Technical Field

The present invention relates, in general terms, to systems for monitoring hygiene compliance.

Background

As stipulated by the World Health Organization and enforced by every healthcare institution, the universal standard of proper hand hygiene compliance requires all healthcare workers to sanitize their hands before and after touching any patient. This is because basic hand hygiene (HH) compliance is effective in preventing healthcare associated infections (HAIs) such as COVID- 19. However, HH compliance among healthcare workers (HCW) remains alarmingly low even with widespread education and awareness, with average compliance at 40-50%.

The current COVID-19 pandemic itself is a HAI of phenomenal impact where at least 90,000 healthcare workers have been infected, and more than 200 of them have died because of COVID-19. Even before the COVID-19 pandemic, HAIs have always been a serious problem around the world. In the United States alone, 1 in 25 hospitalized patients are estimated to catch a HAI. HAIs also leads to significant economic, operational, and legal consequences for healthcare institutions.

The current system of auditing HH compliance through direct manual observation is resource intensive, yet inadequate and inconsistent due to the Hawthorne effect. This results in an inefficient deployment of resources that could be better allocated to more pressing needs.

It would be desirable to overcome or [alleviate/ameliorate] at least one of the above-described problems, or at least to provide a useful alternative.

Summary

Disclosed is a system for monitoring hygiene compliance in an environment, comprising: an action detector for detecting a hygiene compromising event involving interaction between a healthcare worker and subject, the action detector being configured to: capture a sequence of images of the environment; and process the sequence of images to detect the hygiene compromising event; and a sanitiser system for confirming a compliance event corresponding to the hygiene compromising event

The system may further comprise a timer for detecting whether the compliance event occurred within a predetermined time period relative to the hygiene compromising event.

As used herein, the term "relative to" when used in relation to the time between the hygiene compromising event and hygiene compliance event, includes the hygiene compromising event occurring either before or after the hygiene compliance event. It can be a requirement that a healthcare worker sanitises their hands before treating a patient or subject, to ensure the patient or subject does not catch infection from the healthcare worker. Similarly, it can be a requirement that a healthcare worker sanitises their hands after treating a patient or subject, to ensure the healthcare worker does not catch infection from the patient or subject does or does not transfer an infection between patients or subjects.

The action detector may detect a hygiene compromising event by: detecting the healthcare worker and subject in a plurality of images in the sequence of images; identifying a subset of the images in which the healthcare worker performs an action indicative of a hygiene compromising event in relation to the subject; and detecting the hygiene compromising event based on the subset of images.

In a clinical setting, a hygiene compromising event will require the healthcare worker to be in proximity of the patient. However, mere proximity of the healthcare worker to the patient does not necessarily mean a hygiene comprising event has occurred. Therefore, some embodiments of the present system detect particular actions performed by the healthcare worker that are indicative of a hygiene compromising event. These actions can include actions by the patient, such as the healthcare worker supporting the patient during performance of an exercise by the patient. Thus, such embodiments detect performance, by the healthcare worker, of an action indicative of a hygiene compromising event in relation to the subject (i.e. patient).

The action detector may identify images in the subset of images by identifying sequential images in which the healthcare worker is performing the action.

The action detector may detect the hygiene compromising event based on a timestamp of an earliest image in the subset of images and a timestamp of a latest image in the subset of images. The action detector may detect the hygiene compromising event based on a time difference between the timestamps.

The system may further comprise a notification device for notifying the healthcare worker of the need to perform the compliance event if the compliance event has not taken place within a reminder period starting from the hygiene compromising event.

The sanitiser system may comprise a hand sanitiser station, the hand sanitiser station confirming a compliance event on detecting dispensing of hand sanitiser.

The action detector may process the images to detect a hygiene compromising event by: detecting an interaction between the healthcare worker and subject; and determining if there is a match between the interaction and a set of hygiene compromising interactions.

The action detector may process the images by applying a machine learning model to the images, the machine learning model being trained to detect hygiene compromising events. The system may further comprise a transceiver for communicating with a remote server, the remote server comprising data for training the machine learning model, the system intermittently updating the machine learning model based on an updated model supplied by the remote server.

Also disclosed is a network for hygiene compliance monitoring comprising: a plurality of systems as described above; and a remote server for: receiving images comprising hygiene compromising events from the plurality of systems; training a machine learning model based on the received images; and transmitting the machine learning model to one or more of the systems.

The remote server may be configured to update the machine learning model and transmit one or both of an updated model and updated weights to the one or more systems, the one or more systems implementing the updated model and/or updated weights in the action detector for processing subsequently captured images. Advantageously, by avoiding the use of wearable devices to monitor the relative locations of a healthcare worker and patient (i.e. subjects), the present invention is generalisable to a greater variety of situations in which there may be multiple healthcare workers and/or patients, different clinical settings, and can be used on-the-fly without requiring a physical device to be allocated to each person in the environment being monitored. This also avoids disruption to clinical workflows.

Advantageously, embodiments of the present system detect hygiene compliance events (i.e. hygiene compromising event) of a particular nature, such as those specified by the World Health Organisation (WHO). By autonomously monitoring such events, the present system assists hosts of clinical environments (clinic owners) in reducing the incidence of disease transmission on their premises.

Brief description of the drawings

Embodiments of the present invention will now be described, by way of nonlimiting example, with reference to the drawings in which:

Figure 1 is a schematic representation of a system in accordance with the present disclosure, and a network comprising multiple such systems;

Figures 2a to 2d show screenshots of an action detector detecting interactions performed in a doctor-patient consultation session;

Figure 3 is a partial exploded view of a sanitiser system in accordance with present teachings;

Figure 4 is an assembled view of the sanitiser system of Figure 3 with a hand sanitiser dispenser therein; Figures 5a to 5d illustrate compliance and non-compliance events; and

Figure 6 shows a notification device mounted to a monitor.

Detailed description

The systems disclosed below enable autonomous monitoring of hygiene compliance events in particular settings such as clinics. The systems include an action detector for detecting hygiene compromising/compliance events and a sanitiser system for identifying hygiene compliance events corresponding to the hygiene comprising events. Some embodiments also include a timer. The timer is used to monitor the period of time between the hygiene compliance event taking place and the time at which a corrective action (e.g. sanitisation event using the sanitiser system) takes place.

The sanitiser system is a sensor component to detect hand hygiene activity. It may be embodied by a soap dispenser at a sink in a clinical setting, a rotary encoder built into a tap to indicate that water has been used to wash hands, a pressor, other devices and any combination thereof. The embodiments below will be described with reference to a pressor. Unless contact dictates otherwise, the pressor may be replaced with another system without departing from the purpose of the devices and systems disclosed herein and such other systems are intended to form part of the present disclosure.

The present systems avoid the need for markers in the field of vision of the action detector. Markers for calibration and depth perception, such as fudicial markers, are not needed. Instead, a machine leaning model of various embodiments determines specific patient-doctor/subject-healthcare worker interactions from an image stream of depth images taken from a fixed mounting point with a top-down perspective. That machine learning model analyses the relative positions of the patient and healthcare worker (HCW) to extract specific patient-HCW interactions, and study hand hygiene compliance. Contact events detected by present systems are occurrences of one or more types of physical interaction between doctor (herein interchangeably used with healthcare worker although the term "healthcare worker" is broader) and patient (herein interchangeably used with subject) that necessitates hand hygiene compliance - i.e. occurrence of a compliance event. As the action detector detects interaction in a plurality of frames/images, in some embodiments a contact event consists of a continuous sequence of frames that evidences a minimum period of time over which an interaction occurred for that interaction to be considered a contact event. This is distinguished from contact events detected by systems based on only a single frame or image that may falsely trigger detection of a contact event when a healthcare worker passes by another person in a clinic without interacting with that person.

Figure 1 illustrates an embodiment of a system 100 capable of detecting contact events and monitoring hygiene compliance in an environment such as a clinical setting. The system 100 includes an action detector 102 and sanitiser system 104. The action detector 102 detects hygiene compromising events, also referred to herein as hygiene compliance events, involving interaction between a healthcare worker and subject. The sanitiser system 104 confirms the occurrence of a compliance event such as hand washing or hand sanitising, corresponding to the hygiene compromising event. In this sense, "corresponding to" refers to a hygiene compliance event occurring before and/or after a hygiene compromising event. It can be a requirement that a healthcare worker sanitise their hands prior to interacting with a patient so as not to infect the patient with a disease carried by the healthcare worker. It can similarly be a requirement that the healthcare worker sanitise their hands post interaction with the patient, to preserve their own health.

The system 100 also includes a time 106. The timer 106 is used for detecting whether the compliance event confirmed by sanitiser system 104 occurred within a predetermined time period relative to the hygiene compromising event detected by the action detector 102. In the event that the compliance event did occur within that predetermined time period, then hygiene compliance is confirmed. If the compliance event occurred outside of the predetermined time period or did not occur, then a compliance failure is confirmed. For illustration purposes, the following description will generally be given with reference to a hygiene compliance event occurring after the hygiene comprising event. Though, it should be understood that the same teachings apply to a hygiene compliance event occurring before the hygiene comprising event.

The system comprises at least one image capturing device 112 that scans or captures images of the environment being monitored. The image capturing device 112 may be part of the action detector or may be a system from which the action detector processes the sequence of images of the environment. In either case, the action detector 102 processes the sequence of images to detect hygiene compromising events. Relevantly, a sequence of images is used to ensure that a compliance event occurred over time rather than determining occurrence of the compliance event on the basis of a single frame which may lead to false detections.

In some embodiments, the image capturing device 112 is a depth sensing camera in which the pixel data corresponds to the distance to the objects in the scanned environment from the camera. In other embodiments, the image capturing device 112 is a stereoscopic camera images from which enable ascertainment or detection of actions performed by a healthcare worker in relation to a subject. In other embodiments, the image capturing device 112 may take the form of a RGB, digital, or IR camera. Moreover, there may be a system of image capture devices 112 located throughout the environment particularly in circumstances where there is a lot of equipment that would create blind spots for a single image capturing device 112.

The image capturing device or devices 112 may be positioned in the environment at any location suitable for capturing interactions between a healthcare worker and subject. In general, it can be advantageous to position image capturing devices 112 above individuals over the area in which healthcare interactions mainly take place. In cases were a single image capture device 112 is employed, it is desirable that the single image capture device 112 has a field of view in which all healthcare interactions take place. In cases where multiple image capturing devices 112 are employed, each may have a partial view of the area in which healthcare interactions take place, the multiple image capturing devices 112 together having a field of vision covering the full portion of the environment in which all healthcare interactions take place.

Where multiple image capture devices are used, the action detector 102 may process images from the plurality of image capturing devices 112 and use a sequence of images from one image capturing device 112 to confirm or disconfirm compliance events detected in a sequence of images taken at the same time from a different image capturing device 112.

The action detector 102 detects hygiene compromising events by detecting a healthcare worker and subject in a plurality of images in the sequence of images. The action detector 102 then identifies a subset of the images in which the healthcare worker performs an action indicative of a hygiene compromising event in relation to the subject - e.g. checking the heartbeat of the subject using a stethoscope, feeling limbs of the subject and so on. Based on that subset of images, the action detector 102 detects whether or not a hygiene compromising event has occurred.

In general, for a compliance events to occur, the healthcare worker must be within the predetermined proximity of the subject. Moreover, the healthcare worker must interact with the subject (i.e. perform an action) while in the predetermined proximity (e.g. 1.5m) of the subject rather than, for example, walking past the subject multiple times. To that end, the action detector 102 may detect sequential images in which the healthcare worker performs an action in relation to the patient, which will typically occur within a predetermined proximity of the subject - e.g. within 1.5m. In this sense, sequential images are images taken in succession where it can be assumed the healthcare worker has insufficient time to separate and return from the subject - i.e. sequential images represent a period of time over which the healthcare worker was interacting with the subject.

Not all interactions are hygiene compromising interactions. Moreover, if the healthcare worker is only in proximity of the subject for a very short period of time, it may be possible that no hygiene compromising event occurred. To ensure that only those events that can constitute a hygiene compromising event are analysed, the action detector 102 may detect hygiene compromising events based on timestamp of the earliest image in the subset of images, or sequential images, and a timestamp of the latest image in that subset or sequence. If the time difference between the earliest and latest images is sufficient, a hygiene compromising event is confirmed or detected.

Some clinical settings do not care to monitor all interactions between healthcare workers and subject. To that end, a hygiene compromising event may only be determined if an interaction between the healthcare worker and subject is one of a specific set of hygiene compromising interactions. For example, interactions detected by the action detector 102 between healthcare worker and subject may be matched or compared to the World Health Organisation's five Moments of Hand Hygiene. If there is a match, then a hygiene compromising event is detected.

The action detector 102 may process images in any suitable manner. In some embodiments, the action detector 102 applies a machine learning model to the images, where the machine learning model has been trained to detect hygiene compromising events/interactions. While it can be desirable that such a machine learning model is trained based on images captured from the particular environment being monitored, so as to account for lighting and other conditions that are particular to that environment, this can be a computationally expensive way of developing a machine learning model and it also means that the machine learning model cannot be implemented until sufficient number of images are captured to make the machine learning model accurate. Instead, the system 100 comprises a transceiver 108. The transceiver communicates with a remote server 110. The remote server 110 has data used for training the machine learning model and the system 100 intermittently or periodically updates the machine learning model based on an updated model supplied by the remote server 110.

In some embodiments, a network 118 is employed that comprises the central processing unit or remote server 110. The network 118 may include a plurality of systems 100 each of which supplies to the remote server 110 images comprising hygiene compromising events. Of course, it is often best to train a machine learning model using positive and negative data so images that do not comprise a hygiene compromising event may also be supplied from the systems 100 to the remote server 110. In either case, the remote server 110 trains the machine learning model based on the images received from the systems 100 and transmits the machine learning model to one or more of the systems 100. The remote server 110 may also update the machine learning model and transmit one or both of an updated model and updated weights for an existing model to the systems 100 so that those systems can implement the updated model and/or updated weights in the respective action detector 102. This enables the action detector 102 to be updated, for example to improve accuracy, for processing subsequently captured images of the respective clinical environment.

In this sense, each system 100 in the network 118 operates as an edge computer (typically a single-board computer), that takes in the data from a continuous sequence of images captured from the image capturing device and performs machine learning algorithms to detect contact events (i.e., specific interactions between the doctor and patient that indicate compulsory hand hygiene to be performed). These interactions can be required to consist of a continuous sequence of images to reduce ambiguity of an actual physical interaction that should occur over a certain period of time. This period can have a minimal value (i.e. a minimum predetermined time period over which continuous interaction must be detected).

Such interactions include (but are not limited to):

(1) The doctor touching the patient while sitting; and

(2) The doctor touching the patient while standing.

Figures 2a to 2d shows instances where a healthcare worker 200 and subject 202 are interacting, where Figure 2a is a frame comprising no physical interaction that will therefore not form part of the sequence of images constituting a hygiene compliance event whereas each of Figures 2b to 2d are instances where there is contact between the healthcare worker and subject that may form part of a sequence of images from which a hygiene compliance event is detected.

Hygiene compliance and compromising events form part of the normal set of interactions expected in a clinical setting. The systems disclosed herein seek to confirm that healthcare workers and others in clinical settings in other environments comply with hygiene requirements. To that end, events detected by the action detector 102 are combined with data from the sanitiser system 104 to detect compliance events. In some instances, an event-based rule is used that determines that a compliance event has occurred if an instance of dispenser usage is detected after the latest known contact event instance, or if a contact event instance comes after an instance of dispenser usage, which constitutes the After Touching and Before Touching events respectively according to 5 Moments of Hand Hygiene stipulated by the World Health Organization.

For present purposes, a compliance event is an act of carrying out a hand hygiene activity such as washing or sanitising hands, either before or after a contact event, by a healthcare worker who is captured in a sequence of images processed by the action detector 102. For every contact event, compliance event for before and after the contact event can, and in many cases should, be accounted for. In addition, while the present system 100 is used for detecting a compliance event for each health compromising or contact event, the lack of a compliance event indicates a non-compliance event - where the healthcare worker does not perform the required hand hygiene step before or after a contact event.

The sanitiser system 104 can include any system for detecting a compliance event such as a hand sanitising event. For example, the sanitiser system 104 may be a hand sanitiser station, the hand sanitiser station confirming a compliance event on detecting dispensing of hand sanitiser. In other embodiments, the sanitiser system 104 is a sensor for detecting usage of a cat at a basin for hand washing.

One embodiment of a sanitiser system is a pressor 300 is shown in Figure 3. The pressor 300 comprises a platform, presently embodied by non-malleable platform 302, that sits on top of a sensor 304. The platform 302 exerts pressure on the sensor 304 whenever pressure is exerted on the platform 302 itself. The sensor 304 thereby detects usage of a hand sanitizer dispenser 400 as shown in Figure 4. By depressing the plunger on top of the dispenser 400, the dispenser 400 exerts pressure on the platform in the base of column 402, the platform in turn exerting pressure on a sensor, the sensor indicating that hand sanitiser has been dispensed. Indications of hygiene compliance events such as the dispensing of hand sanitiser may be transmitted by the sanitiser system 300/404 to a processor of a system such a system 100.

The system 100 may comprise a network of components one of which is a central processing unit for monitoring instances of compliance events measured by sanitiser system 104 relative to events detected by the action detector 102. The network may be a wired network or a wireless network or a combination of different network types. In one embodiment of the system, a wireless communication network is established between the sanitizer system 104 and a central processing unit 114, while image capturing device(s) 112 is/are directly connected to the central processing unit 114 for data transmission.

The central processing unit 114, which may form part of the action detector 102 or any other system component as necessary, may adopt a range of rules for detecting hygiene or compliance events that require one or more inputs from the action detector 102 (e.g. contact events) and/or the sanitiser system 104 (e.g. usage of hand sanitizer). In some cases, the timer 106 enables the system 100 to implement a time-based rule that monitors hygiene compliance by comparing the timestamps of the signals received from the action detector 102 and sanitiser system 104. For example: a compliance event may be detected if a healthcare worker sanitizes his/her hands (1) within x seconds before a contact event and (2) within z seconds after a contact event. To remind the healthcare worker to comply with hygiene requirements, another time variable y can be set where y < z, such that a reminder is activated y seconds after a current of a hygiene compromising/contact event, particularly if no compliance event has occurred within that y seconds. Therefore, z - y should represent a reasonable amount of time for the healthcare worker to sanitize his/her hands so that it can be considered a compliance event after the reminder is activated.

In addition, lack of compliance events (according to the determined rule) should indicate a non-compliance event for the healthcare worker and be captured (for auditing purposes) as well. Figures 5a to 5d illustrate timelines of events for time-based scenarios. Figures 5a and 5b show compliance events taking place before and after a contact, without and with a y-second reminder respectively. In each case, the dark spike is a compliance event and the lighter line indicates an interaction. The time taken for the healthcare worker to sanitize his/her hands (as recorded by the sanitiser system) before and after the contact event are within the acceptable threshold, and each scenario therefore comprises a compliance event. The y-second reminder in Figure 5b is activated as the healthcare worker has not yet sanitized his/her hands within y seconds from the contact event. Figures 5c and 5d show non-compliance events for before or after the contact event as the time taken for the healthcare worker to sanitize his/her hands (as recorded by the Pressor) is not within the threshold x or z respectively. Non-compliance event will also be recorded if the healthcare worker does not even sanitize his/her hands at all.

A reminder may be presented in any recognisable form. To that end, the system 100 of Figure 1 comprises a notification device 116 for notifying the healthcare worker of the need to perform the compliance event if the compliance event has not taken place within a reminder period starting from the hygiene compromising event - i.e. y-seconds. The notification device 116 may comprise a speaker for emitting an audible alarm, a visible indicator such as a light emitting diode or other device for display notifications to remind the healthcare worker of the need to sanitise.

The notification device 116 activates an alert to remind the doctor, in real-time, to sanitize his or her hands if they are required to perform hand hygiene and have not yet complied according to the time-based rules. In one such embodiment, the reminder system uses a visual alert mechanism utilizing a LED that blinks, placed in a location that is easy for the healthcare worker to notice. For example, the notification device may be an alert mechanism 600 mounted on the doctor's desktop monitor 602 (Figure 6).

The system 100 uses a computer vision-based machine learning model to intelligently recognize actions of people in the image stream. That model can detect, with a very high probability, contact events belonging to the WHO 5 Moments of Hand Hygiene (particularly before and after contact with patients) which undoubtedly requires the healthcare worker to perform hand hygiene.

Embodiments of the present system 100 integrate peripheral sensors to detect hand hygiene compliance. To that end, the system 100 deviates from non- systems that, for example, only monitor hand hygiene compliance when a healthcare worker enters or exits the environment rather than in relation to health compromising events. Those peripheral sensors are sensors of the sanitiser system 104 that can include one or more sensors each of which may be a pressor such as that depicted in Figure 4 for determining the use of a hand sanitiser dispenser, a tap or faucet usage sensor or other device. Thus the system 100 determines when a hygiene compliance event should occur and also detect whether that event has subsequently occurred.

An advantage of the system 100 is that it employs computer vision without requiring healthcare workers or patients to wear a sensor, tag or other wearable device and avoids the need for the fiducial markers in the field of view. As such, disruption to the normal clinical workflows is largely avoided.

The system 100 is intended to reduce the infection rate for both patients and healthcare workers. By recording incidences of hygiene compromising events and compliance events, the system 100 or the network 118 may be able to compare the rate of compliance with the rate of infection for multiple sites to determine the efficacy of hygiene compliance requirements. Improving hygiene compliance results in a reduced burden on healthcare systems and administrative processes, thereby reducing cost.

In addition, contact tracing can be implemented by identifying the healthcare worker and subject in various images and subsequently determining interactions between those individuals and other individuals.

It will be appreciated that many further modifications and permutations of various aspects of the described embodiments are possible. Accordingly, the described aspects are intended to embrace all such alterations, modifications, and variations that fall within the spirit and scope of the appended claims. Throughout this specification and the claims which follow, unless the context requires otherwise, the word "comprise", and variations such as "comprises" and "comprising", will be understood to imply the inclusion of a stated integer or step or group of integers or steps but not the exclusion of any other integer or step or group of integers or steps.

The reference in this specification to any prior publication (or information derived from it), or to any matter which is known, is not, and should not be taken as an acknowledgment or admission or any form of suggestion that that prior publication (or information derived from it) or known matter forms part of the common general knowledge in the field of endeavour to which this specification relates.