TRACING OF COVID-19 VACCINE VIALS

- Wiliot, Ltd.

A system and method for tracing vaccine vials are provided. The method includes receiving, from a gateway of a plurality of gateways, frequency words from tags attached to vaccine vials, wherein each tag is configured to transmit a plurality of frequency words; extracting at least one data feature from the plurality of frequency words, wherein each data feature changes in response to a change in a state of a vaccine vial; classifying the extracted data feature based on a machine learning model trained with respect to a location of the gateway, wherein the classifier is trained to label a trace parameter indicative of a state of a vaccine vial; and sending a semantic event indicating a value of the trace parameter.

Skip to: Description  ·  Claims  · Patent History  ·  Patent History
Description
CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims the benefit of U.S. Provisional Application No. 62/977,433 filed on Feb. 17, 2020, the contents of which are hereby incorporated by reference.

TECHNICAL FIELD

The present disclosure relates generally to a system and method for detecting materials in containers, more particularly to detecting materials in vials using low-energy sensing and machine learning.

BACKGROUND

There are many ways to detect materials, or the make-up of a material on a surface. For example, drawing from spectroscopy, reflection of light from a surface of a road may be detected, and the intensity of the detected light may be measured to indicate whether there is ice on a road surface.

Other electromagnetic spectrums may be used to detect materials. For example, infrared signals reflected off of materials or surfaces may be detected, and based on heat signatures unique to materials, the material may be identified. Other modalities, including ultrasound, may be used.

However, the devices used under these modalities are often expensive and complicated, and use signals that are incompatible with signals used among devices to communicate with each other. Also, these devices require a lot of energy to operate, further adding to their cost. In addition, such devices cannot detect materials in containers that are in transit. Further, it is difficult to make sense of the data collected, and a lot of resources need to be used to interpret the data.

Vaccination and, in particular, vaccination for COVID-19, requires special procedures to store, thaw, prepare, and administer. The messenger ribonucleic acid (mRNA) based COVID-19 vaccine now offered by Prizer® and Moderna® requires storage at −80° C. and −20° C., respectively. Then, the Vaccine has be thawed in a refrigerator or at room temperature. It may take a number of hours to thaw in the refrigerator or at a room temperature.

The Center for Disease Control (CDC) provides a number of steps to prepare the vaccine. The steps include removing the vaccine from the freezer or refrigerator and allowing the vaccine to come to room temperature. Vials can be held at room temperature for up to 2 hours before mixing. After 2 hours, return unmixed vials to the refrigerator. With the vaccine at room temperature, the vial has to be generally inverted a number of times. Then, using a 21-gauge (or narrower) needle, 1.8 mL of 0.9% sodium has to be withdrawn into a mixing syringe. After use, the diluent vial and any remaining diluent has to be discarded. Vials at room temperature must be mixed within 2 hours or returned to the refrigerator. Also, a single vial now can be used to administrate 5 or 6 doses.

As noted, there are a number of preparation steps that may yield for mistakes performed by the nurses administrating the vaccine. So far, it has been reported on patients that have been administered with a complete vial (6 doses instead of 1), vials that have not been thawed to room temperature have been administered.

To ensure that the preparation and administration is performed without mistakes or with minimal mistakes, there is a need to trace the preparation and administration process (or pipeline). Currently, this is performed by medical personnel. However, due to the high number of vaccines being administered, even well-trained personnel can make mistakes. One solution to monitor the temperature of a vial is a vaccine vial monitor (VVM). This is a thermochromic label put on vials containing vaccines which gives a visual indication of whether the vaccine has been kept at a temperature which preserves its potency. However, as the COVID-based vaccines should be kept at different temperatures (deep freeze, refrigerator, and room temperature), a single VVM may not be a feasible solution.

It would therefore be advantageous to provide a solution that would overcome the challenges noted above.

SUMMARY

A summary of several example embodiments of the disclosure follows. This summary is provided for the convenience of the reader to provide a basic understanding of such embodiments and does not wholly define the breadth of the disclosure. This summary is not an extensive overview of all contemplated embodiments, and is intended to neither identify key or critical elements of all embodiments nor to delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more embodiments in a simplified form as a prelude to the more detailed description that is presented later. For convenience, the term “some embodiments” or “certain embodiments” may be used herein to refer to a single embodiment or multiple embodiments of the disclosure.

Certain embodiments disclosed herein include a method for tracing vaccine vials are provided. The method includes receiving, from a gateway of a plurality of gateways, frequency words from tags attached to vaccine vials, wherein each tag is configured to transmit a plurality of frequency words; extracting at least one data feature from the plurality of frequency words, wherein each data feature changes in response to a change in a state of a vaccine vial; classifying the extracted data feature based on a machine learning model trained with respect to a location of the gateway, wherein the classifier is trained to label a trace parameter indicative of a state of a vaccine vial; and sending a semantic event indicating a value of the trace parameter.

Certain embodiments disclosed herein include a system for developing a treatment plan using multi-stage machine learning. The system includes a processing circuitry; and a memory, the memory containing instructions that, when executed by the processing circuitry, configure the system to: receive from a gateway of a plurality of gateways, frequency words from tags attached to vaccine vials, wherein each tag is configured to transmit a plurality of frequency words; extract at least one data feature from the plurality of frequency words, wherein each data feature changes in response to a change in a state of a vaccine vial; classify the extracted data feature based on a machine learning model trained with respect to a location of the gateway, wherein the classifier is trained to label a trace parameter indicative of a state of a vaccine vial; and send a semantic event indicating a value of the trace parameter.

BRIEF DESCRIPTION OF THE DRAWINGS

The subject matter disclosed herein is particularly pointed out and distinctly claimed in the claims at the conclusion of the specification. The foregoing and other objects, features, and advantages of the disclosed embodiments will be apparent from the following detailed description taken in conjunction with the accompanying drawings.

FIG. 1 is a schematic diagram of a secure low energy communication system utilized to describe the various embodiments.

FIG. 2 is a flow diagram illustrating a training phase for a supervised model utilized for detecting a material, according to an embodiment.

FIG. 3 is a graph showing a frequency word telemetry corresponding to labeled items in the container, according to an embodiment.

FIG. 4 is a flowchart for a method of training a model for tracing vaccine vials, according to an embodiment.

FIG. 5 is a flowchart of a method for tracing vaccine vials using low energy sensing according to an embodiment.

FIG. 6 is a circuit diagram of a gateway used for detecting a material using low energy sensing, according to an embodiment.

FIG. 7 is a schematic diagram of a wireless device, designed according to the disclosed embodiments.

FIG. 8 is a schematic diagram of a server according to an embodiment.

FIG. 9 is a picture of a vaccine vial and an IoT tag in a form factor of a label attached thereon according to an embodiment.

DETAILED DESCRIPTION

It is important to note that the embodiments disclosed herein are only examples of the many advantageous uses of the innovative teachings herein. In general, statements made in the specification of the present application do not necessarily limit any of the various claimed embodiments. Moreover, some statements may apply to some inventive features but not to others. In general, unless otherwise indicated, singular elements may be in plural and vice versa with no loss of generality. In the drawings, like numerals refer to like parts through several views.

The various disclosed embodiments include a method and system for tracing vaccine vials based on machine learning techniques and signals received from IoT tags attached to the vials. Tracing of vials is achieved through the preparation and administration of the vaccines. In an embodiment, the vaccine vials being traced are messenger ribonucleic acid (mRNA) vials used for COVID-19 immunization. According to the disclosed embodiments, the tracing is of level of doses in a vial, if the vial is empty or full, a temperature of a vial, and a location of vial.

FIG. 1 is an example schematic diagram of a secure low energy communication system 100 utilized to describe the various embodiments. The system 100 includes a plurality of Internet of Things (IoT) tags 110-1 through 110-n (collectively referred to as a IoT tag 110 or IoT tags 110), a plurality of gateways 120-1 through 120-5, and a cloud computing platform 130. The system 100 also includes at least one server 140 that may be deployed in the cloud-based platform 130. The server 140 may be realized as a physical machine, a virtual machine, or a combination thereof. An example block diagram of the server 140 is provided in FIG. 2. The cloud computing platform 130 may be a public cloud, a private cloud, or a hybrid cloud. A database 145 may also deployed in the platform 130 and may be connected to the server 140. The database 145 may store events generate by server 145, identifiers (IDs) of IoT tags 110, and other signals.

In an embodiment, the IoT tag 110 is a battery-free IoT tag as discussed in FIG. 7. Also, communication among the IoT tag 110 and the gateway 120 may be performed using a low-energy communication protocol. The communication between the cloud computing platform 130 and the gateway 120 is over, for example, the Internet.

In an example embodiment, the low-energy communication protocol includes a Bluetooth Low Energy (BLE) protocol, which are short-wavelength radio waves operating at a range of about 2.40 to 2.485 MHz, and commonly used among portable wireless devices. The cloud computing platform 130 may include a public cloud, a private cloud, a hybrid cloud, or combination thereof.

In the example embodiment illustrated in FIG. 1, five (5) gateways 120-1 through 120-5 are illustrated, each of which is deployed at a different station in the preparation and administration of a vaccine. For example, gateway 120-1 may be deployed in a deep freeze station 101, gateway 120-2 may be deployed in a refrigeration station 102, gateway 120-3 may be deployed in a preparation station 103, gateway 120-4 may be deployed in an administration station 104, and gateway 120-5 may be deployed in a discarding station 105. The various stations are in different physical locations. As will be discussed below, different parameters are traced at different stations.

Each gateway 120 is configured to receive signals from the IoT tags 110-1 through 110-n in the respective station, encapsulate those signals together with additional data in data packets, and transmit the data packets to the cloud computing platform 130 to be processed by the server 140. The communication with the IoT tags may be performed over low-energy communication protocol (e.g., BLE) while the communication with the cloud computing platform 130 may be, for example over the Internet. An example implementation of a gateway 120 is discussed below.

Each IoT tag 110 may be attached, glued, or printed to a vaccine vial 150. The form factor of an IoT tag 110, in an embodiment, may include a label. The vaccine vial 150 is a container. An example picture showing an IoT tag 900 in a form factor of a label is placed on a vaccine vial 910 is provided in FIG. 9.

Returning to FIG. 1, the IoT tags 110 sense a particular radio frequency (RF) activity relative to each other and relative to changes in in the vials they attached to so. The sensing is performed at a certain location. When a vaccine vial 150 (e.g., the vial 150-1) is placed proximate to one of the IoT tags 110 (e.g., the IoT tag 110-1), an interference to the ambient RF field causes a difference between the electrical properties of the material in the vial and that of the ambient surroundings (e.g., air) which leads to a change in a calibration frequency of the IoT tag, such change is translated into a frequency word. As will be explained in more detail below, the IoT tag 110-1 is configured to send frequency words along with other information to the gateway 120. The gateway 120 is further configured to relay the combined information to the server 140. The server 140 is configured to perform further processing to trace vials 150-1 through 150-n in each station.

In another embodiment, each IoT tag 110 is configured to send data packets to a gateway 120. The information in such data packets is later processed by the server 140 to detect the material of the item in the vial 150 proximate to one of the IoT tags 110. It should be noted that a vial 150 may be any type of container including a capsule, a box, a bottle, and the like. The material in the vaccine vial 150 may be in a form of liquid, solid, or gas, and may include any medicated formula.

In an embodiment, a data packet transmitted by a gateway 120-i (i=1, . . . , 5) includes a digital frequency word and an Identifier (ID) of an IoT tag 110. The frequency word is measured by an IoT tag 110 depending on a frequency calibration of an IoT tag 110. Any changes to the temperature in the ambient environment or changes of the material contained in the vaccine vial placed among an IoT tags 110, will change the values detected among an IoT tags 110 from synchronization, thereby changing the value of the frequency word. In an example embodiment, the frequency calibration is discussed in more detail with reference to FIG. 7.

In an embodiment, the transmission frequency may be determined based on a digitally controlled oscillator (DCO) signal. The DCO is utilized to calibrate an internal capacitor that changes its capacitance value to counteract the capacitance that a new material imposes, the DCO signal is a control signal of the DCO. That is, the IoT tag senses whether the transmission frequency is corrected and, if not, the DCO is adjusted to the main frequency of the transmission. When a different material is placed in the vaccine vial, the tag recalibrates itself to the same transmission frequency by changing the capacitance value of its internal capacitor. In an embodiment, the transmission frequency may be one of the channels of the BLE transmission protocol. It should be noted that other low energy communication protocols are also applicable.

In an embodiment, the ID is a unique identifier (ID) of the IoT tag 110 which is set during production of the IoT tag 110. The data packets sent by the IoT tag 110 are received at a gateway 120 which, in turn, is periodically configured to send the data packets to the server 140. An example block diagram of a gateway 120 is shown in FIG. 6. In an embodiment, a gateway 120 may be installed with an agent or application (not shown in FIG. 1). The agent, when executed, is configured to control the communication between the IoT tag 110 and the server 140. The agent is also configured to receive a semantic event caused by the placement of a vaccine vial proximate to the IoT tag 110. Such an event is received by the server 140 and displayed as a notification on a display on a handheld device 170. The connection between the device 170 and the server 140 is over the Internet. For example, a semantic event may be a measurement of a temperature of a vial 150. Another semantic event is an instruction to discard a vial.

The server 140 is configured to process the data packets received by the gateway 120 to detect a semantic event for each of the IoT tags 110. A semantic event may be indicative of, for example, a change in the level of contents in a vial; a temperature change of a vial; a change in location of an IoT tag 110 attached to a vial 150; and the like. The semantic event may also include an alert, for example, not to change a specific vial associated with a tag 110 in a specific location.

In an embodiment, the detection of a semantic event is based on analysis of the frequency words. At least two data features can be extracted from such words: a first data feature is extracted from the difference between two consecutive frequency words (or content of consecutive packets) received from the same tag, and a second data feature is a rate of receiving the frequency words. The first data feature can indicate the type of change, such as location, temperature, and proximity. The second data feature can indicate the number of doses in the vial. The analysis is performed using machine learning processes performed by the server 140. These processes or techniques are discussed in greater detail below. For example, tracing temperature is based on content of packet (auxiliary measurement signal for which a regression model that correlates it with temperature was built. Tracing location is based on packet content (e.g., RSSI value) and rate, and the material is based on frequency words and rate.

In the alternative, detection of RF values from multiple co-located IoT tags 110 may be analyzed within various temporal windows to determine which of the respective frequency words have anomalies. Co-location sensing may be detected by the gateway 120. When no vaccine vial is present, the IoT tags 110 may have similarly sensed RF values.

In an example embodiment, upon detection of change in a vial or vials are, a sematic event may be sent as notification from the server 140 to a handheld device 170. The semantic event may include an ID of the respective IoT tag 110 and information to be displayed. The handheld device 170 may be any monitor or device used by medical personnel, and can be realized as a smartphone, a tablet device, and the like.

In an embodiment, different machine learning models are used to analyze packets received from the gateways 120. In an embodiment, a first model may include tracing the temperature of vials in a deep freeze station 101 based on packets received from a gateway 120-1, a second model may include tracing the temperature of vials in a refrigeration station 102 based on packets received from a gateway 120-2, a third model may include tracing the temperature of vials to be a room temperature at a preparation station 103 based on packets received from a gateway 120-3, a fourth model may include tracing a level of material (vaccine dose) in each vial at an administration station 104 based on packets received from a gateway 120-4, and a fifth model may include tracing discarded vials at a discarding station 105 based on packets received from a gateway 120-5. The tracing are based on data features extracted from at least the frequency words.

FIG. 2 is a schematic diagram of a machine learning framework 200 utilized to trace a vaccine vial having an IoT tag attached thereon. The tracing includes determining the contents of the material in a vial. In an embodiment, a level of contents may be binary (empty or full), a percentage volume of the material in the vaccine vial, or a number of doses left. The framework 200 can be further configured to detect the type of surface that the vaccine vial is placed on. For the sake of simplicity and without limitation on the disclosed embodiments, FIG. 2 will be discussed also with reference to the elements shown in FIG. 1.

The supervised machine learning framework 200 operates in two phases: learning and detection. In the learning phase, a trained model 201 is generated and trained, while in the detection phase, the trained model 201 is utilized for identification of the type of contents of the vaccine vial 150. The trained model 201, in an embodiment, is generated and trained using semi-supervised machine learning techniques.

In the learning phase, packets which include frequency words are received from the IoT tags 110. Such received words are aggregated and saved as a learning dataset 210. The aggregation of frequency words may be for a predefined time window (e.g., all words during an hour-time window) or per IoT tag ID.

The learning dataset 210 is input to a filter 220. The filter 220 is configured to filter out “noisy” readings of the frequency words. The filter 220 may be implemented using a boundary decision technique, outlier detection, and the like. In an embodiment, upon filtering noisy readings, the forwarding of information of the filtered noise is propagated. Such information can be further analyzed by the feature's extractor, such that eventually the trained model refers also to a noise level which can be different with respect to different material (e.g., air may gain more noise than condensed material).

The filtered words are fed into a features extractor 230 configured to generate a set of features representing an interface event. Depending on a state of vaccine vial 150, the values of the frequency words provided by the respective tag may vary. The state of the vaccine vial may be any one of: a temperature of a vaccine value and a number of doses in the vaccine vial. By modeling the values of the frequency words, and attributing the values to specific materials, the material make-up of the vaccine vial 150 may be identified. In an example embodiment, a noise level of received frequency word may be one of the extracted features. That is, upon filtering, information on the filtered noise is fed into the features extractor 230, such that eventually the trained model 201 refers also to a noise level which can be different with respect to different material (e.g., air may gain more noise than condensed material).

In an embodiment, the set of features include two data features. The feature extraction may include computing a value of a frequency word or words using, for example, a Fast Fourier transform (FFT), delta values between two consecutive frequency words, and the like. The feature extraction may further consider the proximity between IoT tags based on co-located group of tags. The two data features can be extracted from such words include the different between two consecutive frequency words received from the same tag and a rate of receiving the frequency words from the same tag.

The extracted features are fed to the correlator 240. In an embodiment, the correlator 240 is configured to implement a semi-supervised machine learning algorithm for analyzing the features and generate, using external supporting data, the trained model 201. Examples for the semi-supervised machine learning algorithm includes, and/or based on, decision trees, label propagation, local outlier factor, isolation forest, and the like.

The supporting data includes labels of certain features previously identified as a specific type of material, information as when the learning dataset was gathered. For example, when the learning dataset was gathered in a room at a certain temperature, an empty room, a crowded room, and the like.

The trained model, generated by the correlator 240, is used to map or correlate unseen frequency words (features) to labels various trace parameters related at least to a state of each vaccine vial 150. The unseen frequency words are analyzed during the detection phase. The trained model 201 can also be trained to label the rate at which the frequency words are received. Such rate can be labeled to indicate a level of material (doses) within the vaccine vial. For example, A high transmission rate frequency words would be labeled as empty vaccine vial, while a low transmission rate frequency words would be labeled as full. The values of “high” and “low” and the values in between are trained.

In an embodiment, a supervised multiclass model can be trained with data that samples the different level of doses of interest. Such a model can be based on extension of existing binary classifiers, such as Naive Bayes, Neural Network, k-nearest-neighbors, decision tress or support vector machine. With this, given new input from an IoT tag, the model can directly detect the level of dose currently seen. A semi-supervised model can be also used in which specific levels, preferably empty and full, are being used to train a regression model, such that given new input from a tag a level of doses (that was not necessarily well defined in advance) can be estimated.

In an embodiment, a semi-supervised model can be further extended to refer to time-line, utilizing the fact that the level of doses is expected to gradually decrease with time. In yet another embodiment, a graphical model or a state machine, for example, can be used to connect properly between states that represent different levels. Detections of other events, like location of vial, proximity of a person, or its temperature, can be also processed with in such framework to improve detection rates. The ability to refer to other events can even facilitate the training of a weaker detector, for which explicit labelling of dose levels is not presented, but weakly inferred from the location and temperature of vial.

According to an embodiment, the trained model 201 is generated when the IoT tags are at rest and in a controlled environment (e.g., a lab). It should be noted that the trained model 201 may be generated and trained based on information received from a plurality of co-located IoT tags.

In an embodiment, during the detection phase, frequency words received from the gateway 120 are aggregated and saved as a detection dataset 250. The aggregation is performed over a predefined detection time window or per tag ID.

The detection dataset 250 is input to the filter 260 that performs the filtration process. The filtered frequency words are fed to a features extractor 270. The feature extraction may include computing a value of frequency word or words using, for example, a FFT, delta values between two consecutive frequency words, statistical information, and the like. As noted above, at least two data features are extracted by the extractor 270.

The set of features (values of frequency words) are fed to the classifier 280. The classifier 280 is configured to label the values of the frequency words to identify the type of material in a vaccine vial. Further, the classifier 280 may be configured to determine the level of the material in the vaccine vial. The labeling is based on the trained model 201 and the set features vector of the respective IoT tag.

The classifier 280 can be implemented using known classifying techniques utilized in supervised or semi-supervised machine learning. For example, the classifier 280 can be implemented using techniques that spread the semi-supervised to supervised range. Examples for such techniques include a K-means, a gaussian mixture model (GMM), a random forest, manifold learning, decision trees, and the like. In an embodiment, the labels provided by the classifier 280 may be fed into the correlator 240 to improve, or otherwise update the trained model 201. In yet another embodiment, the labels provided by the classifier 280 may be fed to the detection dataset for labeling input words.

In an example embodiment, the correlator 240 and/or the classifier 280 can be realized by one or more hardware logic components utilized to process artificial intelligence (AI) tasks, such as a digital signal processor (DSP), a tensor processing unit, or other AI accelerators.

The datasets 210 and 250 can be stored in a memory, which can be volatile (e.g., RAM, etc.) memory, non-volatile (e.g., ROM, flash memory, etc.) memory, or a combination thereof. Alternatively, or collectively, the datasets 210 and 250 can be stored in a storage, or any other medium which can be used to store the desired information.

The machine learning framework 200 discussed herein can be trained to detect other semantic events related to the trace parameters. The events may include proximity to a vaccine vial including an IoT tag, motion of the IoT tag, location of a vaccine vial including an IoT tag, and temperature changes in material in a vaccine vial including the IoT tag, and the temperature within such IoT tag proximity.

FIG. 3 is an example graph of frequency word telemetry corresponding to labeled items in the vaccine vial, according to an embodiment. Various items such as water, air, and vaccine liquid were put in the vaccine vial 150 and placed near one of the IoT tags 110 over time. During the training phase, a learning dataset 210 resulting from the measured disturbances proximate a particular IoT tag 110 with values that reflect the different contents of the vaccine vial 150 are labeled as one of the identified materials based on the method described in FIG. 2.

In an embodiment, the model classifies a vial that is almost empty, that is, condensed vaccine liquid at very low level. Thus, the vial is either with air, or full. That is, the condensed material was diluted with some liquid material.

As an example, a frequency word of approximately 18 dBC/Hz may be labeled as water; frequency word of approximately 15 dBC/Hz may be labeled as air; and frequency word frequency word of about 21 dBC/Hz may be labeled as vaccine liquid during the training process. The dBC unit is decibels relative to the carrier, i.e., the power ratio of a signal to a carrier signal, expressed in decibels. Over time, with repeated collection of frequency words, the trained model 201 may be developed using semi-supervised machine learning described in FIG. 2.

The trained model 201 may be applied during the detection phase to trace various parameters related to the vaccine vials 150 based on data features. The data features include changes in value between two consecutive frequency words, a rate of reception of the frequency words, or both. A trace parameter may be any one of material of the content in the vaccine vial 150, a level of doses in the vial 150, a temperature of a vial 150 or vials 150, a location of vials 150, and so on. It should be noted data features may fed back to the correlator 240 to improve the trained model 201 over time.

FIG. 4 is an example flowchart 400 for a method of training a model for tracing vaccine vials according to an embodiment. At S405, a plurality of frequency words gathered from measurements by the IoT tag 110 and transmitted by the gateway 120 is received. At S410, the plurality of frequency words is filtered based on an established decision boundary, or filter, to retain the frequency words that are outside of the defined boundary, or outliers where interference to the low energy field where the tag is considered to be at rest is detected.

At S420, a sample of the filtered frequency words is obtained, and each of the sample of the filtered frequency words is labeled as a particular material associated with the sample. Once all of the samples of the filtered frequency words are labeled, the method proceeds to S425, where the labeled and filtered frequency words are recombined with the unlabeled and filtered frequency words to create a labeled data set.

At S430, the combined data is further processed, and is averaged and sharpened by a semi-supervised algorithm or a supervised algorithm. At S435, a model for tracing a vial based on the detected frequency word is developed, based on the combined and processed data including the labeled and unlabeled frequency words. In an embodiment, different models can be trained, each of which is designed to trace vials at the different phase of the preparation and/or administration of the vaccine. This can be performed by repeating FIG. 4 for each type of model.

In an embodiment, a first model may include tracing the temperature of vials in a deep freeze station, a second model may include tracing the temperature of vials in a refrigeration station, a third model may include tracing the temperature of vials to be a room temperature at a preparation station, a fourth model may include tracing a level of material (vaccine dose) in each vial at an administration station, and a fifth model may include tracing discarded vials at a discarding station. The various stations are located in different physical locations. The utilization of the different machine learning models at the various location is discussed below.

FIG. 5 is an example flowchart 500 of a method for tracing vaccine vials using low energy sensing according to an embodiment. At S510, packets are received. The packets are sent by IoT tags attached vials and relayed to a server (e.g., server 140) by a gateway (e.g., gateway 120) located at a station. A packet may include a frequency word detected or measured by an IoT tag and an ID associated with the tag. In an embodiment, the packet may be also encapsulated with a gateway identifier (or address), which may be unique for the gateway. Alternatively, such gateway identifier may include a source IP address of the gateway. Each gateway is associated with a specific station location in the vaccination process. In another embodiment, a packet includes a signal strength indicator (RSSI) sent by an IoT tag and encapsulated by the gateway. The RSSI is a measurement of the power present in a received radio signal.

At S515, the frequency words are extracted from the packets and the gateway sending such packets is identified. As each gateway is deployed in a different station, the respective model trained for this model is retrieved. As noted above, a first model may include tracing the temperature of vials in a deep freeze station, a second model may include tracing the temperature of vials in a refrigerate station, a third model may include tracing the temperature of vials to be a room temperature at a preparation station, a fourth model may include tracing a level of material (vaccine dose) each vial at an administration station, and a fifth model may include tracing discard vial at a discard station. The various stations are located in different physical location.

At S520, a respective model, previously trained, is associated with the station from where the packets are received is called.

At S525, each called model is applied to the respective received frequency words. At S530, based on the applied model, the traced parameter that corresponds to the received frequency word is identified. The trace parameter may be determined on the station and the called model. For example, if the fourth model is called, then a dose level in a vial may be determined at an administration station. As another example, a temperature of a vial may also be a trace parameter. The traced temperature may be different at different stations. As described previously with regard to FIG. 4, the model is developed based on a combination of labeled sample of frequency word that is filtered based on an established decision boundary, and an unlabeled frequency word that is filtered based on the established decision boundary described at S425.

At S540, the result of the determined trace parameter is fed back, so the model may be further adjusted and customized to more accurately identify material corresponding to the frequency word detected.

In an embodiment, the traced parameters may be further analyzed to provide semantic events. The semantic events may be sent to the gateway or directly to a monitor at the respective station. The monitor may be the handheld device 170, FIG. 1. For example, a trace parameter from the discard station may be an ID of tag attached to a vial that has been removed, thus such vial should not be used again. Thus, the semantic event may include discarding of the respective vial. As another example, the semantic event may be a temperature derived from the respective traced parameter.

The monitor may be, for example, a handled device (170, FIG. 1) connected to the server through the Internet.

It should be appreciated that, since existing BLE low energy, already widely used for data transmission among devices, is used, according to the disclosed embodiments, for tracing vaccine vials 150, there is no need to use various expensive equipment incompatible with low energy modality to detect temperature and/or contents of such vials 150. Also, since the IoT tag 110 is battery-less, low energy consumption is achieved.

FIG. 6 is an example diagram of the gateway 120 includes a BLE communication card 650 and a network interface card (NIC) 620, the BLE card 650 communicates with the IoT tag 110 over a BLE network (not shown), while the NIC 620 allows communication with the server 140 over the Internet (not shown), or other type of network.

In an embodiment, the gateway 120 may be installed with an agent or application executed by the processor 670 and stored in the memory 680. The agent, when executed, is configured to control the communication with the IoT tag 110 and the server 140.

It should be noted that processor 670 may be realized as one or more hardware logic components and circuits. For example, and without limitation, illustrative types of hardware logic components that can be used include field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), Application-specific standard products (ASSPs), system-on-a-chip systems (SOCs), general-purpose microprocessors, microcontrollers, digital signal processors (DSPs), and the like, or any other hardware logic components that can perform calculations or other manipulations of information.

The memory 680 may be volatile (e.g., RAM, etc.), non-volatile (e.g., ROM, flash memory, etc.), or a combination thereof. The agent (or application) are realized in software. Software shall be construed broadly to mean any type of instructions, whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise. Instructions may include code (e.g., in source code format, binary code format, executable code format, or any other suitable format of code). The instructions, when executed by the processor 670, cause the execution of the agent over the processor 670.

FIG. 7 shows an example schematic diagram of an IoT tag 110, designed according to the disclosed embodiments. The form factor of the IoT tag 110 is an on-die package-less. The IoT tag 110, as schematically demonstrated in FIG. 7, includes an energy harvester 701, coupled to an on-die capacitor 702 and an external passive capacitor 702′, a power management unit (PMU) 703, a microcontroller 704, a system on chip (SoC) 705, and a retention memory 706. The IoT tag 110 further may include at least one antenna 710 glued to a substrate 720, for example. In another embodiment, the antenna 710 may be printed on the substrate or etched to the substrate. In a further embodiment a passive external capacitor may take the place of the antenna 710. In an embodiment, the substrate 720 is made of a low-cost material, such as, but not limited to, polyethylene (PET), polyimide (PI), and polystyrene (PS). In another embodiment, the substrate 720's pattern (layout) can be any of aluminum, copper, or silver. The glue utilized to glue to die and/or antenna 710 may be include materials such as an anisotropic conductive film (ACP), any type of conductive glue, solder past, and the like.

In the embodiment shown in FIG. 7, the antenna 710 is coupled to the harvester 701 and may be utilized for energy harvesting as well as wireless communication. In some embodiments, multiple antennas may be utilized to harvest energy at multiple frequency bands. Other embodiments may include one or more antenna for energy harvesting and an antenna to receive/transmit wireless signals at the BLE frequency band.

The SoC 705 includes a number of execution functions realized as analog circuits, digital circuits, or both. Examples for such execution functions are provided below. The SoC 705 is also configured to carry out processes independently or under the control of the microcontroller 704. Each process carried out by the SoC 705 also has a state, and processes can communicate with other processes through an IPC protocol. In the configuration illustrated in FIG. 7, the SoC 705 and/or the microcontroller 704 loads the context of processes and reads data from the retention memory 706.

The SoC 705 is partitioned into multiple power domains. Each power domain is a collection of gates powered by the same power and ground supply. To reduce the power consumption, only one power domain is turned on during execution. The SoC 705 can perform functions, such as reading from and writing to memory, e.g., of peripherals and can execute simple logic operations; tracking power level of the SoC 705; generating and preparing data packets for transmission; cyclic redundancy check (CRC) code generation; packet whitening; encrypting/decrypting and authentication of packets; converting data from parallel to serial; and staging the packet bits to the analog transmitter path for transmission.

In a preferred embodiment, the SoC 705 includes an oscillator calibration circuit (OCC) 705-A. The OCC 705-A includes at least one frequency locking circuit (FLC), each of which is coupled to an oscillator (both are not shown). The FLC calibrates the frequency of an oscillator using an over-the-air reference signal. In an embodiment, the calibration of the respective oscillator is performed immediately prior to a data transmission session and remains free running during the data transmission session. The FLC can be realized using frequency locked loop (FLL), a phased locked loop (PLL), and a delay locked loop (DLL). An example implementation of an oscillator calibration circuit 705-A is discussed in U.S. Pat. No. 10,886,929 to Yehezkely, assigned to the common assignee.

According to the disclosed embodiments, the energy harvester 701, the capacitor 702, PMU 703, microcontroller 704, SoC 705, and retention memory 706 are integrated in a die 730. The die 730 is glued to the substrate 720. The IoT tag 110 does not include any external DC power source, such as a battery.

In an embodiment, the microcontroller 704 implements electronic circuits (such as, memory, logic, RF, etc.) performing various functions allowing communication using a low energy (power) communication protocol. Examples for such a protocol includes, but are not limited to, Bluetooth®, LoRa, Wi-Gi®, nRF, DECT®, Zigbee®, Z-Wave, EnOcean, and the like. In a preferred embodiment, the microcontroller 704 operates using a Bluetooth Low energy (BLE) communication protocol.

In some embodiments, the microcontroller 704 is integrated with wireless sensors (not shown) to a complete an IoT device functionality.

The harvester 701 is configured to provide multiple voltage levels to the microcontroller 704, while maintaining a low loading DC dissipation value. In an example implementation, the energy harvester 701 may include a voltage multiplier coupled to the antenna 710. The voltage multiplier may be a Dickson multiplier, while the antenna is a 710 receive/transmit antenna of the microcontroller 704. That is, in such a configuration, the antenna is primarily designed to receive and/or transmit wireless signals according to the respective communication protocol of the low-energy IoT tag 110 (e.g., 2.400-2.4835 GHz signal for BLE communication).

It should be noted that the antenna 710 may also be designed for energy harvesting and may operate on a different frequency band, direction, or both, than those defined in the standard of the respective communication protocol. Regardless of the configuration, energy can be harvested from any wireless signals received over the air. Alternatively, energy can be harvested from any other sources, such as solar, piezoelectric signals, and the like.

The harvested energy is stored in the on-die capacitor 702 and/or the external capacitor 702′. The PMU 703 is coupled to the capacitor 702 and is configured to regulate the power to the microcontroller 704 and SoC 705. Specifically, as the capacitance of the capacitor 702 is very limited, the power consumption should be carefully maintained. This maintenance is performed to avoid draining of the capacitor 702, thus resetting the microcontroller 704. The PMU 703 can be realized using a Schmitt trigger that operates on a predefined threshold (Vref), e.g., Vref=0.85V.

In another embodiment, the PMU 703 may be further configured to provide multi-level voltage level indications to the microcontroller 704. Such indications allow the microcontroller 704 to determine the state of a voltage supply at any given moment when the capacitor 702 charges or discharges. According to this embodiment, the PMU 703 may include a detection circuitry controlled by a controller. The detection circuitry includes different voltage reference threshold detectors, where only a subset of such detectors is active at a given time to perform the detection.

The IoT tag 110 does not include any crystal oscillator providing a reference clock signal. According to an embodiment, the reference clock signal is generated using over-the-air signals received from the antenna 710. As noted above, in a typical deployment, a free running oscillator is locked via a Phase-Locked Loop (PLL) to a clock, originating from a crystal oscillator. According to the disclosed embodiments, the OCC 705-A calibrates the frequency of an oscillator using an over-the-air reference signal. The oscillator(s) implemented in the tag 730 are on-die oscillators and may be realized as a digitally controlled oscillator (DCO).

The retention memory 706 is a centralized area that is constantly powered. Data to be retained during low power states is in the retention memory 706. In an embodiment, the retention area is optimized to subthreshold or near threshold voltage, e.g., 0.3V-0.4V. This allows for the reduction of the leakage of the retention cells.

FIG. 8 is an example schematic diagram of a server 140 according to an embodiment. The server 140 includes a processing circuitry 810 coupled to a memory 820, a storage 830, and a network interface 840. In an embodiment, the components of the server 140 may be communicatively connected via a bus 850.

The processing circuitry 810 may be realized as one or more hardware logic components and circuits. For example, and without limitation, illustrative types of hardware logic components that can be used include field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), Application-specific standard products (ASSPs), system-on-a-chip systems (SOCs), general-purpose microprocessors, microcontrollers, digital signal processors (DSPs), and the like, or any other hardware logic components that can perform calculations or other manipulations of information.

The memory 820 may be volatile (e.g., RAM, etc.), non-volatile (e.g., ROM, flash memory, etc.), or a combination thereof. In one configuration, computer readable instructions to implement one or more embodiments disclosed herein may be stored in the storage 830.

In another embodiment, the memory 820 is configured to store software. Software shall be construed broadly to mean any type of instructions, whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise. Instructions may include code (e.g., in source code format, binary code format, executable code format, or any other suitable format of code). The instructions, when executed by the processing circuitry 810, cause the processing circuitry 810 to perform the various processes described herein.

The storage 830 may be magnetic storage, optical storage, and the like, and may be realized, for example, as flash memory or other memory technology, CD-ROM, Digital Versatile Disks (DVDs), or any other medium which can be used to store the desired information.

The network interface 840 allows the server 140 to communicate with the gateways (e.g., gateways 120, FIG. 2) and with the handheld device (e.g., device 170, FIG. 1) for the purpose of, for example, receiving data, sending data, and the like. Further, the network interface 840 allows the server 140 to communicate with the IoT tag 110 for the purpose of collecting frequency words.

It should be understood that the embodiments described herein are not limited to the specific architecture illustrated in FIG. 8, and other architectures may be equally used without departing from the scope of the disclosed embodiments.

The various embodiments disclosed herein can be implemented as hardware, firmware, software, or any combination thereof. Moreover, the software is preferably implemented as an application program tangibly embodied on a program storage unit or computer readable medium consisting of parts, or of certain devices and/or a combination of devices. The application program may be uploaded to, and executed by, a machine comprising any suitable architecture. Preferably, the machine is implemented on a computer platform having hardware such as one or more central processing units (“CPUs”), a memory, and input/output interfaces. The computer platform may also include an operating system and microinstruction code. The various processes and functions described herein may be either part of the microinstruction code or part of the application program, or any combination thereof, which may be executed by a CPU, whether or not such a computer or processor is explicitly shown. In addition, various other peripheral units may be connected to the computer platform such as an additional data storage unit and a printing unit. Furthermore, a non-transitory computer readable medium is any computer readable medium except for a transitory propagating signal.

All examples and conditional language recited herein are intended for pedagogical purposes to aid the reader in understanding the principles of the disclosed embodiment and the concepts contributed by the inventor to furthering the art, and are to be construed as being without limitation to such specifically recited examples and conditions. Moreover, all statements herein reciting principles, aspects, and embodiments of the disclosed embodiments, as well as specific examples thereof, are intended to encompass both structural and functional equivalents thereof. Additionally, it is intended that such equivalents include both currently known equivalents as well as equivalents developed in the future, i.e., any elements developed that perform the same function, regardless of structure.

It should be understood that any reference to an element herein using a designation such as “first,” “second,” and so forth does not generally limit the quantity or order of those elements. Rather, these designations are generally used herein as a convenient method of distinguishing between two or more elements or instances of an element. Thus, a reference to first and second elements does not mean that only two elements may be employed there or that the first element must precede the second element in some manner. Also, unless stated otherwise, a set of elements comprises one or more elements.

As used herein, the phrase “at least one of” followed by a listing of items means that any of the listed items can be utilized individually, or any combination of two or more of the listed items can be utilized. For example, if a system is described as including “at least one of A, B, and C,” the system can include A alone; B alone; C alone; 2A; 2B; 2C; 3A; A and B in combination; B and C in combination; A and C in combination; A, B, and C in combination; 2A and C in combination; A, 3B, and 2C in combination; and the like.

Claims

1. A method for tracing vaccine vials, comprising:

receiving, from a gateway of a plurality of gateways, frequency words from tags attached to vaccine vials, wherein each tag is configured to transmit a plurality of frequency words;
extracting at least one data feature from the plurality of frequency words, wherein each data feature changes in response to a change in a state of a vaccine vial;
classifying the extracted data feature based on a machine learning model trained with respect to a location of the gateway, wherein the classifier is trained to label a trace parameter indicative of a state of a vaccine vial; and
sending a semantic event indicating a value of the trace parameter.

2. The method of claim 1, wherein each data feature is any of: a change between two consecutive frequency words and a rate of reception of frequency words.

3. The method of claim 1, wherein the state of the vaccine vial is any of: a temperature of a vaccine value and a number of doses in the vaccine vial.

4. The method of claim 1, wherein the gateway is located in a deep freeze station, and the first machine learning model is trained to trace temperature of the vaccine vials at deep freeze temperature range.

5. The method of claim 1, wherein the gateway is located in a refrigerate station, and the machine learning model is trained to trace temperature of the vaccine vials at a warming temperature range.

6. The method of claim 1, wherein the gateway is located in a preparation station, and the machine learning model is trained to trace temperature of the vaccine vials at a room temperature range.

7. The method of claim 1, wherein the gateway is located in an administration station, and the machine learning model is trained to trace temperature of a number of doses in each vaccine vial.

8. The method of claim 1, wherein the gateway is located in a discard station, and the machine learning model is trained to trace discard of vaccine vials.

9. The method of claim 1, wherein each frequency word is based on a frequency calibration of the tag.

10. The method of claim 1, wherein a frequency word changes when the tag is out of a calibration.

11. The method of claim 1, wherein each received frequency word is measured as dBC/Hz, wherein the dBC is dBC unit is decibels relative to the carrier.

12. The method of claim 1, further comprising: training a classifier to classify the extracted data feature, wherein training the classifier includes:

receiving a learning dataset of unseen frequency words;
filtering frequency words demonstrating at least abnormal readings;
extracting data features from the filtered frequency words; and
training the machine learning model based on the extracted data features.

13. The method of claim 12, wherein the machine learning model is any one of: a supervised model any an unsupervised model.

14. The method of claim 1, wherein the frequency words are transmitted over a Bluetooth Low Energy (BLE) protocol.

15. The method of claim 1, wherein tag is a battery-less internet of things (IoT) tag in a form factor of a label attached to a vaccine vial.

16. A non-transitory computer readable medium having stored thereon instructions for causing a processing circuitry to execute a process, the process comprising:

A method for tracing vaccine vials, comprising:
receiving, from a gateway of a plurality of gateways, frequency words from tags attached to vaccine vials, wherein each tag is configured to transmit a plurality of frequency words;
extracting at least one data feature from the plurality of frequency words, wherein each data feature changes in response to a change in a state of a vaccine vial;
classifying the extracted data feature based on a machine learning model trained with respect to a location of the gateway, wherein the classifier is trained to label a trace parameter indicative of a state of a vaccine vial; and
sending a semantic event indicating a value of the trace parameter.

17. A system for developing a treatment plan using multi-stage machine learning, comprising:

a processing circuitry; and
a memory, the memory containing instructions that, when executed by the processing circuitry, configure the system to:
receive from a gateway of a plurality of gateways, frequency words from tags attached to vaccine vials, wherein each tag is configured to transmit a plurality of frequency words;
extract at least one data feature from the plurality of frequency words, wherein each data feature changes in response to a change in a state of a vaccine vial;
classify the extracted data feature based on a machine learning model trained with respect to a location of the gateway, wherein the classifier is trained to label a trace parameter indicative of a state of a vaccine vial; and
send a semantic event indicating a value of the trace parameter.

18. The system of claim 17, wherein each data feature is any of: a change between two consecutive frequency words and a rate of reception of frequency words.

19. The system of claim 17, wherein the state of the vaccine vial is any of: a temperature of a vaccine value and a number of doses in the vaccine vial.

20. The system of claim 17, wherein the gateway is located in a deep freeze station, and the first machine learning model is trained to trace temperature of the vaccine vials at deep freeze temperature range.

21. The system of claim 17, wherein the gateway is located in a refrigerate station, and the machine learning model is trained to trace temperature of the vaccine vials at a warming temperature range.

22. The system of claim 17, wherein the gateway is located in a preparation station, and the machine learning model is trained to trace temperature of the vaccine vials at a room temperature range.

23. The system of claim 17, wherein the gateway is located in an administration station, and the machine learning model is trained to trace temperature of a number of doses in each vaccine vial.

24. The system of claim 17, wherein the gateway is located in a discard station, and the machine learning model is trained to trace discard of vaccine vials.

25. The system of claim 17, wherein each frequency word is based on a frequency calibration of the tag.

26. The system of claim 17, wherein a frequency word changes when the tag is out of a calibration.

27. The system of claim 17, wherein each received frequency word is measured as dBC/Hz, wherein the dBC is dBC unit is decibels relative to the carrier.

28. The system of claim 17, wherein the system is further configured to: train a classifier to classify the extracted data feature by:

receiving a learning dataset of unseen frequency words;
filtering frequency words demonstrating at least abnormal readings;
extracting data features from the filtered frequency words; and
training the machine learning model based on the extracted data features.

29. The system of claim 28, wherein the machine learning model is any one of: a supervised model any an unsupervised model.

30. The system of claim 17, wherein the frequency words are transmitted over a Bluetooth Low Energy (BLE) protocol.

31. The system of claim 17, wherein tag is a battery-less internet of things (IoT) tag in a form factor of a label attached to a vaccine vial.

Patent History
Publication number: 20210257074
Type: Application
Filed: Feb 9, 2021
Publication Date: Aug 19, 2021
Applicant: Wiliot, Ltd. (Caesarea)
Inventors: Ido ZELMAN (Hod-Hasharon), Matan EPSTEIN (Tel-Aviv), Shay MOSHE (Pardes Hanna-Karkur), Tsvika RABKIN (Pardes Hanna-Karkur), David LIPSHITZ (Raanana), Tal SZPRUCH (Kfar Saba), Dotan ZIV (Tel-Aviv), Alon YEHEZKELY (Haifa)
Application Number: 17/171,774
Classifications
International Classification: G16H 20/17 (20060101); G06N 20/00 (20060101);