ES2769451T3 - System to monitor the use of a device - Google Patents

Abstract

A system for monitoring the user technique of an inhaler device configured for the administration of a medicament, said system comprising: a microphone adapted to detect the sound produced during the operation of an inhaler device; processing circuitry operable to process a data signal obtained from the microphone, wherein said data signal comprises sensed acoustic information, the processing circuitry is adapted to determine the characteristics of inhalation and exhalation breathing that occur during use, analyzing the temporal and spectral components of the detected and processed acoustic information to differentiate between an inhalation and an exhalation, as a function of the temporal and spectral components, characterized in that the processing circuitry further comprises means for identifying a priming or blistering characteristic of the drug device to identify that a drug is about to be administered prior to an inhalation by analyzing an acoustic signal generated by a lever mechanism of the inhaler device, adapted to release the drug that generates a unique energy profile of the mechanism. or lever related to drug release noise.

Description

DESCRIPTION

System to monitor the use of a device

Field of the Invention

The invention relates to a monitoring device for monitoring the use of a device.

In particular, the invention relates to controlling the correct use of devices for the administration of drugs and inhalers.

Background of the Invention

Most chronic respiratory and nasal diseases, as well as some systemic diseases, are treated with inhaled medications. However, one problem with administering medication by inhalers is that patients often do not take their medications as directed. Non-adherence to medication regimens refers to loss of medication dose, incorrect technique when using the inhaler, and potentially over-medication. Non-adherence is a major problem rooted in a lack of understanding of the medication and misunderstanding of the instructions for use. For inhaled medication, adherence involves both using the inhaler at the correct time of day (temporary adherence) and the correct manner (adherence of the technique). Non-adherence rates only among asthma patients range from 30% to 70%. Detecting when and how an inhaler is used by analyzing acoustic recordings of inhaler use can provide evidence of patients' adherence to their inhaled medication regimen. Manually listening to acoustic recordings of inhaler use to assess adherence is a tedious and time-consuming process, and therefore a device or system that can perform this task automatically and accurately would be of great value.

There are several challenges associated with analyzing the patient's inhaler technique using acoustics in uncontrolled settings. Often, the patient or the background environment in which the patient uses their inhaler generates large amounts of artifacts. In many cases it has been found that patients unintentionally exhale into the mouthpiece of an inhaler, dispersing some or even all of the medication. There are many systems that try to use acoustic information to analyze flow rates in terms of inhaler performance.

One of these systems has been disclosed by British Patent Publication Number GB 2395437, assigned to Profile Respiratory Systems, which discloses a system that uses sensors to identify sounds in device operation and is primarily associated with flow research in terms of performance of the inhaler, however, the system does not effectively identify errors in the operation of the inhaler.

Another system is disclosed by PCT patent publication number WO2011 / 135353, assigned to Sagentia Limited, which discloses a drug delivery device that includes sensors and processing circuits that can detect operational events, such as flow rates and drug delivery. , in various types of inhalers, such as dry powder inhalers, metered dose inhalers, nasal inhalers, and nebulizers. The information determined by the processing circuits can be used to provide feedback to the user or can be stored or transmitted for later analysis. However, a problem with this approach is that no errors are identified in the technique. After manually sorting through several hundred audio recordings of inhaler use by a wide range of people, it is clear that there are a large number and variety of possible user technique errors. The Sagentia product identifies whether the maximum inhalation flow profile is achieved, as indicated by the manufacturers. However, if the patient has exhaled on the inhaler prior to inhalation, the dose of available medication will be reduced, which could result in a missed dose. Exhales in the device can disperse the medication and introduce a source of moisture into the inhaler. The hydrotropic nature of the drug means that it is likely to combine with water vapor and agglomerate in the device. This can lead to overdosing on subsequent inhalations. Other user errors include not holding the breath for the required duration after an inhalation and inappropriate inhalations, such as short repeated inhalations instead of a long continuous inhalation.

Other systems in the art include WO2011 / 083377, assigned to Philips Electronics; EP 0 387 222, Draco AB; and US2011 / 290240, Meyer Adam et al., which identify a technical error, namely the flow rate to achieve the correct flow profile, however, these systems inherently suffer from the same problems described above. The article "Segmentation of expiratory and inspiratory sounds in baby cry audio recordings using hidden Markov models" by Kentaro Katahira et al. describes an application of machine learning techniques to identify the exhalation and inspiration phases of the audio recording of human baby cries.

Therefore, it is desirable to provide an automatic detection system and procedure that is capable of identifying the full protocol of the drug, or the lack thereof, when using an inhaler. This includes recording the number of medication doses taken, the timing of these doses, and adherence to the prescribed use protocol for that inhaler.

It is an object of the invention to provide a device that can record relevant acoustic information on inhaler use and one that provides valuable information on patients' adherence to their medication.

Summary of the invention

In accordance with the invention, there is provided, as set forth in the appended claims, a suitable system for controlling the user technique of an inhaler device configured for the administration of a medicament, said system comprising:

a microphone adapted to detect the sound produced during the operation of the inhaler device; processing circuitry operable to process a data signal obtained from the microphone, wherein said data signal comprises detected acoustic information, the processing circuitry is adapted to determine the breathing characteristics by inhalation and exhalation that occur during use, analyzing the components and spectral parameters of the acoustic information detected and processed to differentiate between an inhalation and an exhalation, based on the temporal and spectral components.

The invention is defined by the attached claim 1. There are several unique aspects of the invention: the use of acoustic signals to interpret how the inhaler is used by combining timing information and quantitative measures of the effectiveness of the inhaler technique. It will be appreciated that in the context of the present invention the invention can be used to analyze information from any type of inhaler or respiratory device. Other devices use mechanical methods to assess inhalation power without recording inhaler use. Nor do they provide a record of how well a patient was using their inhaler during certain periods of time to inform the practitioner of consistent problems. The data processing function of the invention can be performed on the inhaler device or a separate processing device, such as a separate personal computing device, or shared between devices depending on how the data processing function is configured. Inhalation records, associated with inhaler use, can be used to directly infer the patient's lung function and response to treatment. In particular, changes in inhalation volume throughout treatment are related to the acoustic characteristics of inhalation, including acoustic energy, maximum and median amplitude, increasing amplitude time, duration of inhalation. The exhalation characteristics, recorded when the patient exhales, are also related to the degree of limitation of air flow. Additionally, based on direct evidence, the device can inform the user and others about the patient's health through information on lung capacity and air flow, clinical changes in response to medications. Finally, the device can provide information on various aspects of cognition, including executive cognitive function.

In accordance with the invention, the processing circuitry further comprises means for identifying the priming or blistering characteristics of the drug to identify that a drug is about to be administered prior to inhalation.

According to the invention, the priming characteristics of the drug comprise an acoustic signal generated by a mechanism, for example a lever, adapted to release the drug which generates a unique energy profile of the lever movement involved with the blistering noise in a capsule to release the medicine.

In one embodiment, analysis of the temporal components provides an indicator of temporal adherence indicating that the correct volume of medication is administered at the correct temporal intervals. In one embodiment, the system is adapted to be reconfigured for different temporary adherence requirements depending on the device and medication to be used.

In one embodiment, means are provided to identify multiple inhalations; or an exhalation before inhalation after blistering of the capsule.

In one embodiment, means are provided to identify an insufficient inhalation volume to deliver the medication or deliver the correct volume of the medication after blistering of the capsule.

In one embodiment, the processing circuitry comprises means for identifying different frequency components to differentiate between inhalations and exhalations.

In one embodiment, the device comprises means for comparing a decomposition of singular values of the short-term power spectrum of the signal with a predefined threshold to identify possible respiratory sounds. The zero crossing rate of breath sounds is compared to a predefined threshold to confirm them as breath sounds using such comparison means.

In one embodiment, the signal is processed to determine the frequency domain components of the identified breath sounds and the means to classify the detected sounds as inhalation or exhalation.

In one embodiment, the processing circuitry comprises means for tracking the inhaler usage time of the device for adhesion analysis.

In one embodiment, means are provided to calculate the median amplitude and duration of an identified inhalation to determine the amount of drug released from the device.

In one embodiment, means are provided for calculating the amplitude of an identified exhalation to determine if the medication has dispersed from the inhaler, before inhalation has occurred. In one embodiment, means are provided for calculating the minimum energy to disperse the medication that has been achieved on exhalation.

In one embodiment, means are provided to calculate the duration of any respiratory sound. In one example, means are provided to record and store a time stamp indicating when the device is used.

In one example, the device comprises a gyroscope adapted to identify that the inhaler device is held in the correct plane in use. The gyroscope ensures that the inhaler is in the correct orientation when used.

In an MDI inhaler, part of the protocol is to shake the device beforehand and this can be detected for proper adherence to the protocol by an accelerometer. In another example, the accelerometer can be used to detect unwanted shaking of the device, for example, in a DPI device, shaking of the device after it has been primed is undesirable as the drug will disperse.

In one example, the device comprises a thermistor adapted to identify the differentiation between inhalations and exhalations.

In one example, the device comprises means for recording after closure of the inhaler to identify exhalations after inhalation.

In one embodiment, a second adapted microphone is provided to determine the inhalation or exhalation characteristics.

In one embodiment, the device comprises memory means adapted to store information for later use.

In one embodiment, the device comprises a second microphone in line with the first microphone to differentiate between exhalations and inhalations.

In one example, means are provided for uploading data from the device to a separate computing device using one or more of a physical memory card; a cable, for example, using a USB connection; or a wireless connection, for example, using a Bluetooth protocol.

In another embodiment of the invention, an algorithm is provided to process the audio downloaded from the device. The algorithm processes the audio to confirm that the inhaler has been used correctly. It does this by identifying events that determine whether the inhaler was used correctly or not, for example, inhalations should be differentiated from exhalations to ensure that the user has inhaled the drug. The movement of the lever must be identified to ensure that the drug has been released. The algorithm not only identifies events, but also identifies the order in which they occur, as this has a direct effect on whether the inhaler event is successful or not. For example, an exhalation before the blister forms does not affect drug delivery, but an exhalation after the blister of medication has formed can profoundly affect the delivery of the drug to the lungs of the patient. patient. The ability to identify exhalations after mandatory breath hold after inhalations is also unique in this invention. Recordings of exhalation acoustics not only confirm that the patient has held their breath for a sufficient period of time to allow for the correct delivery of the medication, but also allows for acoustic analysis that relates the caliber of the airways, such as the presence of wheezing signatures and the duration of exhalation. The current system can perform post-processing of data offline and presents information about timing errors and technical errors on a visual display that can be easily read by a doctor or patient. This information will be used to educate the patient on the proper way to use their inhaler.

Non-adherence to medication and errors in inhaler use are the problems addressed by this invention. The timestamp of audio files provides a record of inhaler usage, allowing the system to identify when the inhaler is being over / underused. The identification of events such as inhalation and exhalation allows the identification of technical errors in the use of the inhaler by the subject. These errors can be equivalent to the missed doses.

The invention keeps a record of each performance of the inhaler and can report statistics on medication adherence in terms of timing and errors. A doctor can observe any constant mistakes that the patient is making and intends to educate them on the correct use of this device. In another example, an inhaler device suitable for administering a medication is provided, said device comprising:

a body with a mouthpiece and a microphone adapted to detect the sound produced during the operation of the device;

operable processing circuits for processing a data signal obtained from the microphone, wherein said data signal comprises detected acoustic information; and

the processing circuits are adapted to determine the inhalation and exhalation characteristics that occur during use.

In another example, an implemented computer system is provided to monitor the user's technique of an inhaler device configured for the administration of a medicine, said system comprising: a module or means to process a data signal obtained from a microphone, in which said data signal comprises detected acoustic information, characterized by

a module or means for determining the respiratory characteristics of inhalation and exhalation that occur during the use of the inhaler, by analyzing the temporal and spectral components of the acoustic information detected; and

a module or means for differentiating between an inhalation and an exhalation based on the temporal and spectral components.

In another example, a method is provided for controlling the user technique of an inhaler device configured for administration of a medicament, said method comprising:

processing a data signal obtained from a microphone, in which said data signal comprises detected acoustic information, characterized by the steps of:

determine the respiratory characteristics of inhalation and exhalation that occur during use, analyzing the temporal and spectral components of the acoustic information detected; and

differentiate between an inhalation and an exhalation based on the temporal and spectral components.

Also provided is a computer program comprising program instructions for causing a computer program to perform the above procedure that can be performed on a recording medium, carrier signal, or read-only memory.

Brief description of the drawings

The invention will be more clearly understood from the following description of an embodiment thereof, given by way of example only, with reference to the accompanying drawings, in which:

Figure 1 illustrates an example inhaler device;

Figure 2 illustrates a flow chart illustrating the steps of the processing circuit to identify respiratory sounds;

Figure 3 illustrates the relationship between acoustic energy and the amount of drug dispersed as a result of the puff using the "old" version of the device;

Figure 4 illustrates the relationship between acoustic energy and the amount of drug dispersed as a result of the puff using the "new" version of the device;

Figure 5 illustrates the relationship between acoustic energy and inhalation flow rate;

Figure 6 illustrates the relationship between acoustic energy and the amount of drug withdrawn from the inhaler as a result of inhalations;

Figure 7 illustrates an experimental setup to examine the relationship between inhalation energy and drug released from the inhaler;

Figure 8 illustrates on the left an adherence control device and on the right an adherence control device attached to the side of a Seretide Accuhaler / Diskus inhaler according to one embodiment; Figure 9 illustrates an acoustic signal from a typical inhaler register with an inhalation present between 3.5 to 5.5 s;

Figure 10 illustrates the identification and detection of temporary start / end of an inhalation (indicated by arrows) in a typical inhaler record by the algorithm;

Figure 11 shows the presence of blisters, exhalation and inhalation events in (a) time domain and (b) signal spectrogram illustrating the ideal use of an inhaler;

Figure 12 illustrates a block diagram of the basic steps of the algorithm to analyze the inhaler registers according to an embodiment of the invention;

Figure 13 illustrates a flow chart of the steps used to detect blister events and how they can be displayed;

Figure 14 illustrates a flow chart of the processing steps of the algorithm used to detect inhalations and exhalations in the audio files;

Figure 15 illustrates how the algorithm checks what events have occurred, the frequency of such events, and the order in which these events have occurred;

Figures 16 and 17 illustrate typical results of the algorithm that can be used to evaluate or show the temporary adherence of patients;

Figure 18 illustrates a block diagram of the components for configuring the inhaler device according to an embodiment of the invention; and

Figures 19 and 20 illustrate a sample network architecture to facilitate the implementation of the invention.

Detailed description of the drawings

In an embodiment of the present invention, Figure 1 illustrates an inhaler device, for example, a Diskus inhaler. The inhaler comprises a system consisting of a microphone, a microprocessor, a battery, and a memory medium or card (not shown). The microprocessor activates and begins recording every time the inhaler is opened. It will be appreciated that the system of the present invention can be performed on an inhaler device or adapted to an existing inhaler device. The data can be uploaded to a remote location for subsequent post-processing that will be described in more detail below. Referring to the flow chart in Figure 2, the signal is processed to evaluate the operation of the inhaler device by determining the existence and order of certain breathing characteristics that can occur during use of the inhaler. A decomposition of unique values of the signal's short-term power spectrum (Mel Frequency Cepstral coefficients (MFCCs)) is compared with a predefined threshold to identify possible respiratory sounds. The zero crossing rate of these breath sounds is also compared to a predefined threshold to confirm them as breath sounds.

The signal is processed to determine the frequency domain components of the identified breath sounds to classify them as an inhalation or an exhalation. A ratio of the high and low frequency bands is used to classify each respiratory sound as an inhalation or an exhalation.

The signal is processed to identify drug release by identifying the unique energy profile of lever movement related to the blistering noise in the drug capsule. The median amplitude and duration of the identified inhalation is calculated to determine the amount of drug released by the inhaler. The amplitude of the exhalation is also identified to determine if the drug has dispersed from the inhaler, before inhalation occurs.

The algorithm then uses this information to determine whether a successful, partial, or unsuccessful inhalation of the drug has been performed. An inhaler event is classified as successful if lever movement is identified and if there is sufficient inhalation without prior exhalation. A partial classification is given to an inhaler event if the mean inhalation amplitude is below a threshold and / or sufficient exhalation occurs before inhalation and after lever release. A failed classification will refer to an inhaler event where no inhalation has been identified.

Sample operation

A Knowles Acoustics SPM0204HE5 microphone was used in the design as it has low power consumption (250 ^{| j} A), high sensitivity (42dB ± 4dB), low signal to noise ratio (59dB) and is physically very small (4.72mm) L x 3.76mm W x 1.25mm H). It will be appreciated that any suitable microphone can be used. A microSD card, a NAND flash with a serial interface, is used due to its relatively low power consumption (15 mA), small form factor and removable nature.

A 1/10 D cell lithium-thionyl chloride (Li-SOCl2) battery was chosen as this chemistry allows for high current consumption relative to the physical size of the battery, while providing sufficient capacity to last one month of operation and up to 24 months of device waiting time before use.

The device can be turned on by flipping a switch and the microprocessor receives power, the firmware starts running, and the device wakes up. The device initially sets the internal and external clocks. Second, the memory card can be initialized, this confirms the correct operation of the memory card, reads the initial value of the real time clock (RTC) and sets the boot sector on the card in which the next recording. The microcontroller has a 16-bit £ -A Analog to Digital Converter (ADC) with an internal reference. At startup, the firmware must configure the ADC. The ADC receives a clock signal from the digitally controlled internal oscillator of the microcontroller, which is calibrated at 8MHz. A 4-stroke clock splitter reduces this speed to 2MHz, this combined with a 256 oversampling rate produces 7874 samples per second, roughly equivalent to the 8kHz target capture resolution originally specified for the device. Although the ADC has a resolution of 16 bits, half of these are discarded and only the most significant 8 bits are registered. This provides a data rate of just under 64 kBits per second and audio quality comparable to an ISDN (digital landline) phone call, which is more than enough for analytics purposes.

When the user subsequently opens the inhaler device, a magnetic reed switch is activated and an interrupt request is generated, this activates the device from low power mode and activates a function, which starts a new recording. Before a new recording can start, the device performs three integrity checks, firstly, it makes sure the device isn't recording yet, secondly, it checks that the RTC is working properly, and thirdly, it checks that the card works correctly. The microprocessor continues recording until the inhaler is closed and the recording is saved on the memory card. The device then goes into low power mode to conserve battery power.

When the device is in low power mode, the internal clocks stop to save energy. During this time, an external 32,768 Hz crystal oscillator maintains the time. Once every second (every 32,768 cycles), a watchdog timer triggers an interrupt in which an integer representing the time elapsed since the device was powered is incremented by one. Every time a recording starts, this time stamp is stored as part of the recording's metadata. During a recording, a watchdog interrupt function increases a timer variable that monitors the duration of the recording, if the duration of the recording exceeds a preset duration (currently 120 seconds), the recording will stop regardless of the position of the switch to conserve battery power. This timer ensures that if an inhaler is accidentally opened or left open after use, it will not continue recording indefinitely depleting the limited battery capacity.

When the patient returns the device, it is essential that the data can be easily read from the device. The SD card can be removed from the device, placed in a card reader, and downloaded to a desktop PC. Alternatively, the data can be downloaded with a direct cable to a USB. From here it is uploaded to a web server via an encrypted link. The device file metadata is interpreted to calculate the times the inhaler was used.

An important aspect of the invention is the positioning of the hardware and firmware in the design of the device. In a first aspect, the orientation of the microphone was changed downwards; This improved the fidelity of the audio recordings. Typically, the microphone location is up inside the device and results in recording a significant amount of ambient noise. A second aspect of the invention was the replacement of the removable SD card to a solid state memory module. This reduced the processor noise involved in writing data to memory and improved the reliability of the memory module. The recording protocol has also changed slightly, recording continues for 10 seconds after closing the inhaler to record exhalation. Finally, the way data is extracted from the device changed; data is removed from the device through direct memory access and USB connection.

The algorithm processes the recorded audio signal of each inhaler use to identify a selection of events that may be related to the inhaler technique. The events to identify are: inhalations, exhalations and lever movement.

The signal is filtered to remove all frequency elements of approximately 1300Hz.

The signal is segmented into 700ms frames.

Overlap frame starts 20 ms after previous frame

The adaptive threshold of the singular vector was set at 14% higher than the lowest calculated value. The zero crossing speed threshold is set to 0.002. This was found to reduce false> ossives through trial and error.

Breathing events are identified from this process.

The mean amplitude of these breath samples is extracted.

Spectral density measurements are used to differentiate between inhalations and exhalations.

Acoustic threshold values are used to assess the probability of the breathing phase when releasing the drug from the inhaler.

A decision tree is used to classify the general event of the inhaler as Pass, Failure or Incomplete. The output file is a text file that contains information about the presence of inhaler events, when they occur relative to others, and a classification of them in terms of the known thresholds for each type of event.

• Output file changed to user friendly presentation and easier data import in Microsoft Excel.

After respiratory sounds have been identified from the above, the algorithm is set to find "puffs" in the inhaler. Exhales or "puffs" are identified by examining frequencies between 20-200 Hz. from observations that exhalations have a characteristic pattern in this frequency band. The signal power spectral density is calculated and subsequently converted to decibels. If the average signal power is greater than a fixed threshold of 65dB, then it is labeled as an exhalation. The fixed threshold was calculated from empirical observations.

The same procedure is used to identify blister sounds. However, this movement of the Diskus inhaler lever causes a distinctive pattern at high frequencies. The blisters can be identified by calculating the power spectral density in the 2000-3000 Hz frequency band. A fixed threshold of 77 dB was established, again established by empirical observations, with any section of the signal above this threshold classified as a blister event. An additional restriction was established in the sense that a blister can only occur after 0.5 seconds of the Diskus inhaler, any possible blister event that occurs before this time is considered false positive.

From the identification of events and their synchronization in relation to other events, we can identify the following technical errors

Blow into the device after lever movement and before inhalation

Multiple inhalations less than 5 seconds apart

No lever movement

Double lever movement

No inhalation or very weak

Exhaling within five seconds after inhaling

If any of these events are identified, the algorithm will indicate that an error has occurred.

Finally, a traffic light system can be shown that has a light that represents each event of the inhaler and the color of the light represents whether it was correct in time and technique, if it was incorrect in time or OR technique, or if it was incorrect in time and technique. . This information can highlight any constant errors that occur in the use of the inhaler by a patient.

It will be appreciated that the system hardware and software can be modified to accommodate various inhalers and medication protocols according to your application.

Experimental data

The relationship between acoustic energy and the amount of drug dispersed in a Seretide Diskus DPI when exhalation occurs in the mouthpiece direction is now described with respect to the experimental data. Blowing into the mouthpiece of a dry powder inhaler (DPI) such as the Diskus is presumed to reduce the amount of medication available for inhalation, if the ampule containing the medication has been pre-punctured. Patients who inhale reduced amounts of prescription medication are at risk of not receiving the expected full therapeutic effects. Investigating how the intensity of blowing into an inhaler can affect the amount of drug available to inhale can provide empirical evidence on what level of exhalation is considered harmful. Patients using DPI are advised to exhale normally but not to or near the DPI (particularly the mouthpiece). Exhaling in a DPI can create two problems. Firstly, the drug dose can be lost. Second, exhaling at a DPI can introduce moisture, which can cause the medication to clump or stick to the side of the mouthpiece. Exhalation to FRC (functional residual capacity) is considered sufficient before inhalation, as previous research has found that forced expiration is not more effective and is also likely to cause broncho spasm (i.e. cough). The use of an inadequate technique with DPI reduces the efficacy of the drugs (bronchodilators, etc.) that are administered. A critical mistake for all DPI is to cause the puff on the device before inhalation. However, it should be noted that exhaling in a DPI would be more important with bulk deposit devices such as the Turbuhaler, but less so for the Diskus where the unit doses are sealed until priming (blistering).

In a previous study by Melani et al., Patients were found to exhale through a DPI nozzle in up to 22% of cases. Being able to detect these exhalations through the use of an algorithm will allow doctors to investigate whether patients are using their DPI correctly. The severity of the exhalation is important. If the drug has been released into the mouthpiece of a DPI, the exhalation power required to disperse the drug should be quantified. Knowing how powerful an exhalation must be before adversely dispersing the drug from the mouthpiece allows establishing an effective threshold for detecting such events using an algorithm.

One aspect of the invention is to know the relationship between the intensity of an exhalation at the mouthpiece of a DPI and the amount of drug dispersed.

A Seretide Diskus DPI inhaler was used for experimental purposes. The murmurs were classified as either hard or soft in nature. The following simple procedure was performed for this experiment:

1. The inhaler was first weighed before placing it directly in front of the subject's mouth.

2. The subject then proceeded to exhale or blow directly into the mouthpiece.

3. The inhaler was subsequently weighed once more to see the amount of drug that was dispersed.

4. Each time the inhaler was used, the device recorded the acoustics of the puff.

5. Steps 1 through 3 were repeated for hard and soft puffs.

Figures 3 and 4 illustrate the relationship between acoustic energy and the amount of drug dispersed from the Diskus DPI inhaler. Exhales or puffs were classified as either hard or soft in nature. Hard puffs were exhalations to residual volume, whereas soft puffs were exhalations to functional residual capacity.

The Diskus DPI inhalers used in this experiment contained the drug Seretide. This contained 50 mcg of salmeterol and 500 mcg of fluticasone propionate per dose, plus 12.5 mg of lactose (vehicle) in each dose. Overall, this was equivalent to 13.05 mg of powder per dose. Other variations of Seretide can be used that contain a reduced amount of fluticasone propionate (250 mcg or 100 mcg).

After each individual puff, the Diskus DPI inhaler was weighed on a scale to calculate the amount of Seretide drug dispersed. Acoustic energy was calculated using an algorithm designed to detect exhalations. The algorithm searches for energy bands between 20Hz and 200Hz. Acoustic energy is calculated for each section by multiplying the square of the average amplitude of each exhalation by its duration in seconds. The relationship between acoustic energy and dispersed drug is depicted in Figure 3 and Figure 4, respectively.

Figure 3 contains data obtained from the use of the "old" device, while Figure 4 contains data from the "new" device. Two different versions of the device were tested to investigate whether the findings of this experiment would be different.

Examining Figure 3, it can be seen that there is a clear difference in acoustic energy between hard blows and soft blows, as expected. However, in terms of the amount of drug dispersed, there were no significant differences between strong and soft murmurs. The results are shown in figure 4.

It will be appreciated that the energy of a puff has the potential to affect the amount of drug dispersed. However, from this experiment, it is clearly appreciated that regardless of the intensity of the puff, the Seretide drug will disperse from the mouthpiece of the Diskus inhaler.

There are some cases where the dispersed drug weighs more than the amount of drug available in the vial (13.05 mg). The reasons for this may lie in the fact that for each puff, a small amount of the drug will remain in the mouthpiece. Moisture introduced by exhalation has the potential to cause the powder to agglomerate and adhere to the side of the mouthpiece. While one puff may not displace this powder, a second puff may do so, which explains the abnormally high dispersion of the drug.

Another observation to keep in mind is that the acoustic energy of the murmurs is significantly higher in the new device compared to the previous device. This may be due to the fact that the orientation of the microphone was changed from the top face to the bottom face.

The following conclusions can be deduced from this experiment:

• Acoustic energy differs between hard and soft puffs for the Seretide Diskus DPI used in this experiment.

• The type of murmur does not affect the amount of drug dispersed, since a relatively soft murmur has the potential to disperse the same amount of drug as a strong murmur.

• The only real way to prevent drug from escaping from the nozzle is not to blow directly into the nozzle.

• Puffs are unfavorable events and a robust algorithm is required that can detect them, regardless of how powerful they are. The question remains whether a murmur that occurs in the vicinity of a device is classified as harmful, even if the drug is not dispersed. More trials are required to investigate this.

Experiment 2

This experiment is concerned with obtaining information regarding the inspiratory flow / acoustic energy required to remove the drug from the DPI during inhalations. This relationship is important in understanding how patients actually use their inhalers. Previous studies have indicated that for inhaler medication to be clinically effective, it must have a minimum flow rate of 30 l / min. While these studies have focused primarily on the effectiveness of drugs at different flow rates, this experiment will simply aim to discover what acoustic energies remove how much drug from the inhaler. Figure 5 illustrates the relationship between acoustic energy and flow rate for inhalations. Figure 6 illustrates the relationship between acoustic energy and the amount of drug withdrawn from the inhaler as a result of inhalations.

Although being able to accurately identify inhalations is an important step in monitoring adherence, being able to extract additional pertinent information from each inhalation and inform the doctor about the effectiveness of each inhalation is a much bigger step.

This experiment will seek to investigate and validate such relationships by looking at the flow rates, associated acoustic characteristics, and drug mass removed from a Diskus DPI inhaler.

This experiment involved the use of an air vacuum to remove the drug from the Diskus inhaler. A rotameter was used to measure the vacuum flow rate in liters per minute. A specially designed attachment accessory was used to hold the Diskus inhaler in place, as shown in Figure 7.

The experiment was designed to ideally create a realistic flow at the required level. However, simply connecting the device to the inhaler, while switched on, would give uncertain results. This is because the vacuum flow would drop dramatically from a high flow once connected to the DPI inhaler due to a change in resistance. A simple on / off valve was used to connect the vacuum and aspirate the powder from the mouthpiece of the inhaler. This simulated the ramp of an inhalation since the flow would start at a low rate (0 l / min) and increase at a maximum rate of flow. The following sections will give details of the experimental protocol followed and the results obtained in addition to a discussion and conclusion of the experimental results. Figure 7 illustrates the equipment configuration required to carry out these experiments. The protocol for this experiment was:

• A dummy Diskus inhaler was used to establish the required flow rate. When the required flow rate was reached, the on / off valve was closed and the dummy Diskus inhaler was replaced with the Test Diskus.

• The Diskus Test Inhaler was weighed prior to blistering and was subsequently placed in the Attachment to hold it in place. The on / off valve was opened for a set period of time depending on the required inhalation duration. The times used in this experiment were 0.5s, 1s, 2s and 3s respectively.

• The test Diskus inhaler was removed after each test and weighed, measuring the mass of the drug removed during sham inhalation.

• The Diskus Test Inhaler was cleaned after each test or every four tests, depending on the test being performed.

• Flow rates were adjusted using the dummy Diskus inhaler and results were obtained for different inhalation times for flow rates ranging from 10 l / min to 100 l / min.

Regarding the flow rate with respect to the drug withdrawn, it can be seen that the amount of drug withdrawn from the inhaler is drastically reduced below 30 l / min. This finding is in agreement with the previous literature that suggests that 30 l / min is the minimum flow for an inhalation to be clinically effective. For this test, the inhaler was only cleaned every four runs, to mimic actual use of the inhaler by a patient.

For the next test the inhaler was cleaned after each run, it can be seen that even at flow rates greater than 30 l / min the full weight of drug is not removed. This may have implications for the clinical efficacy of inhalers, since the required dose is not reached.

Acoustic amplitude has an approximate linear relationship to flow. In addition, acoustic energy has a relation of power with the flow. These findings may help to extract relevant information on inhalations. If an algorithm was used to detect inhalations from asthma patients using inhalers, then investigating the amplitude and energy of the inhalations would provide clinicians with information about the effectiveness of each inhalation relative to the amount of drug drawn from the inhaler.

Figure 6 shows the DPI inhaler data that is output after every 4 tests. The tests ran from 3 seconds to 0.5 seconds and then the DPI inhaler was cleaned. The importance of duration at low acoustic energy values is particularly evident here. An interesting result is that the mass increase of the drug eliminated does not always occur during the duration of the tests. It can be seen in figure 6 that the tests carried out for 1 second extracted more drug than the test carried out for 2 seconds in 5 cases. This shows that a user can take more drug in some doses than others. This finding suggests that it may lead to excessive amounts of drug being given unintentionally, as it may be stuck in the inhaler but added to a new dose giving a higher dose than required.

It will be appreciated that duration is an important factor in determining the amount of drug withdrawn from the inhaler during inhalations. The ability to determine the amount of drug released by the inhaler by combining measures of duration and acoustic amplitude will increase the power of the system to assess the quality of the phases of breathing during the use of the inhaler, that is, it will improve the identification of technical error. This experiment provides minimal acoustic thresholds to assess the quality of inhalation.

Experiment 3

Twenty asthma patients (11 women and 9 men) attending an outpatient respiratory clinic were recruited for this study. The age range was from 20 to 68 (mean 43.5 ± standard deviation 14.2). All subjects had previously been prescribed Seretide inhalers and were very familiar with the mechanics of using such inhalers. Each of the subjects received a Seretide Accuhaler / Diskus inhaler with the adherence control device attached and were instructed to use the dry powder inhaler as they would normally do at a clinical visit. Figure 8 illustrates on the left an adherence control device and on the right an adherence control device attached to the side of a Seretide Accuhaler / Diskus inhaler according to one embodiment. The addition of the adherence monitoring device had no impact on the normal operation of the inhaler. Each time the inhaler is opened, the adhesion device turns on and records the acoustic signal until the inhaler is closed later. Subject's inhaler use was recorded over a three-month period, and each subject returned to the clinic at monthly intervals to upload their records of the inhaler to a database.

Adhesion monitoring device

This study uses the Seretide Accuhaler / 'Diskus "inhaler together with an attached adherence monitoring device (manufactured by Vitalograph (Ireland) Limited). The adherence device consists of a microphone, microcontroller, battery and micro SD card The microphone is a medium quality Knowles Acoustics SPM0204HE5 Microphone The adhesion device was securely attached to one side of the Diskus inhaler, as seen in Figure 8, allowing patient use of the inhaler to register No problem.

The adhesion device is activated, that is, it begins to record, the first time the disc inhaler is opened. Each time the patient uses the inhaler, an audio file of the event is recorded and saved as a mono wav file, sampled at 7913 samples / second with a bit depth of 8 bits / sample, on the memory card. The adhesion device enters sleep mode, to conserve energy, when the inhaler is closed.

The acoustic signal of a typical patient recording is shown in Figure 9. The correct procedure for using the Seretide Diskus inhaler involves first sliding the device to expose the mouthpiece (t = 0s), sliding a lever that releases a dose of medication at the mouthpiece (t = 1s), inhale (t = 3.5 -5.5s), hold your breath for about 10 seconds (t = 5.5-15s) and finally slide the device to close it.

Signal processing

The algorithm used to identify and detect the start / end of inhalations can be divided into two different sections. The first section involves identifying and demarcating inhalation type events on recordings, while the second stage involves eliminating false positives, that is, false inhalation classifications.

Obtaining Cepstral Mel Frequency Coefficients (MFCC) is a parameterization procedure for vocalization, due to the fact that MFCC models the known variation of critical bandwidth of human ears with frequency. Respiratory sounds are known to have a characteristic pattern that allows them to be distinguished from other sounds. Based on this observation, an algorithm was designed to detect this pattern.

The algorithm first went through a training procedure on a set of 20 randomly selected inhaler recordings. Each signal was separated into 700ms long frames that overlap every 20ms. 12 MFCC were calculated for each frame in the signal, forming a short-term signal cepstrogram. Using Singular Value Decomposition (SVD), a normalized singular vector was calculated from the signal cepstrogram. Singular vectors can be used to capture the most important characteristics of respiratory sounds obtained from MFCC calculations. An adaptive threshold is automatically set that is 14% higher than the lowest singular vector in the inhaler recording. Singular vectors above the adaptive threshold were marked as possible inhalation events, while those below were discarded. This adaptive threshold was found empirically to produce the most accurate detection of events and, subsequently, inhalations in the training series. In the second stage of the algorithm, the zero crossing rate (ZCR) and the mean amplitude were calculated to reduce the number of false positives detected by the algorithm, that is, artifacts classified as inhalations. Inhalations were empirically found to have a characteristically high ZCR compared to non-inhalations in the training series. Therefore, a fixed threshold constant of 0.17 was introduced to reflect this fact. In the training series, inhalations always had a ZCR above this threshold value, while false positives were successfully removed.

The mean amplitude of the proposed inhalation event was also calculated. Similar to the ZCR threshold, a fixed threshold was introduced to eliminate false positives based on empirical observations from the training series. Inhalations were found to have a mean amplitude threshold value greater than 0.012, while any artifact below this threshold was discarded. It was empirically found that this combination of threshold values produced the most accurate detection of inhalations in the training series and was therefore applied to a new validation series of 255 separate files.

RESULTS

The algorithm was applied to acoustic signals obtained from asthmatic outpatients who attended a respiratory clinic. Figure 10 shows the identification and detection of temporary start / end of an inhalation in a typical inhaler register. The algorithm was designed so that various artifacts such as speech, inhaler manipulation, and background noise are not detected as events. The algorithm calculates both the start and end times of the inhalation.

To validate the algorithm, 255 audio files were randomly selected to be analyzed from the database of inhaler recordings. The audio files were randomly selected from 12 of 20 subjects who were part of the study and the files were also randomly selected from the three months of recordings of each subject. Two human evaluators, trained by a respiratory clinician experienced in identifying inhalations, independently classified each of the 255 audio files by visual and auditory inspection. Human assessors first identified whether there was an inhalation and secondly marked the start and end of inhalation. Human evaluators agreed the presence of inhalations in 100% of the audio files. The average difference between the evaluators in detecting the start time of inhalations was ± 19 ms, while the average difference in the end times was ± 15 ms.

Table I shows the performance of the algorithm in detecting inhalations, compared to that of human evaluators. The results were classified as True Positive (TP), False Positive (FP) and False Negative (FN), according to the classification of the human evaluators. The algorithm was found to have a sensitivity (Sen) of 95%, specificity (Spe) of 94%, and an accuracy (Acc) of 89% in detecting inhalations.

TABLE I

The result of the algorithm to accurately identify the start and end of inhalations is shown in Table II and Table III, respectively. For this analysis, only true inhalations were considered. positive. For the inhalation onset time, the average difference between the human assessors was ± 57ms and ± 61ms respectively. For the inhalation completion time, the average difference was ± 104 ms and ± 107 ms. Considering that an average inhalation was found to be 1.8 seconds in duration, the algorithm inhalation onset time classification ranged ± 3.16-3.38%, compared to the rater classification humans. Furthermore, it was found that the completion time of inhalation algorithms varies by ± 5.77-5.94%, compared to the classification of human evaluators.

TABLE II

TABLE III

The system of the invention has been designed to automatically detect and demarcate inhalations from recordings of inhaler use in real-world settings. Validation of the algorithm was completed by running it on 255 audio files obtained from actual inhaler records of asthma patients and comparing it to the results of manual classification. The results have indicated that the algorithm was able to detect, on average, inhalations in 95% of the audio recordings containing inhalations according to human evaluators. The algorithm had a specificity of 94%, while accurate identification of inhalations occurred, on average, in 89% of audio files. This high level of precision is a promising result if this approach is included in a fully automated system for identifying inhalations of audio recordings.

From the inhalations that the algorithm detected, it was observed that it was able to identify the start / end times of the inhalations with a high degree of precision.

Compared to human assessors, the algorithm differed in the inhalation start time by ± 57ms and ± 61ms and in the inhalation completion time by ± 104ms and ± 107ms. A possible explanation for why the algorithm was not as accurate in detecting the end time of the inhalation, compared to the start time, can be found in the mechanics of inhaler use.

Inhalation of asthma medications using inhalers involves deep and constant inhalation of the user, in order to successfully inhale the drug into the small airways of the lungs. These inhalations have a characteristic pattern, both in the time domain and in the frequency domain, when users follow the correct inhalation technique. The start of an inhalation is usually accompanied by a period of silence in the period prior to inhalation. Although artifacts may occasionally interfere with the accuracy of identification of inhalation onset time, the algorithm achieves a good correlation compared to that of human evaluators.

Precise identification of inhalation completion time from inhaler recordings represents a more challenging task. As patients inhale the drug from their inhalers, there is a tendency to gradually decrease the inhalation flow rate in the last few hundred milliseconds of inhalation. At the end of inhalation, the patient will withdraw his lips from the mouthpiece of the inhaler device before closing his mouth and holding his breath. The reduction in the rate of inhalation to completion, the sound artifacts produced by the removal of the lips from the mouthpiece, in addition to the artifacts associated with the clumsiness of the inhaler as it is removed from the mouth area, are a series of factors that contribute to hindering the accurate identification of inhalation completion times.

The following modifications may be included in a version of the algorithm:

• Audio bandpass filtering above 1300Hz

• Comparison of the mean amplitude of the identified inhalations with the thresholds calculated from experiment 2, generating a classification of the quality of the identified inhalation

• Measurement of the peak inspiratory flow rate (PIF) of the acoustic signal to provide an objective measure of lung health from inhalers.

It will be appreciated that the invention provides an algorithm that can accurately detect and demarcate inhaler registers and has a wide range of implications for both practitioners and asthma sufferers. Incorporating this algorithm into devices that can record audio signals from patients taking their asthma medications opens the door to a whole new approach to monitoring adherence. The algorithm provides a quick and easy procedure to analyze patients 'inhaler use, and therefore can provide clinicians with strong empirical evidence of patients' adherence to their medication. This information can be used to give active feedback to patients. Such feedback can encourage patients to take better control of their asthma and lead to an overall improvement in their adherence to their medication. This in turn can improve the effectiveness of the drug treatment regimen, reduce the occurrence of asthma attacks and decrease hospitalizations. It will be appreciated that the applications of the algorithm include the extraction of relevant inhalation characteristics that can be used to provide real-time information about the patient's lung conditions in a remote monitoring scenario.

In one embodiment, the algorithm is adapted to automatically analyze adherence to the patient's inhaler. A patient's temporary adherence to their inhaler medication can be analyzed from the date and time stamp of each audio file obtained. Adherence to the technique can be analyzed by detecting the breathing and blistering events in the audio signal, the number of each event present, and the order in which the events take place. The invention provides a means for all this information on the use of the inhaler to be compiled in a format that is easy to understand and accessible to both the physician and the patient.

Acoustic recording device

As described above, the device according to an embodiment of the present invention can be used in conjunction with a commonly used Diskus inhaler. The device starts recording once the Diskus inhaler is opened and turns off when it is closed. The acoustics of inhaler use are recorded as mono WAV files, at a sample rate of 8000 Hz and a resolution of 8 bits / sample. The device has sufficient battery life to record the use of the patient's inhaler for up to a period of one month.

To validate the performance of an algorithm, data from 12 asthmatic patients residing in the community (6 women and 6 men) were recorded. The age range of the recruited patients was 20-83 (mean 49 ± 18 years). All patients had previous experience using the Diskus DPI inhaler. The Diskus inhaler used contained the combined drug Seretide, which is made up of salmeterol and fluticasone propionate. There were 60 doses of Seretide drug in each inhaler.

Each patient received a Diskus inhaler from their doctor along with the device for a period of one month. The device was securely attached to the inhaler where it had no affect on the mechanics of inhaler use. Patients were instructed to use their inhaler normally and were not given any additional advice or special training. Patients used their inhaler as part of their normal daily routine. After using their device-enabled inhaler for a month, the patients returned to their clinic where the device was removed from the inhaler and the audio files were uploaded to a database for analysis.

Proper use of Diskus inhaler

The Diskus inhaler was originally designed to facilitate patient use and acceptability and is illustrated in Figure 1. When patients receive a Diskus inhaler, their doctor instructs them on how to use the device correctly. To validate the device of the present invention, patients were instructed to use their inhaler twice daily. As there were 60 doses in each inhaler this corresponds to 30 days of correct use. The Diskus inhaler is opened by sliding a thumb grip to expose the mouthpiece. When this occurs, the device turns on and begins recording audio. Then a lever is pulled that opens a blister sheet containing the medication inside the mouthpiece (blister event). A clicking noise indicates that the vial sheet has been punctured and medication is available at the mouthpiece for inhalation. The patient is then instructed to gently exhale away from the mouthpiece, taking special care not to exhale into the mouthpiece. They should then seal their lips tightly around the mouthpiece, inhale steadily and deep and hold your breath for 10 seconds. Then, after 10 seconds, the patient should exhale slowly. Once this is complete the patient should use the thumb grip again to slide the Diskus inhaler back to its original position. When the Diskus inhaler is completely closed, the device will turn off and save the audio file in its internal memory.

Technical adherence algorithm

The algorithm designed to detect common inhaler events initially went through a training phase. The 12 recruited patients provided 609 audio files in total. Of these, 202 (33% of the total files available) were randomly selected and used in the algorithm training phase. This specific number of files was selected for the training phase in order to develop a robust estimate of typical inhaler use. The inhaler events that will be specifically detected from the audio recordings are blisters and breaths (both inhalations and exhalations). To detect blister events, characteristics such as mean power in selected frequency bands, amplitude, and duration are calculated. The presence of blisters, exhalation and inhalation events are shown in Figure 11 (a) time domain and (b) signal spectrogram illustrating the ideal use of the inhaler. A Cepstral Frequency Coefficients (MFCC) approach was used to detect breaths in this study due to the fact that breaths have a characteristic MFCC pattern that allows them to be distinguished from other sounds. Figure 12 shows a block diagram of the basic steps the algorithm takes to analyze inhaler records.

As mentioned above, the algorithm calculates the MFCCs to detect breaths in this study. MFCC extraction is a parameterization procedure for vocalization, due to the fact that MFCC models the known variation of the critical bandwidth of the human ear with frequency. It will be appreciated that there are other methods that can be used to detect breaths, such as support vector machines (SVM) and Gaussian Mix Models (GMM).

The algorithm automatically examines each audio file in four different stages. The algorithm first identifies the perforation of the ampoule containing the drug (stage 1), before identifying the sounds of breathing (stage 2). Then, differentiate each breath sound detected as inhalation or exhalation (stage 3). Finally, the algorithm calculates a user technique score (step 4) for each individual audio file. Scoring for this technique is based on the presence of blistering and breathing events in stages 1-3, the order in which they occur, and the number of each event.

The first stage of the algorithm involves detecting the perforation of the sheet of the ampoule containing the medication. The audio signal is segmented into windows of 100 ms in length, with an overlap of 10 ms. The mean power spectral density (PSD) is calculated for frequencies between 2000 Hz-3000 Hz. For this frequency band it was found from empirical observations in the data set that the ampoule sounds had an average power greater than -65 dB. The reason the power in this frequency band was higher for blisters compared to non-blisters is due to the intrinsic sound associated with blistering on the drug sheet in the Diskus inhaler. A fixed threshold was established with any frame greater than this threshold considered as possible blister sounds. The algorithm then examines the proposed ampoule sounds to remove any false positives. Possible blister sounds with a maximum normalized amplitude less than 0.7 are eliminated, in addition to possible blister sounds of more than one second in duration. Finally, the mean PSD is calculated in the 20 Hz-200 Hz frequency band. It was discovered in the training data set that the ampoules had a higher mean PSD than any false positives in this frequency range, due to the distinctive sound from a blister. Potential blisters with a power less than -62 dB are considered false positives and are eliminated, leaving only the true blister positive events. Figure 13 shows a flow chart of the steps used to detect blister events.

Stage 2 of the algorithm involves detecting breath sounds. The audio signal is first filtered to remove high frequency components above 1400 Hz using a low-pass type sixth order Chebyshev filter. Each signal is separated into 700 ms long frames with 20 ms overlap. Twelve MFCCs are calculated for each frame, forming a cepstrogram of the short-term signal. Then Singular Value Decomposition (SVD) is used to obtain a normalized singular vector of the signal cepstrogram. Singular vectors capture the most important characteristics of respiratory sounds obtained from MFCC calculations. An adaptive threshold is established that is 7% higher than the lowest singular vector in the inhaler recording. Singular vectors above this adaptive threshold are marked as potential respiration events, while those below are discarded. This adaptive threshold was found empirically to produce the most accurate detection of breaths in the training set. The average zero crossing rate (ZCR) is calculated to reduce the number of false positive breaths detected by the algorithm using equation 1, mentioned above.

Respirations were found to have a characteristically high ZCR compared to non-breaths in the training data set. Therefore, a fixed threshold constant of 0.1 to reflect this fact. In the training data set, breaths always had a ZCR above this threshold value, while false positives were successfully removed.

Stage 3 involved differentiating breaths into inhalations and exhalations. To do this, the mean power spectral density (PSD) of the identified breaths is calculated for frequencies between 2520 Hz-4000 Hz in the original unfiltered signal. It was discovered from empirical observations in the training data set that inhalations had greater power in this specific frequency band compared to exhalations. Based on this fact, a fixed threshold can be established. Inhalations were found to have a mean power greater than -80 dB and exhalations had a mean power below this value. The standard deviation of the ZCR was also found to be higher for inhalations compared to exhalations in the training data set. A fixed threshold of 0.045 was established with inhalations having a value greater than this threshold and exhalations with a value below this threshold. Figure 14 shows a flow chart of the processing steps that the algorithm used to detect inhalations and exhalations in the audio files.

The last stage of the algorithm (stage 4) is to analyze all the events that took place in the audio file and make a decision regarding the quality of a patient's inhalation technique. This information is issued as a score that can be one of three possibilities; (1) used correctly, (2) technical error, or (3) not used. To arrive at this decision, the algorithm checks what events have occurred, the frequency of such events and the order in which they have occurred, as shown in Figure 15.

The inhaler is considered to have been used correctly if a patient first blisters the foil and then inhales the medication. For the purpose of this algorithm it was decided that there does not need to be an exhalation event for the inhaler to be used correctly. The exhalations can take place before the ampoule or after the inhalation, which still leads to a “used correctly” score from the algorithm. However, if the patient exhales vigorously in the time between blister formation and inhalation, they are considered to have made a "technical error" as they may have exhaled into the mouthpiece of the inhaler and dispersed some of the medication. Such a scenario is seen as a critical error. Any other sequence of events is considered a technical error. For example: An inhalation followed by an ampoule, an ampoule but without inhalation, inhalation but not inhalation, etc. If the algorithm detects two or more inhalations or blisters, a technical error will also be considered to have occurred.

To test the performance of the algorithm, 407 new audio files were selected from the 12 asthmatic patients recruited (67% of the total audio files obtained). Two human evaluators, trained by an experienced breathing physician to identify correct / incorrect use of the Diskus inhaler, independently evaluated each of the 407 audio files using the Audacity® audio tool. Each human evaluator manually examined the audio files using visual and auditory methods and rated each individual audio file as one of three possible outcomes: (1) used correctly, (2) technical error, or (3) not used.

Temporal adhesion analysis

As mentioned above the device also provides information on the exact time and date the Diskus DPI inhaler was used. Using these data the algorithm automatically calculated the number of doses taken each day and represented this information in a bar graph format. Any audio file less than a second long is discarded for this analysis due to the fact that this is not a sufficient period of time to use the inhaler properly.

The designed algorithm is adapted to detect ampoule, inhalation and exhalation events, analyze the frequency of each event, in addition to the order in which they occurred and generate a score in the user's technique each time the inhaler is used. The algorithm also analyzed the time and date the inhaler was used to generate feedback on the patient's temporary adherence. Figure 16 illustrates a typical algorithm output that can be used to assess the patient's temporary adherence. This bar graph shows whether a patient is overdosing, underdosing, or taking the correct number of doses of their medication for each day they must use their inhaler.

The patient user technique score for each inhaler audio file, as calculated by the algorithm, is stored in a text file. However, in order to present the specific technical score that the algorithm calculates for both clinicians and patients, it was decided that a more interpretable version of the presentation of this information would be necessary. Previous research has suggested that people perceive visual cues more accurately than information positioned along a common scale. Based on this information, the best procedure to visualize data is through the use of scatter diagrams and bar graphs. For this reason, a bar graph was used to display information on temporal adherence. Colors are also widely used in data visualization to indicate appropriate levels of risk (i.e., green = safe, red = danger). A traffic light scatter diagram was created to show the results of the algorithms in technical adherence.

An example of such an arrangement is shown in Figure 17. This output graph shows information about the time and date the inhaler was used, as well as how the inhaler was used. A green color indicates that the inhaler was used correctly, while an orange color indicates that there was a technical error. This allows clinicians to examine a patient's adherence to their inhaler medication, while also providing a procedure for patients to easily understand when and how they are using their inhaler. Figure 18 illustrates a block diagram of the components to form the inhaler device according to an embodiment of the invention as described above.

In another aspect of the invention, the device of the invention can be used with a web portal service as shown in Figures 19 and 20. The web portal is an online system for collecting, organizing and displaying the results of the protocols of inhaler monitoring. The web portal uses the audio recordings, their file names and the output of the processing algorithm as inputs to provide the user with detailed information on the use of the inhaler.

Recordings taken from the device are uploaded to the web portal server. The time and date of each recording are extracted from the time stamp of the audio recording, that is, the file name and this information is used to present a visual representation of the times the inhaler was used. Plotting these data reflects events of the inhaler that are outside the recommended use of the inhaler, i.e. doses of medication taken too early, indicating excessive doses and doses of medication taken too late, indicating missed doses.

The web portal makes it easy to manually classify audio recordings. Manual classification allows the user to listen to each recording and classify it according to a set number of options such as: pass, fail, "exhalation present after ampoule and before inhalation" and "breath not held for 5 seconds". These options have been established based on research that has been conducted in terms of Diskus inhalers, but can be modified to include options that are appropriate for other inhalers. The result of manual sorting will be illustrated in the diagram. The web portal can also read the output results of the processing algorithm and display the results in a visual format. The algorithm classifies each record as pass, failed, or incomplete and the associated reasons for incomplete and failed, all are illustrated in the generated diagrams. The web portal also provides an over-reading function, all files can be reviewed by an over-reader and indicates an agreement or disagreement. The web portal can be adjusted to reflect various inhaler usage protocols and additional errors that might be seen with different inhalers.

A typical use of the web portal according to the invention is when a patient returns the device to a physician who uploads the data to the web portal. The algorithm processes the loaded data and provides a diagram of the latest inhaler events ranked in terms of temporal adherence and technical adherence. The physician could visually inspect the data to determine if the patient is using their inhaler correctly, both in terms of dose taking and inhaler technique, and advise on any strategies to improve adherence. The physician can also view the adherence profiles from the previous months to determine if there are long-term problems with adherence that may require additional intervention. It will be appreciated that some of the functionality of the algorithm can be stored on the inhaler device or on a separate computing device. The data recorded by the inhaler device can be downloaded to the separate computing device for further processing of the data as described above.

The embodiments in the invention described with reference to the drawings comprise a computer apparatus and / or processes performed on a computer equipment. However, the invention also extends to computer programs, particularly computer programs stored on or on a medium adapted to carry the invention into practice. The program may be in the form of source code, object code, or intermediate source code and object code, such as in partially compiled form or in any other form suitable for use in implementing the procedure according to the disclosure. The medium may comprise a storage medium such as ROM, for example, CD ROM, or magnetic recording medium, for example, a memory card or hard disk. The support can be an electrical or optical signal that can be transmitted through an electrical or optical cable or by radio or other means.

In the specification, the terms "comprise, includes, understood and comprising" or any variation thereof and the terms "include, includes, included and including" or any variation thereof are deemed to be fully interchangeable and should be give everyone the widest possible interpretation and vice versa.

The invention is not limited to the embodiments described hereinabove, but can be varied both in construction and in detail.

Claims (14)

1. A system for monitoring the user technique of an inhaler device configured for the administration of a medicine, said system comprising: a microphone adapted to detect the sound produced during the operation of an inhaler device; operable processing circuitry for processing a data signal obtained from the microphone, wherein said data signal comprises detected acoustic information, The processing circuitry is adapted to determine the characteristics of inhalation and exhalation respiration that occur during use, analyzing the temporal and spectral components of the acoustic information detected and processed to differentiate between an inhalation and an exhalation, depending on the temporal and spectral components, characterized in that the processing circuitry further comprises means for identifying a priming or blistering characteristic of the drug in the device to identify that a drug is about to be delivered prior to inhalation by analyzing an acoustic signal generated by a lever mechanism. of the inhaler device, adapted to release the drug that generates a unique energy profile of the lever mechanism related to the drug release noise. 2. The system of any preceding claim in which the analysis of the temporal and acoustic components provides an indicator of temporal adherence indicating that the correct volume of medication is administered at the correct time intervals. 3. The system of claim 2, wherein the processing circuitry is adapted to be reconfigured to process different temporary adherence requirements depending on the device and a drug to be used. 4. The system of any preceding claim comprising means for identifying multiple inhalations; or an exhalation before inhalation after a drug priming mechanism has been activated, for example after blistering in a capsule. 5. The system of any claim preceding comprising means for identifying a volume of insufficient inhalation for release of medication or administering the correct volume after medication has been activated the priming of the drug, for example, after formation of capsule blisters. 6. The system of any preceding claim wherein the processing circuitry comprises means for identifying different frequency components to differentiate between inhalations and exhalations that will be used for the identification of user technique errors. 7. The system of any preceding claim comprising means for comparing a zero crossing rate, a singular value decomposition of a short-term power spectrum of the data signal at a predefined threshold to identify possible respiratory sounds to be used for the identification of user technique errors. 8. The system of claim 7, wherein a speed of zero crossing comparing breath sounds with predefined threshold to confirm these as respiratory sounds using said comparison means. 9. The system of any preceding claim wherein the signal is processed to determine the frequency domain components of the identified breath sounds and the means to classify the detected sounds as inhalation or exhalation. 10. The system of any preceding claim comprising means for calculating a median acoustic amplitude and the duration of an identified inhalation to determine an amount of drug released from the device. 11. The system of any preceding claim comprising means for calculating an amplitude of an identified exhalation in order to determine whether the drug has been dispersed from the inhaler device, before inhalation has occurred. 12. The system of any preceding claim which comprises means for calculating a minimum energy to disperse the medicine that has been achieved in the exhalation in the airways. 13. The system as claimed in any of the preceding claims, comprising a second microphone adapted to determine the breathing characteristics by inhalation or exhalation and / or to cancel ambient noise. 14. A computer program comprising program instructions to cause the computer program, when run on the processing circuitry of a system as claimed in claim 1, to perform the following steps: process a data signal obtained from a microphone, in which said data signal comprises detected acoustic information; determine the characteristics of inhalation and exhalation respiration that occur during use, analyzing the temporal and spectral components of the acoustic information detected; differentiate between an inhalation and an exhalation based on the temporal and spectral components; and identifying the priming or blistering characteristics of the drug from the inhaler device monitored by the system to identify that a drug is about to be released before an inhalation by analyzing an acoustic signal generated by a lever mechanism of the inhaler device. and adapted to release the drug which generates a unique energy profile of the lever mechanism related to the noise of drug release.