WO2024062096A1

WO2024062096A1 - Device for detecting an anomaly in the wearing of a respiratory mask

Info

Publication number: WO2024062096A1
Application number: PCT/EP2023/076216
Authority: WO
Inventors: Thomas Similowski; Capucine MORELOT-PANZINI; Philippe ROGUEDAS
Original assignee: Oso-Ai; Sorbonne Universite; Iinstitut National De La Sante Et De La Recherche Medicale; Assistance Publique Hopitaux De Paris
Priority date: 2022-09-22
Filing date: 2023-09-22
Publication date: 2024-03-28
Also published as: FR3139994A1

Abstract

A method for detecting an anomaly in the wearing of a respiratory mask fitted to the face of a user, comprising: - a) detection, by a microphone, of a sound produced, at different instants, by an expiration or inspiration of the user through the mask; - b) processing of the detected sound, by a processing unit, at each instant so as to extract therefrom at least one characteristic of the sound; - c) as a function of each characteristic extracted during step b), detection, by the processing unit, of an occurrence of an anomaly in the wearing of the mask.

Description

DEVICE FOR DETECTING AN ANOMALY WHEN WEARING A RESPIRATORY MASK

TECHNICAL AREA

The technical field of the invention is the monitoring of users wearing a respiratory mask.

PRIOR ART

The use of a respiratory mask is common in the treatment of respiratory pathologies, for example chronic obstructive pulmonary disease, obesity-hypoventilation syndrome, so-called “restrictive” pulmonary conditions, or sleep apnea.

A possible function of such masks is to connect the patient wearing it to a respiratory assistance machine delivering excess pressure to the respiratory system by covering the nose, or even the nose and mouth. This makes it possible, for example, to carry out treatment by non-invasive ventilation, which consists of providing mechanical assistance to breathing by supplying the mask with pressurized air cyclically. This helps to reduce the work of the respiratory muscles and improve gas exchange. In another type of treatment, the air is pressurized in a non-cyclical manner, this modality, called "continuous positive pressure", can be used to treat sleep apnea.

The effectiveness of the treatment is subject to correct application of the mask to the face, particularly during phases during which the user's alertness is insufficient, for example during sleep. Depending on the treatment, the mask used may be a face mask, covering the mouth and nose, or a nasal mask, covering only the nose.

Documents WO2021243293, WO2010091462, WO2020118311 and WO2021245637 describe respiratory masks intended to treat sleep apnea. In WO2021243293, the detection of a sound is mentioned, for leak detection purposes, in a spectral band below 1000 Hz.

We understand that it is desirable to have an effective, discreet and easy to use device, making it possible to monitor that a respiratory mask is correctly used. The invention meets this need.

STATEMENT OF THE INVENTION A first object of the invention is a method for detecting an anomaly in the wearing of a respiratory mask applied against the face of a user, comprising:

- a) detection, by a microphone, of a sound produced, at different times, by an exhalation or inspiration of the user through the mask;

- b) processing of the sound detected by a processing unit, at each moment so as to extract at least one characteristic;

- c) depending on each characteristic extracted during step b), detection, by the processing unit, of the occurrence of an anomaly in wearing the mask.

Step c) may include:

- cl) taking into account at least one previously memorized criterion;

- c2) comparison of each characteristic extracted during step b) with the criterion or one of the criteria taken into account during sub-step cl).

Treatment may include:

- selection of at least one frequency band of the sound;

- determination of a characteristic in the selected frequency band.

According to one possibility:

- the characteristic is a spectral power;

- the criterion is a spectral power threshold, so that the fault is detected when the spectral power, in the selected frequency band, exceeds the spectral power threshold.

According to one possibility:

- the characteristic is a variation in spectral power;

- the criterion is a spectral power variation threshold, so that the fault is detected when the spectral power variation, in the selected frequency band, exceeds the spectral power variation threshold.

According to one possibility:

- the criterion is a threshold;

- sub-step cl) includes:

• taking into account times at which the characteristic crosses the threshold;

• estimation of a temporal period of crossing the threshold, according to which the characteristic crosses the threshold; during substep c2), the anomaly is detected when the time period of crossing the threshold is within a predetermined range. According to one possibility:

- the mask is a face mask, configured to cover the mouth and nose of the user;

- the frequency band is greater than 100 Hz.

According to one possibility

- the mask is a nasal mask, configured to cover the user's nose without covering the mouth;

- the frequency band is between 300 Hz and 13000 Hz.

According to one possibility, the frequency band extends from 1000 Hz, or from 1100 Hz, up to 10000 Hz or 13000 Hz.

According to one possibility:

- the processing unit implements a supervised learning artificial intelligence algorithm;

- the supervised learning artificial intelligence algorithm is configured by a learning phase taking into account sounds detected in the absence and presence of an anomaly;

A second object is a device for monitoring the wearing of a respiratory mask by a user, the device comprising:

- a microphone, configured to record sounds of the user's breathing through the mask;

- a processing unit, programmed to receive the sounds recorded by the microphone, and to implement steps b) and c) of a method according to the first object of the invention.

The invention will be better understood on reading the presentation of the exemplary embodiments presented, in the remainder of the description, in connection with the figures listed below.

FIGURES

Figure IA represents an example of a device according to the invention.

Figure IB shows the wearing of a face mask and possible air outlets.

Figure 2 represents a frequency decomposition of sounds produced by a facial mask worn by a user.

Figure 3A represents a nasal mask. Figure 3B shows the wearing of a nasal mask.

Figure 4 represents a frequency decomposition of sounds produced by a nasal mask worn by a user.

Figure 5 schematizes the implementation steps of a device according to the invention.

PRESENTATION OF SPECIAL MODES OF REALIZATION

Figure IA represents an example of a device according to the invention. The device comprises a microphone 10, connected to a processing unit 11. The microphone is configured to record sounds produced by a user wearing a face mask 20.

Figure IB shows the face mask 20 worn by a user.

The processing unit 11 is programmed to carry out processing of the sounds collected by the microphone 10. More precisely, these are sounds produced by the breathing of the user through the mask 20, during his inspiration and/or his expiry.

The inventors observed that when the mask is not correctly applied against the face, the sound resulting from the user's breathing, through the mask, varies. Thus, an analysis of the sound recorded by the microphone makes it possible to detect a possible defect in wearing the mask.

The respiratory mask comprises a bubble 21, generally flexible and transparent, delimited by a seal 22, for example a silicone type seal or foam. The seal is intended to be applied against the user's skin. When the mask is worn correctly, the seal forms a waterproof barrier, or considered as such, preventing or limiting air circulation between the internal space, delimited by the bubble and the user's face, and the ambient air. , outside the bubble.

Note that there is also an intentional calibrated air outlet through the mask allowing the user to exhale freely.

However, the mask can be moved from its correct position for use, causing air to pass through either side of the seal. Such a leak is not desirable, because it compromises the possibility of establishing the desired pressure regime inside the bubble, which results in a reduction in the effectiveness of the treatment.

The processing unit 11 is configured to: process the sounds recorded by the microphone 10, so as to extract characteristics therefrom; compare the extracted characteristics with a criterion, the criterion being representative of a defect in wearing a mask; based on the comparison, determine the occurrence of an anomaly in wearing the mask.

The criterion representing the defect in wearing a mask is previously determined during a learning phase. It may be: a spectral power threshold, in a predetermined frequency band; a threshold for variation of the spectral power in a predetermined frequency band; an intensity threshold or an intensity variation threshold.

The criterion can be a passage of a periodic threshold, reflecting the fact that the characteristic studied exceeds the threshold periodically, according to a period compatible with a respiratory period, typically of the order of a few seconds or a few tens of seconds.

According to one possibility, the sound processing is carried out by an artificial intelligence algorithm with supervised learning, as described below in connection with Figure 5. The criterion is therefore “implicit”, in the sense that it is taken into account in the parameterization of the algorithm. The output of the algorithm can be detection or non-detection of an anomaly.

Figure 2 represents a spectrogram of sounds recorded, in the audible range, during a test during which a user used a facial mask correctly during a first time interval Ati. An anomaly was then deliberately introduced, inducing an air leak, particularly during exhalations. The anomaly was maintained during a second time interval At ₂ .

The x-axis corresponds to time. The y axis corresponds to frequency (Hz). The gray level corresponds to the spectral power.

In Figure 2, the periodicity of breathing is observed by an alternation between light bands (high spectral power), which correspond to expirations, and dark bands (low spectral power), which correspond to inspirations. The occurrence of an anomaly results in a periodic increase in spectral power, particularly during expirations. The spectral power is increased in frequencies above 100 Hz.

According to this embodiment, a criterion forming a spectral power threshold can be established, for one or more frequencies greater than 100 Hz. When at least one of said frequencies, the measured spectral power exceeds the threshold, a fault is detected. The method is more robust by simultaneously considering different spectral bands.

According to one possibility, the processing unit is configured to detect the moments during which at least one characteristic extracted from the measured sound satisfies the criterion representative of the anomaly. The processing unit can estimate a time period during which the characteristic satisfies the criterion. When the time period is within a predetermined range, likely to correspond to a respiratory rhythm, the processing unit detects an anomaly. Taking the time period into account makes it possible to limit the occurrence of false detections. The predetermined range can be between 2s (breathing rate of 30 cycles per minute) to 10s (breathing rate of 6 cycles per minute).

Figure 3A shows an example of a nasal mask 20 connected to a device according to the invention. The nasal mask includes a bubble 21 covering only the nose and leaving the user's mouth free.

Figure 3B shows a nasal mask worn by a user. When using this type of mask, the mouth should be kept closed, otherwise a leak will occur through the mouth. An anomaly when wearing a mask may correspond to an opening of the mouth.

Figure 4 represents a spectrogram of sounds recorded during a test during which a user used a nasal mask correctly during a first time interval Ati. A first anomaly was then voluntarily introduced, during a second time interval At ₂ , corresponding to a slight opening of the mouth. A second anomaly was then deliberately introduced, during a third Ata time interval, corresponding to a greater opening of the mouth.

Figure 4 is similar to Figure 2: the x-axis corresponds to time, the y-axis corresponds to frequency and the gray level corresponds to spectral power.

As in Figure 2, the intensity of the sound is modulated according to the respiratory rhythm, the spectral power during the expiration phases being greater than the spectral power during the inspiration phases. We observe that the anomalies result in a variation of the spectrogram, and more precisely an increase in the spectral power in a frequency range between 300 Hz and 10,000 Hz. When the anomaly becomes more significant, we observe a marked increase in the spectral power between 1000 Hz and 8000 Hz. We observe that the increase in spectral power concerns both the expiration and inspiration phases. In Figure 4, the box drawn in white during the Ata period delimits a frequency band of interest.

Figure 5 shows the main stages of sound processing implemented by the processing unit. during a step 100, the processing unit 11 receives the sound detected by the microphone; during a step 110, the processing unit 11 extracts one or more characteristics of the detected sound; during a step 120, the processing unit 11 takes into account a criterion, previously determined during a learning phase 90. The learning phase consists of carrying out tests for a type of mask, with and without anomaly, so as to determine the criterion(s) corresponding to an anomaly. during a step 130, the processing unit 11 detects the occurrence of an anomaly when at least one extracted characteristic, or when each extracted characteristic, corresponds to an anomaly occurrence criterion. If an anomaly is detected, an alarm signal can be generated, so as to alert the user or a user monitoring team.

According to one possibility, the criterion for the occurrence of an anomaly is determined by implementing an artificial intelligence algorithm with supervised learning, for example a neural network. The algorithm is fed either by sound or by previously extracted sound characteristics.

The output of the algorithm is the occurrence, or not, of an anomaly. The algorithm is configured during the learning phase, taking into account sounds recorded respectively in the presence and absence of defects. According to such an embodiment, the criterion is implicitly taken into account in the algorithm. Steps 110 and 120 are merged into a single step 110/120, which corresponds to the implementation of the algorithm.

The spectrograms shown in Figures 2 and 4 show that a leak results in a modification of the spectrum between 100 Hz and 13000 Hz. Between 100 Hz and 1000 Hz, certain fluctuations are detected in the absence of leaks. See for example the fluctuations observed during the Ati periods in Figures 2 and 4. Such fluctuations may correspond to everyday noises, corresponding to an activity carried out by the user, not linked to wearing the mask. It is therefore preferable for the spectral band to extend from a wavelength greater than 1000 Hz, for example beyond 1100 Hz, or 1200 Hz, or 1300 Hz, or 1400 Hz, or 1500 Hz, or 2000 Hz, and below 13000 Hz. Being above 1000 Hz or 1100 Hz makes it possible to limit the impact, on the detected sound, of user activities.

The invention can be implemented for monitoring users, both in a medical environment and at home. It does not require bulky or expensive equipment, which makes it particularly suitable for use at home.

Claims

CLAIMS Device for monitoring the wearing of a respiratory mask by a user, the device comprising:

- a microphone (10), configured to record sounds of the user's exhalation or inspiration through the mask;

- a processing unit (11), programmed to

- a) receive the sounds recorded by the microphone, at different times;

- b) process the detected sound, at each moment so as to extract at least one characteristic;

- c) depending on each characteristic extracted during step b), detect the occurrence of an anomaly in wearing the mask. Device according to claim 1, in which the processing unit is programmed so that step c) comprises:

- cl) taking into account at least one previously memorized criterion;

- c2) comparison of each characteristic extracted during step b) with the criterion with one of the criteria taken into account during sub-step cl). Device according to claim 2, in which the processing unit is programmed to:

- select at least one frequency band of the sound;

- determine a characteristic in the selected frequency band. Device according to claim 3, in which:

- the characteristic is a spectral power;

- the criterion is a spectral power threshold, so that the fault is detected when the spectral power, in the selected frequency band, exceeds the spectral power threshold. Device according to claim 3, in which:

- the characteristic is a variation in spectral power;

- the criterion is a spectral power variation threshold, so that the fault is detected when the spectral power variation, in the selected frequency band, exceeds the spectral power variation threshold. Device according to any one of claims 2 to 5, in which:

- the criterion is a threshold;

- sub-step cl) includes:

• estimation of a temporal period of crossing the threshold, according to which the characteristic crosses the threshold;

- during substep c2), the anomaly is detected when the time period of crossing the threshold is within a predetermined range. Device according to any one of claims 3 to 6, in which:

- the mask is a face mask, configured to cover the mouth and nose of the user;

- the frequency band is greater than 100 Hz. Device according to any one of claims 3 to 6, in which:

- the frequency band is between 300 Hz and 13000 Hz. Device according to any one of the preceding claims, in which:

- the supervised learning artificial intelligence algorithm is configured by a learning phase taking into account sounds detected in the absence and presence of an anomaly.