Leak detection system and method for respirators

Abstract

A method for creating a customised personal respiratory mask. The method includes performing an image recognition analysis to determine if the person is wearing a mask; scanning a mask-wearing person with a thermal imaging camera; performing a leak detection analysis; and displaying results of the leak detection analysis.

Description

[0001] LEAK DETECTION SYSTEM AND METHOD FOR RESPIRATORS

[0002] Field of the Invention

[0003] The present description relates to improvements in leak detection and fit analysis of breathing devices designed to protect against airborne particulates and / or biological pathogens.

[0004] Background of the Invention

[0005] Controlling what is mixed with the oxygen we breath, or the way in which that oxygen is delivered during respiration is fundamental for prevention of disease and maintenance of health. For many, breathing natural air taken is for granted, however when that air is contaminated with pathogens or carcinogens, the best-known defence is to create a physical barrier to block the contaminants while allowing smaller air particles pass. An example of this kind of respiratory control device is a simple disposable filtering facepiece mask made from specialised interlocking fabric mesh to capture and block large particles like viruses and carcinogens. Many in the field will understand that a filtering facepiece respirator (FFR), such as a N95 or P2 respirator, performs filtering of larger particles. In many cases, these control devices can be the wearer’s last line of defence against serious health complications including viruses such as C0VID19, influenza, tuberculosis or protection from carcinogens like asbestos, silica dust, smoke, volatile-organic-compounds.

[0006] However different respiratory control devices may first appear, like a simple disposable facemask or a complex electromechanical pressure controlled airway system of CPAP, and they all share a single failure mode that contributes to an overwhelming reduction in device safety and efficacy: the device-to-face interface. The interface between a filtration respirator mask and the wearer’s face contributes to a dramatic step-reduction in filtration efficacy when conformity between the mask and the face is not achieved. The human facial anatomy is a complex three-dimensional geometry that is unique to each individual with only a 0.08% probability of a match with another, and characterised by as many as 52 standardized Cephalometric landmarks. There are clear differences in the cephalometric measurements across ethnically diverse populations, between the sexes and between the ages. Substantial differences are found in the depth and length of the face (SNA and SNB measurement) between different genders and racial groups. These differences are confounded by the combination of skeletal and soft-tissue difference between individuals.

[0007] When considering the overwhelming diversity in human facial anatomy, it is not surprising that “one-(or few)-size-fits-all” design is subject to substantial air leak failures at the device / face interface. Although this one-size-fits-all design is preferred for streamlined manufacturing and cost reduction for buyers, the obscured cost of these benefits, however, is an accepted level of device failure due to inadequate fit.

[0008] So serious is the issue of inadequate fit of respirator masks for example, legislation has been required to protect the health and safety of workers placed in environments with airborne pathogens. These legislative and safety guideline measures are a clear indictment on the design and subsequent efficacy of respiratory control devices to act as the last-line of defence against disease. Therefore, what is needed is a respirator that is custom-designed for a particular wearer.

[0009] The existing art in respirator fit assessment encompasses a range of practices from manual seal checks performed by users to quantitative fit tests utilizing specialized equipment like the PortaCount devices. However, significant challenges persist in ensuring an accurate fit. For example, inaccuracy of self-assessment. User- performed fit checks are prone to errors, with studies showing a failure to detect leaks in a significant number of instances, which could lead to exposure to harmful pathogens. Another challenge is an infrequency of professional fit testing. While more reliable, quantitative fit tests are typically annual, failing to capture daily variations in fit due to changes in facial structure or differences in respirator batches. A further challenge involves technological limitations. Current technologies have not fully harnessed advancements in infrared imaging and machine learning to provide real-time, actionable feedback on the fit of respirators.

[0010] Studies have illustrated the potential of infrared technology, but also highlight limitations such as overfitting, accuracy issues, and methodological constraints due to insufficient data and less sophisticated analysis.

[0011] Summary

[0012] The present description in one preferred aspect provides for a method for creating a customised personal respiratory mask. The method includes sensing presence of a person in proximity to a camera; performing an image recognition analysis to determine if the person is wearing a mask; scanning a mask-wearing person with a thermal imaging camera; performing a leak detection analysis; and displaying results of the leak detection analysis.

[0013] In another preferred aspect, the present disclosure provides a system for creating a customised personal respiratory mask. The system includes a thermal imaging camera; a database of data including spatial relationships between image pixel intensities; and a processor configured to: analyse a digital image taken with the camera using data from the database; and generate a leak detection result based on the analysis. The system further includes a display to display the result generated by the processor.

[0014] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed. In the present specification and claims, the word “comprising” and its derivatives including “comprises” and “comprise” include each of the stated integers, but does not exclude the inclusion of one or more further integers. It will be appreciated that reference herein to “preferred” or “preferably” is intended as exemplary only. The claims as filed and attached with this specification are hereby incorporated by reference into the text of the present description. The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate several embodiments of the invention and together with the description, serve to explain the principles of the invention.

[0015] Brief Description of the Figures

[0016] Fig. 1 is a front perspective view of a mask fit analysis kiosk in accordance with a preferred method of the present disclosure.

[0017] Fig. 2 is a rear partial perspective view of the mask fit analysis kiosk Fig. 1 .

[0018] Detailed Description of the Drawings

[0019] Reference will now be made in detail to the present preferred embodiments of the invention, examples of which are illustrated in the accompanying drawings. Figs. 1 and 2 show a preferred embodiment of a mask fit analysis kiosk system 100 having a housing 102, display screen 104, camara 106, and processing unit 108. The preferred elements of system 100 and their interrelationship are described below.

[0020] Figs. 1 and 2 show housing 102 configured as a user interface that can be formed for easy transport in a variety of environments. The user interface is designed for clarity and ease of use, featuring a display that functions as a 'mirror', providing real-time visual feedback on the fit of the respirator. The interface preferably employs a traffic light system to communicate the results: red for leaks, green for a secure fit, and orange for indeterminate outcomes. Additionally, the interface can be configured to include visual cues that pinpoint the location of any detected leaks, guiding the user to make necessary adjustments for an optimal fit. Interface 102 includes a display screen 104 that may be of a type typically used for portraying digital images, such as encountered with portable tablet computing devices. System 100 includes camera 106, which in a preferred form, is a thermal imaging camera. More preferably, camera 106 is a multi-spectral camera utilized to capture detailed images (including white, but also other spectra) of the respirator as it is worn by the user, detecting variations in signals that indicate potential leaks. System 100 is preferably equipped with an advanced uncooled infrared thermal camera, featuring a 384x288 resolution with 17pm pixel pitch, coupled with a 7mm infrared optical lens optimized for human body temperature measurement. The camera is preferably capable of high-precision thermal imaging necessary for fever detection and, by extension, fit assessment of respirators.

[0021] Supplementing the infrared camera is a webcam 110, which preferably provides HD 720p / 30fps video, beneficial for standard visual feedback and auxiliary imaging. The kiosk may also include a specialized camera with a global shutter sensor technology, effective for depth sensing within a range of 7 cm to 50 cm, ensuring accurate distance measurement from the user to the kiosk. Preferred and desireable camara features include stereoscopic depth technology providing depth accuracy within + / - 2% at a distance of 50 cm; a wide Depth Field of View (FOV) of 87° x 58°; a wide Depth Field of View (FOV) of 87° x 58°; a RGB frame resolution and rate up to 1280 x 720 and 90 fps respectively, ensuring clear, high-resolution imaging; and an operational temperature range from 0 to 35 °C ambient, and 0 to 55 °C for the camera casing, making the kiosk adaptable for various environmental conditions

[0022] Housing 102 further preferably includes a video camera 110, such as a white-light video camera.

[0023] Referring to Fig. 2, system 100 includes processing unit 108. Processing unit 108 is preferably configured with a suite of machine learning algorithms to analyze the images to discern patterns indicative of a good or poor fit. These algorithms are preferably trained on a diverse dataset to ensure high accuracy and reliability.

[0024] Examples of appropriate algorithms include, but are not limited to any one or more of the following: including K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Neural Networks, Decision Trees, and Kernel methods. These algorithms are trained on a dataset of over 25,000 images validated against gold-standard Condensing Nuclei Counter (CNC) results from a Portacount device, ensuring a robust and accurate fit assessment.

[0025] The processing software conducts a spectral gradient analysis along the respirator's fringe to detect changes in texture indicative of leaks. This involves extracting the respirator image at 60 frames per second, identifying the fringe, and performing textural analysis using methods like the grey-level co-occurrence matrix (GLCM). The statistics on the GLCM are then used to quantify pass or fail conditions for the respirator fit.

[0026] The kiosk preferably operates in a continuous monitoring mode, with no defined endpoint for the scanning process, allowing users to engage and disengage at their convenience without the need for cleaning or manual resetting between uses. This enhances the usability of the kiosk in high-traffic areas and maintains high hygiene standards by minimizing touchpoints.

[0027] Having described the preferred components of system 100, a preferred method of use will now be described with reference to Figs. 1 and 2. The operation of the respirator leak detection kiosk is designed to be intuitive, efficient, and user-friendly, involving the following steps. First, users approach the kiosk located in a common area within a healthcare facility. It remains in an always-on state, with cameras actively scanning to detect readiness for use. Users position themselves in front of the kiosk, with their face approximately 30-50cm from the camera, guided by instructions provided on the screen or through supplementary educational materials.

[0028] Next, during an activation stage, the kiosk automatically activates when a user's presence is detected in the correct position. Alternatively, users can initiate the testing process manually. Prior to activation, the system performs a series of safety checks to confirm the user is properly positioned with the respirator correctly donned and ready for testing.

[0029] During a scanning stage, and upon activation, the kiosk displays a digital 'mirror' image of the user's face, possibly supplemented with an avatar. Indicators on the screen guide the user to maintain correct breathing patterns necessary for accurate evaluation. The scanning is triggered either manually or automatically, capturing images via a wide spectrum of camera sensors. These images are processed by the software, which controls image acquisition and stability during the scan. Users are instructed to remain still, with the system providing feedback if image stability is compromised.

[0030] Thereafter, results are displayed on the screen in real-time, preferably utilizing a traffic light system — e.g., red for a detected leak, green for no leak, and orange or yellow for inconclusive results. Visual feedback is a primary mode of communication, with optional audible signals to indicate different stages or results of the testing process. If a leak is detected, the kiosk provides visual cues indicating the potential source of the leak on the respirator.

[0031] To complete the operation, the scanning process does not have a defined endpoint as it preferably offers continuous real-time measurement. Users conclude the fitting by simply walking away. The kiosk's software automatically recognizes the completion of use and resets to a default state, ready for the next user without the need for cleaning or manual resetting, thereby ensuring a seamless transition between users.

[0032] It will be appreciated that the steps described above may be performed in a different order, varied, or certain steps omitted entirely without departing from the scope of the present description. Multiple studies were done for testing and evaluation. Some of these studies are detailed below.

[0033] Study 1 : Pilot Study

[0034] 1 . Study Design: A study was conducted with healthcare workers from a community care organization to collect data on respirator fit using infrared imaging and quantitative fit-testing.

[0035] 2. Participant Selection: Participants were selected based on inclusion criteria relevant to routine respiratory protection programs. The study adhered to the Declaration of Helsinki and received ethics approval.

[0036] 3. Respirator Selection: P2 FFRs from the institution's supply chain were used, selected by an occupational hygienist based on facial feature assessment.

[0037] 4. Fit Testing Procedure: Fit testing was performed using a PortaCount 8048 device, with a pass / fail threshold set at a fit factor of 100.

[0038] 5. Infrared Imaging: Thermal imaging was conducted using a Flir One Gen 3 camera attached to a table device, positioned to capture the respirator on the participant's face.

[0039] 6. Data Processing: Images were pre-processed using Imaged and Matlab, categorizing them into 'pass' and 'fail' based on the fit-test results.

[0040] 7. Machine Learning Analysis: Custom Matlab scripts created semi-automated regions of interest on the images. Features extracted using grey-level cooccurrence matrix analysis were used to train machine learning models to classify the respirators as pass or fail.

[0041] 8. Model Selection and Evaluation: Multiple machine learning models were assessed. The best-performing model was selected through cross-validation and learning curve analysis to ensure accuracy and generalizability.

[0042] 9. Conclusion: The methodology validates the kiosk's algorithm to predict respirator fit, offering a real-time, objective assessment that surpasses traditional self-fit checks. Study 2: IR Mirror Data Study

[0043] 1 . Study Design: Randomized comparison of infrared (IR) group versus standard self fit-check group among university staff and students.

[0044] 2. Participant Selection: 41 participants were randomized, including a diverse demographic profile with four ethnicities represented.

[0045] 3. Respirator Selection: Participants were randomized to test different commonly available FFR styles: flat-fold, tri-panel, or cup style.

[0046] 4. Fit Testing Procedure: Conducted by a qualified occupational hygienist following the Australian Standard for respiratory protective devices, with participants receiving standardized training.

[0047] 5. Infrared Imaging: The IR group utilized an 'infrared mirror' to view thermal gradient changes on the respirator for assessing fit quality and detecting leaks.

[0048] 6. Data Processing: Demographic and anthropometric data were collected, and fit-testing results were analyzed statistically.

[0049] 7. Results: A statistically significant difference between the two groups, with those who had access to the infra-red video ‘mirror’ had substantially higher likelihood of passing the test (IR pass 63.4% vs fit-check pass 34%, Chi- squared p=0.011 ).

[0050] Conclusion: The study confirms the kiosk's algorithm effectively predicts respirator fit, demonstrating a significant improvement in fit-test pass rates for the IR group compared to the self-fit check group.

[0051] Study 3: IR Training Data

[0052] 1 . Study Design: Prospective recruitment with randomization to either an IR group or a standard care (self fit-check) group.

[0053] 2. Participant Selection: Anticipated recruitment of 309 participants through flyers and digital notices. Consent obtained through informed discussion and agreement, overseen by research coordinator. Respirator Selection: Participants receive individual training and are provided with three different respirator styles, each which has size selection model deemed suitable by the occupational hygienist. Fit Testing Procedure: Participants in the IR group engage with an IR 'mirror' to adjust their respirators for optimal fit, followed by a standard care fit-testing procedure including a series of physical exercises. All participants, after adjusting their respirators, underwent a standard care fit-testing procedure which included a series of movements to assess the fit. They were connected to a PortaCount, a condensing nuclei counter, to quantitatively test for respirator leaks by calculating the respiratory fit factor — the ratio of particulates found inside the respirator versus those in the surrounding environment. This quantitative assessment provides an objective measure of the respirator’s efficacy in filtering out environmental particles. Infrared Imaging: Employed in the IR group to make real-time adjustments to respirator fit, aiming to reduce air leaks as visualized by thermal changes on a computer monitor connected to an IR camera. Data Processing: Data collected includes consent documentation, randomization outcome, fit-testing results, and possible retest information if initial fit-testing fails and infra red video (2 mins per participant @60 FPS). Custom computer vision algorithms were developed to identify the respirator fringe (we called these ‘slips’) and to extract the pixel data from the selected region. Data processing was performed on the extracted ‘slips’ to compute GLCM statistics using various parameters for GLCM distance and angle. GLCM stats and portacount binary classification (Pass (Eo>100), or Fail (E0<100)) were loaded into KNN machine learner classifier and hyperparameters tuned to specifications. Model and accompanying weights were exported after training on >5k ‘slip’ images. Model was then re-loaded into Kiosk software and fed data in real-time (<5ms) to classify the respirator as either ‘leaking’ or ‘not leaking’. Accuracy for leak detection achieved: >96%. Another study was conducted as set forth below.

[0054] Study 4

[0055] The methodology for this study is outlined in the following sections. Part 1 describes how infra-red data was collected using imaging data and the images correlated with fit-test results. Part 2 describes the data preparation and processing steps that were required prior to applying the machine learning algorithms.

[0056] Part 1 - Data collection

[0057] Participants

[0058] The study population consisted of healthcare workers from a medium-sized (-1500 employee) community care organisation in suburban Southern Adelaide (Australia).

[0059] The organisation was prepared for COVID-19 outbreaks following guidance of the Australian Government Infection Control Expert Group, including a respiratory protection program. In this respiratory protection program, workers attended a structured clinic where quantitative P2 FFR fit-testing was performed to select and evaluate the appropriate size and style respirator for the individual.

[0060] Participants were invited to join the study at the beginning of their visit to the clinic by the occupational hygienist / nurse. Inclusion criteria included employees required to undertake quantitative fit testing as part of routine respiratory protection program. Exclusion criteria were those who wore beard greater than 2mm.

[0061] The study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the SALHN Human Research Ethics Committee. Participants provided informed consent before enrolment.

[0062] Selection of P2 filtering facepiece respirator

[0063] P2 FFR’s used for this study were taken from the current supply chain within the institution with four different sizes available at the time of fit-testing clinic. All respirators were flat-fold design with ear-loops that comes with a ‘clip’ that is used to tether the loops together on the head for additional fixation support. The use of the ‘clip’ that is provided with these respirators is required by the manufacturer in accordance with their regulatory certification. It must be noted that not all respirators with ear-loops are supplied with this ‘clip’, and this is a particular feature of the respirators used in this study.

[0064] To select the first respirator to test, the occupational hygienist made a visual assessment of the participants facial features (nason-menton length, and bizygomatic width) according with standard process for this role.

[0065] Fit testing

[0066] During the visit to the clinic, fit testing was conducted to assess the adequacy of P2 FFRs in preventing the ingress of microscopic particles like C0VID19 virus by a trained and experienced occupational hygienist. The fit testing clinic is required under that Australian Standard to assess the suitability of a respirator for an individual.

[0067] Quantitative fit-testing was conducted using a PortaCount 8048 device in N95 mode. In N95 mode, the PortaCount measures the concentration of microscopic, aerosolized particles (40nm-60nm) in the ambient air and compares it to the concentration of particles that leak into the FFR through gaps between the face and the FFR while it is worn (rather than through the filter medium of the FFR). For each fit test conducted in this study, the “Modified Ambient Aerosol CNC Quantitative Fit Testing Protocol for Filtering Facepiece Respirators” was selected, whereby participants are required to: bend at the waist as if going to touch toes for 30 seconds, talk out lout slowly and loud enough so as to be heard by a test conductor for 30 seconds, stand in place and turn head side to side for 30 seconds and finally stand in place and move head up and down for 30 seconds. The occupational hygienist then conducted the fit test and recorded the overall result (pass / fail) and the overall fit factor achieved. A fit factor of 100 was used as the pass / fail threshold for the study.

[0068] Infra-red imaging

[0069] Infra-red imaging was performed using commercially available camera built for a table or smartphone. This camera has a thermal resolution of 160x120 pixels, and thermal sensitivity of 60mK. For this particular study, the IR camera was attached to a tablet with the accompanying software app installed.

[0070] The thermal camera was turned on and allowed to self-calibrate prior to any images taken. Using a floor-mounted stand, the tablet and thermal camera was manipulated into position by the hygienist to position it in front of the participant’s face at a distance between 40cm-60cm. Immediately prior to commencing a fit-test with an individual (i.e. , after tubes connected, PortaCount system prepared and the participant had donned their chosen respirator according to instruction), the hygienist used the FLIR ONE app to record an image of the respirator on end- inspiration. Images were saved to the tablet’s internal memory and given a unique identification number for cross-referencing fit-test results on analysis.

[0071] Part 2 - Data preparation and analysis using machine learning (ML) Thermal Image Data Pre-processing

[0072] Native infra-red images were exported from the study tablet to a study PC with software packages with loaded function for processing FLIR images. All FLIR images were imported in native .jpg format and converted to thermal .tiff format for analysis in Matlab. In Matlab, thermal images were separated into separate folders for ‘pass’ and ‘fail’ categories according to the results of the corresponding quantitative fit-test.

[0073] Region of Interest

[0074] A suite of custom Matlab scripts where used to create semi-automated region of interests (ROI’s) within the thermal images around the boundary of the respirator on the nose and cheeks (where majority of leaks occur in P2 FFR’s). To achieve this, images were displayed on the computer screen and an experienced occupational hygienist was instructed to trace around the upper boundary of the respirator to the zygomatic roll-off location, and then close the ROI making an approximately 2cm wide strip across the cheeks and nose. To control the creation of this ROI as far as possible, the ‘AssistedFreehand’ function was used in Matlab to automatically follows edges in the underlying image. As the morphology of each ROI is determined by underlying facial geometry and not standardised, we used a normalization process to map the upper edge of the ROI to a vertical line to normalise the morphology of the images as far as possible.

[0075] Feature Extraction

[0076] For each normalized ROI, image feature extraction was completed using grey-level co-occurrence matrix analysis. In short, a gray-level co-occurrence matrix (GLCM) is a statistical representation of the spatial relationships between pixel intensities in an image. It quantifies the occurrence of pairs of pixel values at specified distances and angles, providing information about texture, patterns, and relationships within an image. In the context of detecting air leak from the P2 FFR, the spatial relationships between pixel intensities (i.e. , thermal gradient) permit the quantitative detection of distinct temperature changes along the respirator's boundary, which could signal a leak. For each ROI, we extracted 8 GLCM features; Contrast, Energy (or Angular Second Moment), Homogeneity, Correlation, Entropy, Dissimilarity, Autocorrelation (ASM), based on their consistent use and proven efficacy in texture pattern variations in prior research. It is worth noting that the GLCM method can generate a larger set of features, yet the study focussed on eight to reduce the issues related to issues relating to dimensionality, reducing potential overfitting and computational costs, and enhancing the interpretability of the model.

[0077] Machine Learning for Leak Detection Classification

[0078] Machine learning (ML) techniques were implemented to classify respirators as either 'passing' or 'failing' based on quantitative fit test result. A comprehensive ML pipeline was developed, encompassing data pre-processing, model selection, training, performance evaluation, and deployment, all within a graphical computing environment (Classification Learner, Matlab).

[0079] Preliminary Assessment for Model Selection

[0080] A wide variety of algorithms were deployed to identify candidates that provide high accuracy in classifying respirator IR image as either 'pass' or 'fail'. The data matrix of 8 GLMC features were divided using an 80% / 20% partition for training and testing. This stratified partitioning guarantees the model's performance evaluation would be conducted on a separate set of data not used during the model training, providing a more objective measure of its predictive capabilities. An array of 22 models were trained and tested including Decision Trees, Support Vector Machines (SVM), Ensemble Methods (such as Bagged Trees, Boosted Trees, Random Forests, and Gentle Boost), Discriminant Analysis techniques (Linear, Quadratic, Regularized), Nearest Neighbours (k-NN), Naive Bayes, Generalized Linear Models (GLM, including Logistic Regression, Poisson Regression, Gaussian Regression), Deep Learning (Neural Networks), Gaussian Process, Hidden Markov Model (HMM), K- Means Clustering, and Self-Organizing Map (SOM). Models were evaluated using cross-validation with k-folds on 80% of the dataset, with 20% of the dataset reserved as an unseen test set. For all 22 models, the validation accuracy and validation total cost was reported in table form.

[0081] Model Sub-selection

[0082] To further evaluate model performance and generalizability, the best performing model were selected from each of model type: Decision Tree, Discriminant, Logistic Regression, Naive Bayes, Support Vector Machines, Nearest Neighbour, Kernel Approximation, Ensemble Classifiers and Neural Network Classifiers.

[0083] Learning curves were used to evaluate the performance of well performing models with the augmented combined with original dataset and divided into training and validation sets using k-fold cross-validation. The training error and validation error were calculated for each iteration, where the size of the training set increased incrementally. The learning curves were plotted to visualize the change in error with increasing training examples. Learning curves were inspected to assess average model error over the folds, providing insights into the model's ability to generalize as the training set size varies.

[0084] Results

[0085] A total of 48 participants (75% female) were recruited with thermal image taken during end inspiration while wearing their professionally selected P2 FFR. There 27 PortaCount failures and 21 passes. All IR images passed visual inspection for suitability (screening for out of focus images, objects obscuring (i.e., hair) the respirator). All participants wore the same style of flat-fold P2 FFR.

[0086] Image processing

[0087] All 48 thermal images were converted to .tiff file format containing only per-pixel temperature values and exported using Imaged for custom processing. Region of interest (ROI) windows were made and exported for all images, with all final ROI data flattened to a rectangular window of 60x 138 pixels. Data augmentation was performed on all rectangular ROI images by flipping along the short axis extending the dataset size to 96 images.

[0088] GLCM features were extracted for each dataset using the following parameters:

[0089] • Offset of: 10x 2 pixels

[0090] • Number of Levels = 50

[0091] Preliminary Assessment

[0092] Twenty two machine learning models were investigated using the full augmented dataset of 96 cases each with 8 GLCM features and classified as ‘pass’ or ‘fail’ according to the quantitative fit test result obtained during fit testing.

[0093] With 100% of the dataset used for model validation with 5 layer k-fold cross- validation, the highest accuracy was found to be with the Ensemble (Bagged Trees) model, with an accuracy of 90.5% at a cost of 17 and the lowest accuracy was found with Ensemble (Boosted Trees) model, with accuracy of 56.8% at a cost of 32. The 6 top performing models were selected for extended analysis of generalizability and accuracy:

[0094] Table 1 : Top 6 performing models all demonstrated an accuracy of greater than 80% with a cost no greater than 13 when the models were trained on 100% of the dataset, and evaluated with 5 levels of k-fold cross verification.

[0095] To investigate the generalisability of the models, training was performed again on these select 6 with data partitioned at 90% for training and 10% as un-seen test data.

[0096] Table 2

[0097] Testing the models on an unseen dataset reduces the likelihood of overfitting to the data. Based on these results, four of the models returned 100% accuracy of predicting if a participant would fail a PortaCount fit test; SVM, KNN, Ensemble Bagged Trees, and Bilayered Neural Network. From these data, the Ensemble RUSBoosted Tree was overfitting in the validation data and resulted in only 55.5% accuracy on the unseen data. Similarly, the Fine Tree model returned only 77.8% accuracy on the unseen data, compared to 83% in the validation. Of note, the total validation cost of both the Ensemble RUSBoosted Tree and the Fine Tree were higher than the other models, at with a cost of 20 and 14 respectively. Of note, the total cost of the test data was highest in these two models, at 4 and 2 respectively, whereas the cost of all other models was 0.

[0098] Discussion

[0099] A thorough investigation was conducted into the utility of using machine learning classification algorithms to predict the binary fit-test result of P2 FFR wearing healthcare workers.

[0100] Key findings of this particular study:

[0101] 1 . Machine learning algorithms are able predict whether a P2 respirator passes or fails a quantitative fit test when using the Detmold D95 Flat-fold respirator.

[0102] 2. Multiple machine learning algorithms return very high accuracy with cross validation for predicting pass or fail quantitative fit test.

[0103] 3. SVM, KNN, Neural Networks and Ensemble Bagged Trees were able to predict fit-test result with 100% accuracy when tested on unseen data.

[0104] Conclusions

[0105] The study successfully employed infra-red (IR) imaging and machine learning to detect leaks in P2 respirators, overcoming the limitations of traditional self fit- checking methods. The techniques used, which focuses on the thermal gradients of the respirator, demonstrated high accuracy, suggesting a novel and more reliable method for leak detection compared to skin gradient measures used by others in the past.

[0106] These findings have substantial implications for enhancing healthcare worker safety, and has promise to provide an objective point-of-use leak detection system for high risk workers. While the study was specific to flat-fold P2 FFRs, these results lay groundwork for broader investigation across various respirator types and populations.

[0107] The foregoing description is by way of example only, and may be varied considerably without departing from the scope of the present description. The features described with respect to one embodiment may be applied to other embodiments, or combined with or interchanged with the features of other embodiments, as appropriate, without departing from the scope of the present invention.

[0108] The kiosk and associated methods of manufacture and use in a preferred form provides the advantages of enhanced accuracy and greater ease of use compared to conventional apparatuses. In a preferred form, the kiosk provides an in situ system capable of analysing mask fit while the user wears the mask, with superior accuracy compared to conventional systems that analyse mask fit separate from the user. Accuracy may be enhanced through thermal analysis of the mask itself, rather than including thermal analysis of areas outside the mask (such as part of a user wearing the mask).

[0109] Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.

Claims

What is claimed is:1 . A method for creating a customised personal respiratory mask, comprising: sensing presence of a person in proximity to a camera; performing an image recognition analysis to determine if the person is wearing a mask; scanning a mask-wearing person with a thermal imaging camera; performing a leak detection analysis; and displaying results of the leak detection analysis.

2. The method of claim 1 , wherein the method is performed at a kiosk so the person can obtain a leak detection analysis in under 10 minutes.

3. The method of either claim 1 or 2, wherein the leak detection analysis includes utilising a machine learning model.

4. The method of claim 3, wherein the machine learning model includes using an ensemble (bagged trees) model.

5. The method of any one of the above claims, wherein the leak detection analysis is confined to a thermal gradient of the mask itself.

6. A system for creating a customised personal respiratory mask, comprising: a thermal imaging camera; a database of data including spatial relationships between image pixel intensities; a processor configured to: analyse a digital image taken with said camera using data from said database; and generate a leak detection result based on the analysis; and a display to display the result generated by said processor.

7. The system of claim 6, wherein said camera is configured for stereoscopic depth detection.

8. The system of either claim 6 or 7, wherein said database includes a gray-level co-occurrence matrix.

9. The system of any one of claims 6 to 8, wherein said processor is configured to analyse the image using a machine learning model.

10. The system of claim 9, wherein the machine learning model includes an ensemble (bagged trees) model.1 1 . The system of either claim 9 or 10, wherein the machine learning model utilises at least 5 features.

12. The system of claim 11 , wherein the features are selected from any one or more of contrast, energy, homogeneity, correlation, entropy, dissimilarity, and autocorrection (ASM).

Images