously, this may provide an initial setting for the sensitivity of the device, or of the initial size of the envelope or hypersphere within the geometric space.

[0035] In an embodiment, the device may be configured to provide an early warning with respect to a fecal event, before the fecal matter has exited the subject, advantageously allowing for earlier triggering and faster staff responses. In further embodiments, after fecal output is successfully detected, the learning algorithm may retrospectively analyze sensormeasurements prior to the detection to determine patterns in gaseous readings associated with ultimately determined fecal events. In further embodiments, both the patterns and the time prior to standard detection at which the pattern becomes evident may be determined and used to trigger fecal event alerts as to future occurrences. In embodiments, the learning algorithm may store such gaseous measurements are used in this process within memory of the device or remotely. In further embodiments, a model may be specifically trained for a particular device or for a particular patient.

[0036] In an embodiment, the device may have an energy efficiency mode. Advantageously, this conserves battery life and power consumption, and improves the longevity of the device due to loss of sensitivity and fidelity of sensor components due to excess “sniffing.” In embodiments, such energy efficiency mode may be achieved through dynamic adjustment of the rate of sensor readings, that is, dynamic adjustment of “sniff” frequency. In embodiments, a default mode may be the taking of sparse measurement at a lower frequency (for example, once per fifteen seconds, once per thirty seconds, once per minute, or once per two minutes), and then, upon detection of gaseous measurements indicative of a potential fecal event, sniff frequency may increase (for example, once per five seconds, two seconds, one second, or two, three or five times per second). The learning algorithm may be configured to only configure an alert based on readings at the increased sniff frequency, thereby allowing additional confidence as to a fecal event occurring. The sniff frequency may return, upon a period of relatively inactivity, for example a period of five minutes, ten minutes, or fifteen minutes, to its baseline lower level.

[0037] In an embodiment, the device may generate reports. In certain embodiments, the report may be generated on a fixed schedule, for example, daily, weekly, biweekly, or monthly. The report may indicate, for a particular subject in highest proximity to the device, one or more or all of frequency, timing, and quality of fecal events (e.g. with respect to physical condition). Advantageously, this report may be reviewed by treating physicians and carekeepers.

[0038] Fig. 1 shows an exemplary design of an artificial-intelligence- enabled system configured to monitor for fecal events in an environment, according to embodiments of the invention. As can be seen, the device may comprise gas-specific sensors 1, for example specific to gases associated with fecal matter and with disease conditions, or for example metal oxide (MOX) sensors, electrochemical sensors, or combustible gas sensors, environmental sensors 2, for example pertaining to one or more or all of temperature, pressure, and humidity, and VOC sensors 3, which in embodiments may be generic or commercially available VOC sensors. The device may further comprise a power supply 4, which may be, for example, a battery or a plug-in power supply, or both, an airflow fan 5 which may be directed to direct air towards some or all of the sensors 1 , 2, 3, a cooling fan or muffin fan 6, directed to avoid overheating of the device and of its processor 7, which may be for example an Al chip, memory storage 8, which may store, for example, training data and the models used, cellular / wifi circuitry 9, or which may be, for example, wireless, Bluetooth, or ethernet circuitry or equipment, or alarm, which may be, for example, a screen, flashing lights, or sound, which may be used as an indicator to transmit or locally indicate, respectively, an alert as to fecal activity and data to acaretaker or a centralized repository, an interface which may comprise one or both of an LCD panel 10 for displaying settings and information pertaining to settings for the device and the patient and a feedback button or buttons11 which may be used by a user to provide feedback as to a false positive (no fecal event occurred) or a true positive (fecal event occurred), or a false negative (a fecal event occurred in the absence of a notification) and an enclosure 12 which may be waterproof. In a further embodiment, enclosure12 may be coated with a waterproof coating, for example a teflon coating.

[0039] Fig. 2 shows a flowchart of an exemplary method for applying a model for detection of fecal events, and for updating the model based on feedback data along with calibration data and training data, according to embodiments of the invention.

[0040] As can be seen, sensor measurements may be obtained, for example via sensors 1 , 2, 3, and the input may be subjected to data preprocessing, for example background subtraction and normalization such as is discussed herein. A model, such as a simple model, a one-class classification model, or a supervised classification model, may be used to determine whether there is a potential indication of a fecal event. Data pertaining to selection or adjustment or use of a radius of sensitivity may be used by such model. Alternately or additionally, a sensitivity threshold may be employed by such model.

[0041] Upon determination of the presumed positive signal, the energy efficient mode may be deactivated, and the sampling frequency of the sensors may temporarily increase. Using the higher frequency input data from the more frequent measurements, a determination may be made as to whether the signal is persistent or merely transient.

[0042] In the event it is determined to be persistent, the associated data may be stored, which may be used to train, for example according to a secondary model, an early activation or early detection mode useful to more quickly determine fecal events for future occurrences.

[0043] Further, upon determination of persistence, a further model may be employed, for example, to more accurately determine the presence of a fecal event in the presence of higher frequency data, to determine the presence of a metabolic condition, physiological condition or health score, or both. Based on this analysis, a notification may be triggered, as to the fecal event, and, where determinable, a health score or the determination of the metabolic or physiological condition.

[0044] Upon the trigger of the notification, the user may provide feedback, for example via a user interface, as to whether the notification represented a true positive (an actual fecal event) or a false positive (no fecal event). The user may, in embodiments, provide feedback, for example via the user interface, as to whether there has been a false negative (an actual fecal event despite the absence of the trigger of a notification). Such feedback may be used to update the model used for the detection of fecal events and triggering of notifications. For example, it may be used to determine sensitivities to be associated with particular calibration settings. In addition, training data may used in conjunction with, or prior to, the new feedback information, in order to create the initial version of such fecal detection model.

[0045] In embodiments, the device and methods as discussed herein may be used in connection with additional applications. For example, they may be used not just for geriatrics in hospitals, nursing homes, and at home,but they may be used, for example with their own training data and associated models, for babies, for example at home or in neonatal intensive care units. They may be used in clinical departments such as burn and wound care units. They may be used with patients who have particular injuries, such as rectal, colon or spinal cord injuries or who have damaged sphincters. They may be used for patients with dementia. The may be used in tele-health and at-home care settings.

[0046] They may be used in animal care facilities. In certain embodiments, the device may be trained to provide indications relating to the animal. The indications may include one or more or all of history, pedigree, hormone levels, and dietary patterns.

[0047] They may be used in public spaces, such as restaurants, bathrooms, and airports.

[0048] In certain embodiments, in addition to or instead of triggering an alert, an automated action may be performed. In some embodiments, the automated action may be the release of air freshener from a communicatively coupled device or from the device itself. In some embodiments, the automated action may be the commencement of cleaning by a robot cleaner, such as a robot vacuum, or a litter box. In embodiments, the robot cleaner or litter box may be communicatively coupled with the device, or may be integrated with the device.

[0049] In embodiments, the device may be integrated with or communicatively coupled with a monitoring camera, such as a hospital camera directed to the patient or a baby monitor camera. Such camera may turn on, or have its feed provided to a caretaker, upon, the determination that a fecal event has occurred.

[0050] There is therefore provided, according to embodiments of the invention, a fecal monitor including a housing, a plurality of sensors, including a plurality of gas-specific sensors contained within the housing, wherein each gas-specific sensor is configured to detect at least one specific gas associated with at least one of feces and urine, a plurality of environmental sensors contained within the housing, and a plurality of volatile organic compound (VOC) sensors contained within the housing, configured to detect at least one volatile organic compound, an airflow fan operatively connected to the housing and configured to direct environmental air towards the plurality of sensors, an indicator configured to indicate to a user that a fecal event has occurred, a processor contained within the housing, operatively connected to the indicator and to memory configured to store fecal event data and a time-series classification model trained based on training data comprising a time series of the following data tagged with fecal event activity: a plurality of gas-specific sensor data, a plurality of environmental sensor data, and a plurality of volatile organic compound sensor data, where processor executable instructions are stored in memory, which, when executed by the processor: receive from each of the plurality of sensors, at a respective time period, respective sensor data reflecting the output of each respective sensor, store the respective sensor data in association with the respective time period in memory, determine fecal event activity using the time-series classification model, where the input is the respective sensor data, the outputted result is a determination of fecal event activity, and at least one of a neural network algorithm, an anomaly detection algorithm, an outlier detection algorithm, and a one-class learning algorithm, is employed, upon determination of fecal event activity, provide the determination to theindicator, and a power supply configured to power the airflow fan, the indicator, and the processor.

[0051] According to further embodiments, the environmental sensors include a humidity sensor to detect humidity, a temperature sensor to detect temperature and a pressure sensor to detect pressure.

[0052] According to yet further embodiments, the neural network algorithm is employed.

[0053] According to other embodiments, the anomaly detection algorithm is employed.

[0054] According to yet other embodiments, the outlier detection algorithm is employed.

[0055] According to embodiments, the one-class learning algorithm is employed.

[0056] According to yet additional embodiments, several of the neural network algorithm, the anomaly detection algorithm, the outlier detection algorithm, and the one-class learning algorithm are employed.

[0057] In embodiments, gas-specific sensors and the respective sensor data measure hydrogen sulfide, indole, skatole, mercaptans, methane, methyl sulfide, 3-Methylindole, furans, ethyl dodecanoate, phenols, butanol, ester, hydrocarbons and ammonia.

[0058] In embodiments, the fecal monitor may have a user interface outside of the housing configured to receive from the user a selectable indication of whether or not a fecal event has occurred, with the time series of data further including a distance from a measurement device, and the processor may be further configured to receive from the user interface aselectable indication of whether or not the fecal event activity has occurred, update the sensitivity of the time-series classification model with respect to the distance based on whether the selectable indication indicates a true positive or a false positive with respect to the fecal event activity, and store within memory, as part of the fecal event data, at least one of the determination and the selectable indication. The processor may be further configured to receive from the user interface a selectable indication that a false negative as to fecal activity has occurred, and to update the sensitivity of the time-series classification model with respect to the distance based on the selectable indication that the false negative has occurred. Alternately or in addition, the processor may be further configured to receive from the user interface a selectable indication that a true negative as to fecal activity has occurred, and to update the sensitivity of the time-series classification model with respect to the distance based on the selectable indication that the true negative has occurred. In embodiments, the user interface may be on the exterior of the housing, or separated from the housing.

[0059] In embodiments, the processor may be further configured to classify at least one background odor based on the output, and the determination of whether a fecal event is likely to have occurred may be further based on the classification of the at least one background odor.

[0060] In yet further embodiments, the power supply may comprise an external power source and a battery configured to power the fecal monitor upon disconnection of the external power source.

[0061] In further embodiments, the power supply may include an external power source, or may include a battery.

[0062] In embodiments, the processor may include, a graphics processing unit (GPU) chip, which may be an ultra-low-power graphics processing unit (GPU) chip.

[0063] In embodiments, the processor may include a central processing unit (CPU), or an ultra-low-power central processing unit (CPU).

[0064] In yet further embodiments, the indicator may be located on the exterior of the housing and may indicate that the fecal event has occurred by a local sound or visual display.

[0065] In further embodiments, the indicator may include a transmitter configured to indicate that the fecal event has occurred by transmitting an alert to a separate device in communication with the user. The transmitter may be further configured to transmit the fecal event data to a centralized repository, and may be further configured to receive updated firmware for storage in memory and execution by at least one processor.

[0066] In embodiments, the housing may be a waterproof enclosure.

[0067] In additional embodiments, the training data may be further tagged with a metabolic or physiological condition. The metabolic or physiological condition may include at least one of: normal stool, diarrhea, C. diff infection, rotavirus infection, Campylobacter stool, bacterial infection, fungal infection, metabolic state abnormality, indication of colon cancer, inflammatory bowel disease, catheter-associated urinary tract infection (CAUTI), hard stool, and soft stool. The processor may be further configured to, upon the determination of fecal event activity, determine a physiological condition based on the training data, which determination of the physiologicalcondition by the processor may be performed using a secondary multi-class classifier stored in memory.

[0068] In embodiments, the processor may be further configured to represent, within the time-series classification model, the distance from the measurement device as data laying within a high-dimensional space, determine a compact envelope, which envelope in embodiments may be a hypersphere, representative of a sensitivity threshold, surrounding a portion of the data laying within the high-dimensional space, and modify the boundaries of the compact envelope based on the selectable indication of whether or not the fecal event has occurred. The at least one of the determination of the compact envelope and the modification of the boundaries of the compact envelope may be performed based on a second one-class learning algorithm in which the class members are exemplars of fecal events. In embodiments, the at least one of the determination of the compact envelope and the modification of the boundaries of the compact envelope may be performed based on calibration information, which calibration information may comprise a user-selected radius of detectability.

[0069] In embodiments, a neural network is employed and is configured to learn, from the output, consistently present neutral background odors, and to subtract a contribution of the learned consistently present neutral background odors from the output. The learning of the consistently present neutral background may be is performed using a short-term habituation algorithm, which short-term habituation algorithm may habituate according to a time period of 10 seconds to 120 seconds.

[0070] In further embodiments, the learning of the consistently present neutral background odors may be performed using a habituation algorithm habituating according to a time period of 12 hours to 24 hours.

[0071] In embodiments, the processor is configured to normalize the output, which normalization may be performed by divisive normalization, batch normalization, min-max normalization, z-score normalization, or unit vector normalization.

[0072] In embodiments, the processor may be further configured to determine, based on the fecal event data, commonalities in output from time periods prior to determinations that the fecal event is likely to have occurred that have been confirmed as true positives based on the selectable indication, and based on the determined commonalities, adjust a time boundary at which the determining the fecal event activity occurs. The determination of the commonalities and the adjustment of the time boundary may be performed by the processor based on a secondary model stored in memory.

[0073] In embodiments, the fecal monitor may be further configured to dynamically adjust a frequency at which measurements are taken by one or more of the plurality of gas-specific sensors, the plurality of environmental sensors, and the plurality of VOC sensors. The dynamic adjustment may involve taking sparse measurements, increasing the frequency of measurement upon detection of output potentially indicative of a fecal event, and then returning to taking sparse measurements upon detection of output unlikely to be indicative of a fecal event.

[0074] In embodiments, the fecal event activity may include production of feces, production of urine, or production of feces and urine.

[0075] In further embodiments, the fecal event activity may comprise an indication that a fecal event occurred, or an indication that the fecal event has not occurred.

[0076] In yet further embodiments, the fecal event activity may include an indication associated with the likelihood that a fecal event occurred.

[0077] In embodiments, the plurality of gas-specific sensors may include an ammonia sensor which is configured to detect the presence and amount of ammonia in a gas sample.

[0078] In further embodiments, the plurality of gas-specific sensors may include a hydrogen sulfide sensor which is configured to detect the presence and amount of hydrogen sulfide in a gas sample, an indole sensor which is configured to detect the presence and amount of indole in a gas sample, a skatole sensor which is configured to detect the presence and amount of skatole in a gas sample, a mercaptans sensor which is configured to detect the presence and amount of mercaptans in a gas sample, a methane sensor which is configured to detect the presence and amount of methane in a gas sample, or a methyl sulfide sensor which is configured to detect the presence and amount of methyl sulfide in a gas sample.

[0079] In yet further embodiments, the fecal monitor may further include a cooling fan or muffin fan configured to cool the processor.

[0080] In embodiments, the processor may include a plurality of processors.

[0081] In further embodiments, the indicator may include an LCD panel.

[0082] In yet additional embodiments, there is a method of fecal monitoring involving training a time-series classification model based ontraining data made up of a time series of the following data tagged with fecal event activity: a plurality of gas-specific sensor data, a plurality of environmental sensor data, and a plurality of volatile organic compound sensor data, directing, via an airflow fan, environmental air towards a plurality of sensors, receiving from each of a plurality of sensors, at a respective time period, respective sensor data reflecting the output of each respective sensor, storing the respective sensor data in association with the respective time period in memory, determining fecal event activity using the time-series classification model, where the input is the respective sensor data, the outputted result is a determination of fecal event activity, and at least one of a neural network algorithm, an anomaly detection algorithm, an outlier detection algorithm, a one-class learning algorithm, and a supervised learning algorithm, is employed, and, upon determination of fecal event activity, providing the determination to an indicator.

[0083] The method may further involve receiving from a user a selectable indication of whether or not the fecal event activity has occurred, updating the sensitivity of the time-series classification model with respect to a distance based on whether the selectable indication indicates a true positive or a false positive with respect to the fecal event activity, and storing within memory at least one of the determination and the selectable indication.Al-enabled detection

[0084] According to embodiments of the present invention, a device may be provided and used for detection of events associated with emission and / or presence of gasses, comprising gas sensors and on onboard or offboard processor or processors employing machine learning or artificial intelligence to detect events (which may be objects such as, withoutHrrdtation, moldy fruit or drugs) associated with emission and / or presence of gasses. According to embodiments, the device may be small, may be reusable, for example across multiple observation subjects. In embodiments, the device may be noninvasive, for example, placed in proximity to, but not touching, the subject. In embodiments, it may be placed a distance, which may be a predetermined or user-selectable distance, from the subject.

[0085] In embodiments, the device may, upon a determination of an event associated with emission and / or presence of gasses, trigger an alert. In embodiments, this alert may be a local alert, such as an alarm or a light such as a flashing light. In embodiments, this alert may alternately or additionally be transmitted to a system or device associated with a person, such that the alert may be delivered remotely. In embodiments, the alert may relate to an indication of the presence of particular substances. In yet additional embodiments, the alert may present information regarding detection or suspicion of an event associated with emission and / or presence of gasses, such as may be detected from the sampling performed by the gas sensors.

[0086] In embodiments, the sensors, including the gas sensors, of the device, may take measurements of gases in the air. These measurements may be taken continuously, or at preset or variable intervals. The measurements may be used to train one or more classification models, which, in preferred embodiments, may be or include a time-series classification model. In embodiments, the classification model or time-series classification model may be a neural network model, an anomaly detection algorithm, an outlier detection algorithm, a one-class learning algorithm, or a supervised learning algorithm.

[0087] In an embodiment, a measurement window of the sensors may be used as an input. In embodiments, the measurement window may be a small measurement window. In further embodiments, an output may be generated from the input. In yet further embodiments, the output may be in the form of a binary output. The output may be indicative of whether or not a particular output is detected in the input. In further embodiments, the output, for example selectively upon determination that the output is a positive signal or positive binary signal, may be fed into a secondary model or secondary classifier. The secondary model or secondary classifier may be a multi-class classifier. The multi-class classifier may be used to ascertain a further feature of the output. The further feature of the output may pertain to classification of the output. In embodiments, the further feature may be a score pertaining to the output which may be on a scale, such as from safe to dangerous.

[0088] According to one embodiment, training data may be used to train a one-class learning algorithm, which algorithm may function as a one-class classifier. The one-class learning algorithm may be a one-class neural network, or a support vector machine (SVM). In embodiments, the one-class learning algorithm may alternately or additionally incorporate one or more or all of distance-based methods, anomaly detection algorithms, and outlier detection algorithms. In embodiments, exemplars from only one class may be provided. For example, the exemplars may be gaseous or other measurements pertaining to events associated with emission and / or presence of gasses in different contexts. In alternate embodiments, exemplars for two classes may be provided (e.g. discriminating particular events associated with emission and / or presence of gasses from otherbackground items). In yet another embodiment, exemplars for additional classes may be provided (e.g., cleaning solutions, garbage cans) so algorithms can distinguish these confounding classes from true events intended to be detected.

[0089] In embodiments, the training data may be represented as a matrix. The matrix may indicate sensor measurements over time, for example by providing sensor measurements as columns and time as rows, or vice versa. Assuming the former, the matrix may include an additional column indicating whether a datapoint represents a specified event associated with emission and / or presence of gasses. Further, if so, such column may further indicate additional items of interest. In embodiments, training data may be collected at various distances from the source, under various background conditions and background smells, or from individuals of various genders, races, ethnicities, dietary habits, or lifestyles. Such items of relevance may be further specified in the training data or column.

[0090] In an exemplary embodiment, sample training data may be according to the following tables 2A-2F:Table 2ATable 2BTable 2CTable 2DTable 2ETable 2F

[0091] In embodiments, the one-class learning algorithm may transform the data to lie in a geometric space, which may be a high-dimensional space. The one-class learning algorithm may then proceed to calculate an envelope hypersphere, which may be a compact envelope or compact hypersphere, surrounding examples (e.g. exemplars within the class) in the geometric space. In further embodiments, the one-class learning algorithm may, for further gas samplings, compute its associated location in the geometric space with respect to the envelope or hypersphere. A determination may be made by the learning algorithm, based on whether the associated location is within the envelope or hypersphere, as to whether an event has occurred in connection with the further gas samplings. A probabilistic determination may also be made based on the distance between the associated location and the center of the envelope or hypersphere. In embodiments, a sensitivity threshold may be customized in order to balance between false positive and false negative rates, for example, with adjustments being made based on user feedback as to whether an event has in fact occurred. In embodiments, the user feedback may adjust the envelope or hypersphere, so as to capture events with sensitivity depending on user selected preferences.

[0092] In embodiments, an adaptive threshold system is employed in the algorithm. For example, in applications where the target smell may emerge slowly and immediate triggers, within seconds to minutes, are not required (e.g. detection of moldy fruit), then algorithm may employ higherthresholds to boost accuracy over speed. In other applications where immediate triggers are required (e.g. detection of drugs or gas leaks), the algorithm may optimize for a user-defined trade-off between speed and accuracy.

[0093] In embodiments, to adjust the learning algorithm to a particular location or context, a user interface may be provided, for example with one or more buttons, allowing a user to provide an indication of one or two or more of a false positive, a true positive and a false negative as to whether a particular event has occurred. Such feedback may be used to refine classification boundaries.

[0094] In further embodiments, adjustments may be made to account for background gas levels, stated differently, background “noise.”

[0095] The inventors of the present invention have noticed that environments may vary, inter alia, in temperature, humidity, and pressure, and such variability can create difficulty in consistent and accurate detection of target odorants, such as those associated with a particular event associated with emission and / or presence of gasses. In order to address this problem, in embodiments, the device may, as a part of calibration phase, sample environmental factors and calculate and develop statistics based thereon, and may also sample neutral background odors present in the environment. Further, as part of the calibration phase, model parameters may be adjusted. Such sampling may be used to set a baseline level of presence of a particular target odorant in order to result in a determination that a particular event associated with emission and / or presence of gasses has occurred. In embodiments, such calibration may occur once a day, for example at a pre-defined time, or at another interval such as twice daily, fourtimes daily, hourly, every two days, or weekly. Advantageously, such calibration may account for changing conditions and sensor drift over time.

[0096] In embodiments, as a further way to account for background gas levels, background subtraction may be employed. The inventors of the present invention have noticed that robust detection of smells requires that background odors, which may be subtracted from sampling measurements, so that they do not obfuscate target foreground odors and interfere with the reliability of event detection. Accordingly, a background subtraction algorithm may be employed. Such background subtraction algorithm may employ one or more or all of temporal average filters, Gaussian mixture models, and a neural network trained to learn neutral background odors that are consistently present, so that their contributions may be subtracted from the samplings.

[0097] One or both of short-term habituation algorithms and long-term habituation algorithms may be employed. The short-term habituation algorithm may involve habituation occurring on the order of five seconds, ten seconds, twenty seconds, thirty seconds, one minute, two minutes, fives minutes, ten minutes, fifteen minutes, or 30 minutes, or a time value within the range of ten to 120 seconds. The long-term habituation algorithm may involve habituation occurring on the order of one hour, two hours, four hours, eight hours, twelve hours, one day, two days, four days, or a week, or a time value chosen from the range of twelve to twenty-four hours. Advantageously, such background subtraction enables otherwise subtle changes in the environment to become more readily detectable.

[0098] In embodiments, as a further way to account for background gas levels, sensor normalization may be employed. The inventors of the presentinvention have realized that the signal with respect to target gaseous odors being sensed can very over ten orders of magnitude in concentration, while sensor responses are often limited to only one or two orders of magnitude. In embodiments of the present invention, for example to address such issue, sensor data may be normalized. In embodiments, such normalization may be carried out by a normalization algorithm, for example divisive normalization, batch normalization, z-score, min-max normalization, or unit vector normalization. Advantageously, normalization may be employed to prevent sensor saturation and enable dynamic range adjustment, better representing the environment with respect to detectability of target gaseous odors.

[0099] In embodiments, as a further way to account for background gas levels, transient gaseous odors may be detected. Thus, in embodiments, when a measurement may be indicative of a potential specified event associated with emission and / or presence of gasses, the learning algorithm may monitor the duration of positive signals. In embodiments, a decay in signal strength, for example within a period of less than 15 seconds, or 20 seconds, or 30 seconds, or 1 minute, or an occurrence of positive signals in intermittent bursts, the learning algorithm may avoid triggering an alert despite the intermittent signal strength.

[0100] In an embodiment, the device may be configured to avoid triggering based on events or gaseous odorants that are located more than a preferred radius of detectability away from the device. In a further embodiment, the radius of detectability may be selected by a user, for example through a user interface. For example, a user may select 3 feet or 6 feet, 12 feet, or 30 feet, or a particular inputted preferred distance, basedfor example on distance from the subject. Advantageously, this may provide an initial setting for the sensitivity of the device, or of the initial size of the envelope or hypersphere within the geometric space.

[0101] In an embodiment, the device may be configured to provide an early warning with respect to particular event associated with emission and / or presence of gasses, advantageously allowing for earlier triggering and faster response. In further embodiments, after a particular event associated with emission and / or presence of gasses is successfully detected, the learning algorithm may retrospectively analyze sensor measurements prior to the detection to determine patterns in gaseous readings associated with ultimately determined events. In further embodiments, both the patterns and the time prior to standard detection at which the pattern becomes evident may be determined and used to trigger event alerts as to future occurrences. In embodiments, the learning algorithm may store such gaseous measurements used in this process within memory of the device or remotely. In further embodiments, a model may be specifically trained for a particular device, a particular patient or a particular environment.

[0102] In an embodiment, the device may have an energy efficiency mode. Advantageously, this conserves battery life and power consumption, and improves the longevity of the device due to loss of sensitivity and fidelity of sensor components due to excess “sniffing.” In embodiments, such energy efficiency mode may be achieved through dynamic adjustment of the rate of sensor readings, that is, dynamic adjustment of “sniff” frequency. In embodiments, a default mode may be the taking of sparse measurement at a lower frequency (for example, once per fifteen seconds, once per thirtyseconds, once per minute, or once per two minutes), and then, upon detection of gaseous measurements indicative of a potential particular event associated with emission and / or presence of gasses, sniff frequency may increase (for example, once per five seconds, two seconds, one second, or two, three or five times per second). The learning algorithm may be configured to only configure an alert based on readings at the increased sniff frequency, thereby allowing additional confidence as to an event occurring. The sniff frequency may return, upon a period of relatively inactivity, for example a period of five minutes, ten minutes, or fifteen minutes, to its baseline lower level.

[0103] In an embodiment, the device may generate reports. In certain embodiments, the report may be generated on a fixed schedule, for example, daily, weekly, biweekly, or monthly. The report may indicate, for a particular subject in highest proximity to the device, one or more or all of frequency, timing, and quality of events.

[0104] Fig. 1 shows an exemplary design of an artificial-intelligence- enabled system configured to monitor for particular events associated with emission and / or presence of gasses in an environment, according to embodiments of the invention. As can be seen, the device may comprise gas-specific sensors 1, for example specific to gases associated with the particular event to be detected, or for example metal oxide (MOX) sensors, electrochemical sensors, or combustible gas sensors, environmental sensors 2, for example pertaining to one or more or all of temperature, pressure, and humidity, and VOC sensors 3, which in embodiments may be generic or commercially available VOC sensors. The device may further comprise a power supply 4, which may be, for example, a battery or a plug-in power supply, or both, an airflow fan 5 which may be directed to direct air towards some or all of the sensors 1 , 2, 3, a cooling fan or muffin fan 6, directed to avoid overheating of the device and of its processor 7, which may be for example an Al chip, memory storage 8, which may store, for example, training data and the models used, cellular / wifi circuitry 9, or which may be, for example, wireless, Bluetooth, or ethernet circuitry or equipment, or alarm, which may be, for example, a screen, flashing lights, or sound, which may be used as an indicator to transmit or locally indicate, respectively, an alert as to event activity and data to a monitoring agent or a centralized repository, an interface which may comprise one or both of an LCD panel 10 for displaying settings and information pertaining to settings for the device and the patient and a feedback button or buttons 11 which may be used by a user to provide feedback as to a false positive (no event occurred) or a true positive (event occurred), or a false negative (a event occurred in the absence of a notification) and an enclosure 12 which may be waterproof. In a further embodiment, enclosure 12 may be coated with a waterproof coating, for example a teflon coating.

[0105] Fig. 2 shows a flowchart of an exemplary method for applying a model for detection of fecal events, and for updating the model based on feedback data along with calibration data and training data, according to embodiments of the invention. As will be appreciated, this exemplary method may be applied to detection of any particular event associated with emission and / or presence of gasses, where instead of a “fecal alert * health score" the alert may be of the particular event and optionally characteristics of the particular event.

[0106] As can be seen, sensor measurements may be obtained, for example via sensors 1 , 2, 3, and the input may be subjected to data preprocessing, for example background subtraction and normalization such as is discussed herein. A model, such as a simple model, a one-class classification model, or a supervised classification model, may be used to determine whether there is a potential indication of a particular event. Data pertaining to selection or adjustment or use of a radius of sensitivity may be used by such model. Alternately or additionally, a sensitivity threshold may be employed by such model.

[0107] Upon determination of the presumed positive signal, the energy efficient mode may be deactivated, and the sampling frequency of the sensors may temporarily increase. Using the higher frequency input data from the more frequent measurements, a determination may be made as to whether the signal is persistent or merely transient.

[0108] In the event it is determined to be persistent, the associated data may be stored, which may be used to train, for example according to a secondary model, an early activation or early detection mode useful to more quickly determine events for future occurrences.

[0109] Further, upon determination of persistence, a further model may be employed, for example, to more accurately determine the presence of an event in the presence of higher frequency data, to determine additional characteristics of the event. Based on this analysis, a notification may be triggered, as to the event, and, where determinable, additional characteristics of the event.

[0110] Upon the trigger of the notification, the user may provide feedback, for example via a user interface, as to whether the notificationrepresented a true positive (an actual event) or a false positive (no event). The user may, in embodiments, provide feedback, for example via the user interface, as to whether there has been a false negative (an actual event despite the absence of the trigger of a notification). Such feedback may be used to update the model used for the detection of events and triggering of notifications. For example, it may be used to determine sensitivities to be associated with particular calibration settings. In addition, training data may used in conjunction with, or prior to, the new feedback information, in order to create the initial version of such detection model.

[0111] In embodiments, the device and methods as discussed herein may be used in connection with additional applications. For example, they may be used, for example, with their own training data and associated models.

[0112] They may be used in public spaces, such as restaurants, bathrooms, airports or anywhere there is need to detect particular events associated with emission and / or presence of gasses.

[0113] In certain embodiments, in addition to or instead of triggering an alert, an automated action may be performed. In some embodiments, the automated action may be the release of a substance from a communicatively coupled device or from the device itself. In some embodiments, the automated action may be the commencement of cleaning by a robot cleaner, such as a robot vacuum, or a litter box. In embodiments, the robot cleaner or litter box may be communicatively coupled with the device, or may be integrated with the device.

[0114] In embodiments, the device may be integrated with or communicatively coupled with a monitoring camera. Such camera may turnon, or have its feed provided to a monitoring agent, upon, the determination that an event has occurred.Aspects of Al-enable detection

[0115] Aspect 1. A monitor comprising: (a) a housing; (b) a plurality of sensors, including: (i) a plurality of gas-specific sensors contained within the housing, wherein each gas-specific sensor is configured to detect at least one specific gas associated with a specified event associated with emission and / or presence of gasses; (ii) a plurality of environmental sensors contained within the housing; and (iii) a plurality of volatile organic compound (VOC) sensors contained within the housing, configured to detect at least one volatile organic compound; (c) an airflow fan operatively connected to the housing and configured to direct environmental air towards the plurality of sensors; (d) an indicator configured to indicate to a user that the specified event has occurred; (e) a processor contained within the housing, operatively connected to the indicator and to memory configured to store event data and a time-series classification model trained based on training data comprising a time series of the following data tagged with event activity: (i) a plurality of gas-specific sensor data; (ii) a plurality of environmental sensor data; and (iii) a plurality of volatile organic compound sensor data, wherein processor executable instructions are stored in memory, which, when executed by the processor, perform the steps of: (i) receiving from each of the plurality of sensors, at a respective time period, respective sensor data reflecting the output of each respective sensor; (ii) storing the respective sensor data in association with the respective time period in memory; (iii) determining event activity using the time-series classification model, where the input is the respective sensor data, the outputted result is a determination of event activity, and at least one of a neural network algorithm, an anomaly detectionalgorithm, an outlier detection algorithm, and a one-class learning algorithm, is employed; and (iv) upon determination of event activity, providing the determination to the indicator; and (f) a power supply configured to power the airflow fan, the indicator, and the processor.

[0116] Aspect 2. The monitor of aspect 1 , wherein the environmental sensors comprise a humidity sensor to detect humidity, a temperature sensor to detect temperature and a pressure sensor to detect pressure.

[0117] Aspect 3. The monitor of aspect 1 , wherein the neural network algorithm is employed.

[0118] Aspect 4. The monitor of aspect 1 , wherein the anomaly detection algorithm is employed.

[0119] Aspect 5. The monitor of aspect 1 , wherein the outlier detection algorithm is employed.

[0120] Aspect 6. The monitor of aspect 1 , wherein the one-class learning algorithm is employed.

[0121] Aspect 7. The monitor of aspect 1 , wherein a plurality of the neural network algorithm, the anomaly detection algorithm, the outlier detection algorithm, and the one-class learning algorithm are employed.

[0122] Aspect 8. The monitor of aspect 1 , wherein the gas-specific sensors and the respective sensor data measure hydrogen sulfide, indole, skatole, mercaptans, methane, methyl sulfide, 3-Methylindole, furans, ethyl dodecanoate, phenols, butanol, ester, hydrocarbons and ammonia.

[0123] Aspect 9. The monitor of aspect 1 , further comprising a user interface outside of the housing configured to receive from the user a selectable indication of whether or not a specified event has occurred,wherein the time series of data further comprises a distance from a measurement device; and wherein, the processor is further configured to, after step iv: (v) receive from the user interface a selectable indication of whether or not the specified event activity has occurred; (vi) update the sensitivity of the time-series classification model with respect to the distance based on whether the selectable indication indicates a true positive or a false positive with respect to the specified event activity; and (vii) store within memory, as part of the specified event data, at least one of the determination and the selectable indication.

[0124] Aspect 10. The monitor of aspect 9, wherein the processor is further configured to receive from the user interface a selectable indication that a false negative as to event activity has occurred, and to update the sensitivity of the time-series classification model with respect to the distance based on the selectable indication that the false negative has occurred.

[0125] Aspect 11. The monitor of aspect 9, wherein the processor is further configured to receive from the user interface a selectable indication that a true negative as to event activity has occurred, and to update the sensitivity of the time-series classification model with respect to the distance based on the selectable indication that the true negative has occurred.

[0126] Aspect 12. The monitor of aspect 9, wherein the user interface is on the exterior of the housing.

[0127] Aspect 13. The monitor of aspect 9, wherein the user interface is separated from the housing.

[0128] Aspect 14. The monitor of aspect 1 , wherein the processor is further configured to classify at least one background odor based on theoutput, and wherein the determination of whether an event is likely to have occurred is further based on the classification of the at least one background odor.

[0129] Aspect 15. The monitor of aspect 1 , wherein the power supply comprises an external power source and a battery configured to power the monitor upon disconnection of the external power source.

[0130] Aspect 16. The monitor of aspect 1 , wherein the power supply comprises an external power source.

[0131] Aspect 17. The monitor of aspect 1 , wherein the power supply comprises a battery.

[0132] Aspect 18. The monitor of aspect 1 , wherein the processor comprises a graphics processing unit (GPU) chip.

[0133] Aspect 19. The monitor of aspect 18, wherein the graphics processing unit (GPU) chip is an ultra-low-power graphics processing unit (GPU) chip.

[0134] Aspect 20. The monitor of aspect 1 , wherein the processor comprises a central processing unit (CPU).

[0135] Aspect 21. The monitor of aspect 1 , wherein the processor comprises an ultra-low-power central processing unit (CPU).

[0136] Aspect 22. The monitor of aspect 1 , wherein the indicator is located on the exterior of the housing and indicates that the event has occurred by a local sound or visual display.

[0137] Aspect 23. The monitor of aspect 1 , wherein the indicator comprises a transmitter configured to indicate that the event has occurred by transmitting an alert to a separate device in communication with the user.

[0138] Aspect 24. The monitor of aspect 23, wherein the transmitter is further configured to transmit the event data to a centralized repository.

[0139] Aspect 25. The monitor of aspect 23, wherein the transmitter is further configured to receive updated firmware for storage in memory and execution by at least one processor.

[0140] Aspect 26. The monitor of aspect 1 , wherein the housing is a waterproof enclosure.

[0141] Aspect 27. The monitor of aspect 1 , wherein the training data is further tagged with a characteristic of the event.

[0142] Aspect 28. The monitor of aspect 27, wherein the processor is further configured to, upon the determination of event activity, determine a characteristic of the event based on the training data.

[0143] Aspect 29. The monitor of aspect 28, wherein the determination of the characteristic by the processor is performed using a secondary multiclass classifier stored in memory.

[0144] Aspect 30. The monitor of aspect 9, wherein the processor is further configured to: (a) represent, within the time-series classification model, the distance from the measurement device as data laying within a high-dimensional space; (b) determine a compact envelope, representative of a sensitivity threshold, surrounding a portion of the data laying within the high-dimensional space; and (c) modify the boundaries of the compactenvelope based on the selectable indication of whether or not the event has occurred.

[0145] Aspect 31. The monitor of aspect 30, wherein at least one of the determination of the compact envelope and the modification of the boundaries of the compact envelope is performed based on a second one- class learning algorithm in which the class members are exemplars of events.

[0146] Aspect 32. The monitor of aspect 30, wherein at least one of the determination of the compact envelope and the modification of the boundaries of the compact envelope is performed based on calibration information.

[0147] Aspect 33. The monitor of aspect 32, wherein the calibration information comprises a user-selected radius of detectability.

[0148] Aspect 34. The monitor of aspect 9, wherein the neural network is employed and is configured to learn, from the output, consistently present neutral background odors, and to subtract a contribution of the learned consistently present neutral background odors from the output.

[0149] Aspect 35. The monitor of aspect 34, wherein the learning of the consistently present neutral background odors is performed using a shortterm habituation algorithm.

[0150] Aspect 36. The monitor of aspect 35, wherein the short-term habituation algorithm habituates according to a time period of 10 seconds to 120 seconds.

[0151] Aspect 37. The monitor of aspect 34, wherein the learning of the consistently present neutral background odors is performed using ahabituation algorithm habituating according to a time period of 12 hours to 24 hours.

[0152] Aspect 38. The monitor of aspect 9, wherein the processor is configured to normalize the output.

[0153] Aspect 39. The monitor of aspect 38, wherein the normalization is performed by divisive normalization.

[0154] Aspect 40. The monitor of aspect 38, wherein the normalization is performed by batch normalization.

[0155] Aspect 41. The monitor of aspect 38, wherein the normalization is performed by min-max normalization.

[0156] Aspect 42. The monitor of aspect 38, wherein the normalization is performed by z-score normalization.

[0157] Aspect 43. The monitor of aspect 38, wherein the normalization is performed by unit vector normalization.

[0158] Aspect 44. The monitor of aspect 9, wherein the processor is further configured to: (a) determine, based on the event data, commonalities in output from time periods prior to determinations that the event is likely to have occurred that have been confirmed as true positives based on the selectable indication; and (b) based on the determined commonalities, adjust a time boundary at which the determining the event activity occurs.

[0159] Aspect 45. The monitor of aspect 44, wherein the determination of the commonalities and the adjustment of the time boundary are performed by the processor based on a secondary model stored in memory.

[0160] Aspect 46. The monitor of aspect 1 , further configured to dynamically adjust a frequency at which measurements are taken by one or more of the plurality of gas-specific sensors, the plurality of environmental sensors, and the plurality of VOC sensors.

[0161] Aspect 47. The monitor of aspect 46, wherein the dynamic adjustment comprises taking sparse measurements, increasing the frequency of measurement upon detection of output potentially indicative of a event, and then returning to taking sparse measurements upon detection of output unlikely to be indicative of an event.

[0162] Aspect 48. The monitor of aspect 1 , wherein the event activity comprises production of feces.

[0163] Aspect 49. The monitor of aspect 1 , wherein the event activity comprises production of urine.

[0164] Aspect 50. The monitor of aspect 1 , wherein the event activity comprises production of feces and urine.

[0165] Aspect 51. The monitor of aspect 1 , wherein the event activity comprises an indication that the specified event occurred, or an indication that the specified event has not occurred.

[0166] Aspect 52. The monitor of aspect 1 , wherein the event activity includes an indication associated with the likelihood that the specified event occurred.

[0167] Aspect 53. The monitor of aspect 1 , wherein the plurality of gasspecific sensors include an ammonia sensor which is configured to detect the presence and amount of ammonia in a gas sample.

[0168] Aspect 54. The monitor of aspect 1 , wherein the plurality of gasspecific sensors include a hydrogen sulfide sensor which is configured to detect the presence and amount of hydrogen sulfide in a gas sample.

[0169] Aspect 55. The monitor of aspect 1 , wherein the plurality of gasspecific sensors include an indole sensor which is configured to detect the presence and amount of indole in a gas sample.

[0170] Aspect 56. The monitor of aspect 1 , wherein the plurality of gasspecific sensors include a skatole sensor which is configured to detect the presence and amount of skatole in a gas sample.

[0171] Aspect 57. The monitor of aspect 1 , wherein the plurality of gasspecific sensors include a mercaptans sensor which is configured to detect the presence and amount of mercaptans in a gas sample.

[0172] Aspect 58. The monitor of aspect 1 , wherein the plurality of gasspecific sensors include a methane sensor which is configured to detect the presence and amount of methane in a gas sample.

[0173] Aspect 59. The monitor of aspect 1 , wherein the plurality of gasspecific sensors include a methyl sulfide sensor which is configured to detect the presence and amount of methyl sulfide in a gas sample.

[0174] Aspect 60. The monitor of aspect 1 , further comprising a cooling fan or muffin fan configured to cool the processor.

[0175] Aspect 61. The monitor of aspect 1 , wherein the processor comprises a plurality of processors.

[0176] Aspect 62. The monitor of aspect 1 , wherein the indicator comprises an LCD panel.

[0177] Aspect 63. The monitor of aspect 30, wherein the envelope is a hypersphere.

[0178] Aspect 64. A method of monitoring comprising the steps of: (a) training a time-series classification model based on training data comprising a time series of the following data tagged with event activity: (i) a plurality of gas-specific sensor data; (ii) a plurality of environmental sensor data; and (iii) a plurality of volatile organic compound sensor data; (b) directing, via an airflow fan, environmental air towards a plurality of sensors; (c) receiving from each of a plurality of sensors, at a respective time period, respective sensor data reflecting the output of each respective sensor; (d) storing the respective sensor data in association with the respective time period in memory; (e) determining event activity using the time-series classification model, where the input is the respective sensor data, the outputted result is a determination of event activity, and at least one of a neural network algorithm, an anomaly detection algorithm, an outlier detection algorithm, a one-class learning algorithm, and a supervised learning algorithm, is employed; and (f) upon determination of event activity, providing the determination to an indicator, wherein the event is an event associated with emission and / or presence of gasses.

[0179] Aspect 65. The method of monitoring of aspect 64, further comprising, after step f, performing the steps of: (g) receiving from a user a selectable indication of whether or not the event activity has occurred; (h) updating the sensitivity of the time-series classification model with respect to a distance based on whether the selectable indication indicates a true positive or a false positive with respect to the event activity; and (i) storing within memory at least one of the determination and the selectable indication.

[0180] Although the invention has been described with reference to a particular arrangement of parts, features and the like, these are not intended to exhaust all possible arrangements or features, and indeed many other modifications and variations will be ascertainable to those of skill in the art. The embodiments and examples shown above are illustrative, and many variations can be introduced to them without departing from the spirit of the disclosure. For example, elements and / or features of different illustrative and exemplary embodiments herein may be combined with each other and / or substituted with each other within the scope of the disclosure. In particular, multiple forms of model training, event identification and alerts based thereon discussed herein may be combined and employed within a particular detectors and methods. For a better understanding of the disclosure, reference should be had to any accompanying drawings and descriptive matter in which there is illustrated exemplary embodiments of the present invention.

Claims

What is claimed is:

1. A monitor comprising:(a) a housing;(b) a plurality of sensors, including:(i) a plurality of gas-specific sensors contained within the housing, wherein each gas-specific sensor is configured to detect at least one specific gas associated with a specified event associated with emission and / or presence of gasses;(ii) a plurality of environmental sensors contained within the housing; and(iii) a plurality of volatile organic compound (VOC) sensors contained within the housing, configured to detect at least one volatile organic compound;(c) an airflow fan operatively connected to the housing and configured to direct environmental air towards the plurality of sensors;(d) an indicator configured to indicate to a user that the specified event has occurred;(e) a processor contained within the housing, operatively connected to the indicator and to memory configured to store event data and a time-series classification model trained based on training data comprising a time series of the following data tagged with event activity:(i) a plurality of gas-specific sensor data;(ii) a plurality of environmental sensor data; and(iii) a plurality of volatile organic compound sensor data, wherein processor executable instructions are stored in memory, which, when executed by the processor, perform the steps of:(i) receiving from each of the plurality of sensors, at a respective time period, respective sensor data reflecting the output of each respective sensor;(ii) storing the respective sensor data in association with the respective time period in memory;(iii) determining event activity using the time-series classification model, where the input is the respective sensor data, the outputted result is a determination of event activity, and at least one of a neural network algorithm, an anomaly detection algorithm, an outlier detection algorithm, and a one-class learning algorithm, is employed; and(iv) upon determination of event activity, providing the determination to the indicator; and(f) a power supply configured to power the airflow fan, the indicator, and the processor.

2. A fecal monitor comprising:(a) a housing;(b) a plurality of sensors, including:(i) a plurality of gas-specific sensors contained within the housing, wherein each gas-specific sensor is configured to detect at least one specific gas associated with at least one of feces and urine;(ii) a plurality of environmental sensors contained within the housing; and(iii) a plurality of volatile organic compound (VOC) sensors contained within the housing, configured to detect at least one volatile organic compound;(c) an airflow fan operatively connected to the housing and configured to direct environmental air towards the plurality of sensors;(d) an indicator configured to indicate to a user that a fecal event has occurred;(e) a processor contained within the housing, operatively connected to the indicator and to memory configured to store fecal event data and a time-series classification model trained based on training data comprising a time series of the following data tagged with fecal event activity:(i) a plurality of gas-specific sensor data;(ii) a plurality of environmental sensor data; and(iii) a plurality of volatile organic compound sensor data, wherein processor executable instructions are stored in memory, which, when executed by the processor, perform the steps of:(v) receiving from each of the plurality of sensors, at a respective time period, respective sensor data reflecting the output of each respective sensor;(vi) storing the respective sensor data in association with the respective time period in memory;(vii) determining fecal event activity using the time-series classification model, where the input is the respective sensor data, the outputted result is a determination of fecal event activity, and at least one of a neural network algorithm, an anomaly detection algorithm, an outlier detection algorithm, and a one-class learning algorithm, is employed; and(viii) upon determination of fecal event activity, providing the determination to the indicator; and(f) a power supply configured to power the airflow fan, the indicator, and the processor.

3. The fecal monitor of claim 2, wherein the environmental sensors comprise a humidity sensor to detect humidity, a temperature sensor to detect temperature and a pressure sensor to detect pressure.

4. The fecal monitor of claim 2, wherein the neural network algorithm is employed.

5. The fecal monitor of claim 2, wherein the anomaly detection algorithm is employed.

6. The fecal monitor of claim 2, wherein the outlier detection algorithm is employed.

7. The fecal monitor of claim 2, wherein the one-class learning algorithm is employed.

8. The fecal monitor of claim 2, wherein a plurality of the neural network algorithm, the anomaly detection algorithm, the outlier detection algorithm, and the one-class learning algorithm are employed.

9. The fecal monitor of claim 2, wherein the gas-specific sensors and the respective sensor data measure hydrogen sulfide, indole, skatole, mercaptans, methane, methyl sulfide, 3-Methylindole, furans, ethyl dodecanoate, phenols, butanol, ester, hydrocarbons and ammonia.

10. The fecal monitor of claim 2, further comprising a user interface outside of the housing configured to receive from the user a selectable indication of whether or not a fecal event has occurred, wherein the time series of data further comprises a distance from a measurement device; and wherein, the processor is further configured to, after step iv:(v) receive from the user interface a selectable indication of whether or not the fecal event activity has occurred;(vi) update the sensitivity of the time-series classification model with respect to the distance based on whether the selectable indication indicates a true positive or a false positive with respect to the fecal event activity; and(vii) store within memory, as part of the fecal event data, at least one of the determination and the selectable indication.1 1. The fecal monitor of claim 10, wherein the processor is further configured to receive from the user interface a selectable indication that a false negative as to fecal activity has occurred, and to update the sensitivity of the time-series classification model with respect to the distance based on the selectable indication that the false negative has occurred.

12. The fecal monitor of claim 10, wherein the processor is further configured to receive from the user interface a selectable indication that a true negative as to fecal activity has occurred, and to update the sensitivity of the time-series classification model with respect to the distance based on the selectable indication that the true negative has occurred.

13. The fecal monitor of claim 10, wherein the user interface is on the exterior of the housing.

14. The fecal monitor of claim 10, wherein the user interface is separated from the housing.

15. The fecal monitor of claim 2, wherein the processor is further configured to classify at least one background odor based on the output, and wherein the determination of whether a fecal event is likely to have occurred is further based on the classification of the at least one background odor.

16. The fecal monitor of claim 2, wherein the power supply comprises an external power source and a battery configured to power the fecal monitor upon disconnection of the external power source.

17. The fecal monitor of claim 2, wherein the power supply comprises an external power source.

18. The fecal monitor of claim 2, wherein the power supply comprises a battery.

19. The fecal monitor of claim 2, wherein the processor comprises a graphics processing unit (GPU) chip.

20. The fecal monitor of claim 19, wherein the graphics processing unit (GPU) chip is an ultra-low-power graphics processing unit (GPU) chip.

21. The fecal monitor of claim 2, wherein the processor comprises a central processing unit (CPU).

22. The fecal monitor of claim 2, wherein the processor comprises an ultra- low-power central processing unit (CPU).

23. The fecal monitor of claim 2, wherein the indicator is located on the exterior of the housing and indicates that the fecal event has occurred by a local sound or visual display.

24. The fecal monitor of claim 2, wherein the indicator comprises a transmitter configured to indicate that the fecal event has occurred by transmitting an alert to a separate device in communication with the user.

25. The fecal monitor of claim 24, wherein the transmitter is further configured to transmit the fecal event data to a centralized repository.

26. The fecal monitor of claim 24, wherein the transmitter is further configured to receive updated firmware for storage in memory and execution by at least one processor.

27. The fecal monitor of claim 2, wherein the housing is a waterproof enclosure.

28. The fecal monitor of claim 2, wherein the training data is further tagged with a metabolic or physiological condition.

29. The fecal monitor of claim 28, wherein the metabolic or physiological condition comprises at least one of: normal stool, diarrhea, C. diff infection, rotavirus infection, Campylobacter stool, bacterial infection, fungal infection, metabolic state abnormality, indication of colon cancer, inflammatory bowel disease, catheter-associated urinary tract infection, hard stool, and soft stool.

30. The fecal monitor of claim 28, wherein the processor is further configured to, upon the determination of fecal event activity, determine a physiological condition based on the training data.

31. The fecal monitor of claim 30, wherein the determination of the physiological condition by the processor is performed using a secondary multi-class classifier stored in memory.

32. The fecal monitor of claim 10, wherein the processor is further configured to:(a) represent, within the time-series classification model, the distance from the measurement device as data laying within a high-dimensional space;(b) determine a compact envelope, representative of a sensitivity threshold, surrounding a portion of the data laying within the high-dimensional space; and(c) modify the boundaries of the compact envelope based on the selectable indication of whether or not the fecal event has occurred.

33. The fecal monitor of claim 32, wherein at least one of the determination of the compact envelope and the modification of the boundaries of the compact envelope is performed based on a second one-class learning algorithm in which the class members are exemplars of fecal events.

34. The fecal monitor of claim 32, wherein at least one of the determination of the compact envelope and the modification of the boundaries of the compact envelope is performed based on calibration information.

35. The fecal monitor of claim 34, wherein the calibration information comprises a user-selected radius of detectability.

36. The fecal monitor of claim 10, wherein the neural network is employed and is configured to learn, from the output, consistently present neutral background odors, and to subtract a contribution of the learned consistently present neutral background odors from the output.

37. The fecal monitor of claim 36, wherein the learning of the consistently present neutral background odors is performed using a short-term habituation algorithm.

38. The fecal monitor of claim 37, wherein the short-term habituation algorithm habituates according to a time period of 10 seconds to 120 seconds.

39. The fecal monitor of claim 36, wherein the learning of the consistently present neutral background odors is performed using a habituationalgorithm habituating according to a time period of 12 hours to 24 hours.

40. The fecal monitor of claim 10, wherein the processor is configured to normalize the output.41 . The fecal monitor of claim 40, wherein the normalization is performed by divisive normalization.

42. The fecal monitor of claim 40, wherein the normalization is performed by batch normalization.

43. The fecal monitor of claim 40, wherein the normalization is performed by min-max normalization.

44. The fecal monitor of claim 40, wherein the normalization is performed by z-score normalization.

45. The fecal monitor of claim 40, wherein the normalization is performed by unit vector normalization.

46. The fecal monitor of claim 10, wherein the processor is further configured to:(a) determine, based on the fecal event data, commonalities in output from time periods prior to determinations that the fecal event is likely to have occurred that have been confirmed as true positives based on the selectable indication; and(b) based on the determined commonalities, adjust a time boundary at which the determining the fecal event activity occurs.

47. The fecal monitor of claim 46, wherein the determination of the commonalities and the adjustment of the time boundary are performed by the processor based on a secondary model stored in memory.

48. The fecal monitor of claim 2, further configured to dynamically adjust a frequency at which measurements are taken by one or more of the plurality of gas-specific sensors, the plurality of environmental sensors, and the plurality of VOC sensors.

49. The fecal monitor of claim 48, wherein the dynamic adjustment comprises taking sparse measurements, increasing the frequency of measurement upon detection of output potentially indicative of a fecal event, and then returning to taking sparse measurements upon detection of output unlikely to be indicative of a fecal event.

50. The fecal monitor of claim 2, wherein the fecal event activity comprises production of feces.51 . The fecal monitor of claim 2, wherein the fecal event activity comprises production of urine.

52. The fecal monitor of claim 2, wherein the fecal event activity comprises production of feces and urine.

53. The fecal monitor of claim 2, wherein the fecal event activity comprises an indication that a fecal event occurred, or an indication that the fecal event has not occurred.

54. The fecal monitor of claim 2, wherein the fecal event activity includes an indication associated with the likelihood that a fecal event occurred.

55. The fecal monitor of claim 2, wherein the plurality of gas-specific sensors include an ammonia sensor which is configured to detect the presence and amount of ammonia in a gas sample.

56. The fecal monitor of claim 2, wherein the plurality of gas-specific sensors include a hydrogen sulfide sensor which is configured to detect the presence and amount of hydrogen sulfide in a gas sample.

57. The fecal monitor of claim 2, wherein the plurality of gas-specific sensors include an indole sensor which is configured to detect the presence and amount of indole in a gas sample.

58. The fecal monitor of claim 2, wherein the plurality of gas-specific sensors include a skatole sensor which is configured to detect the presence and amount of skatole in a gas sample.

59. The fecal monitor of claim 2, wherein the plurality of gas-specific sensors include a mercaptans sensor which is configured to detect the presence and amount of mercaptans in a gas sample.

60. The fecal monitor of claim 2, wherein the plurality of gas-specific sensors include a methane sensor which is configured to detect the presence and amount of methane in a gas sample.

61. The fecal monitor of claim 2, wherein the plurality of gas-specific sensors include a methyl sulfide sensor which is configured to detect the presence and amount of methyl sulfide in a gas sample.

62. The fecal monitor of claim 2, further comprising a cooling fan or muffin fan configured to cool the processor.

63. The fecal monitor of claim 2, wherein the processor comprises a plurality of processors.

64. The fecal monitor of claim 2, wherein the indicator comprises an LCD panel.

65. The fecal monitor of claim 32, wherein the envelope is a hypersphere.

66. A method of fecal monitoring comprising the steps of:(a) training a time-series classification model based on training data comprising a time series of the following data tagged with fecal event activity:(i) a plurality of gas-specific sensor data;(ii) a plurality of environmental sensor data; and(iii) a plurality of volatile organic compound sensor data;(b) directing, via an airflow fan, environmental air towards a plurality of sensors;(c) receiving from each of a plurality of sensors, at a respective time period, respective sensor data reflecting the output of each respective sensor;(d) storing the respective sensor data in association with the respective time period in memory;(e) determining fecal event activity using the time-series classification model, where the input is the respective sensor data, the outputted result is a determination of fecal event activity, and at least one of a neural network algorithm, an anomaly detection algorithm, an outlier detection algorithm, a one-class learning algorithm, and a supervised learning algorithm, is employed; and(f) upon determination of fecal event activity, providing the determination to an indicator.

67. The method of fecal monitoring of claim 66, further comprising, after step f, performing the steps of:(g) receiving from a user a selectable indication of whether or not the fecal event activity has occurred;(h) updating the sensitivity of the time-series classification model with respect to a distance based on whether the selectable indication indicates a true positive or a false positive with respect to the fecal event activity; and(i) storing within memory at least one of the determination and the selectable indication.

Images