Automated control for flow limit detection
The system uses a fuzzy logic control system to enhance the detection of flow limitations in CPAP therapy by incorporating shape indices and respiratory measurements, addressing the inadequacies of existing methods and improving treatment outcomes for obstructive sleep apnea.
Patent Information
- Application Number
- CN202411391557.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2007-08-17
- Filing Date
- 2008-05-09
- Publication Date
- 2024-12-06
AI Technical Summary
Existing methods for detecting and treating obstructive sleep apnea (OSA) are inadequate in addressing certain types of flow limitations and partial obstructions, particularly M-shaped breathing patterns, leading to delayed detection and inappropriate pressure adjustments in continuous positive airway pressure (CPAP) therapy.
A system and method for detecting flow limitations using a fuzzy logic control system that incorporates multiple features such as shape indices, ventilation measurements, and respiratory duty cycles to improve detection accuracy, allowing for more precise and timely pressure adjustments in CPAP therapy.
Enhances the sensitivity and specificity of detecting flow limitations, enabling faster and more appropriate pressure adjustments in CPAP therapy, thereby improving treatment efficacy for obstructive sleep apnea.
Smart Images

Figure CN119075110A_ABST
Abstract
Description
[0001] This application is a further divisional application of the divisional application 202111150974.4, filed on May 9, 2008, with application number 201910046296.3 and name “Automatic Control for Flow Limitation Detection”.
[0002] CROSS-REFERENCE TO RELATED APPLICATIONS
[0003] This application claims priority to Australian Provisional Patent Application No. 2007902561, filed May 11, 2007, and U.S. Provisional Patent Application No. 60 / 965172, filed August 17, 2007, the disclosures of which are incorporated herein by reference. Technical Field
[0004] The present technology relates to methods and apparatus for the detection, diagnosis and / or treatment of breathing conditions such as those associated with sleep apnea hypopnea syndrome (OSAHS) or obstructive sleep apnea (OSA). Background Art
[0005] As described in U.S. Pat. No. 5,199,424, issued April 6, 1993 to Sullivan & Lynch, the application of continuous positive airway pressure (CPAP) has been used as a means to treat obstructive sleep apnea. The patient is connected to a source of positive pressure air through a nasal mask or nasal prongs. The patient breathes air at a pressure slightly greater than atmospheric pressure. It has been found that the application of continuous positive airway pressure provides what is described as an "inflatable splint" that supports and stabilizes the upper airway and thereby eliminates the occurrence of upper airway occlusion. This is effective in eliminating both snoring and obstructive sleep apnea, and in many cases, is effective in treating central and mixed apneas.
[0006] In U.S. Pat. No. 5,549,106, invented by Gruenke and published on August 27, 1996, a device is disclosed for facilitating a patient's breathing in the treatment of mixed and obstructive sleep apnea. The device increases the nasal air pressure delivered to the patient's respiratory passages just before inspiration and then reduces the air pressure to reduce the exhalation effect.
[0007] U.S. Patent No. 5,245,995 to Sullivan discusses how to detect snoring and abnormal breathing patterns during sleep by measuring inspiratory and expiratory pressures, thereby providing an early indication of pre-obstructive events or other forms of respiratory disturbances. In particular, patterns of respiratory parameters are monitored, and upon detection of a predetermined pattern, CPAP pressure is increased to provide increased airway pressure, thereby ideally disrupting the occurrence of obstructive events and other forms of respiratory disturbances.
[0008] Various techniques are known for sensing and detecting abnormal breathing patterns indicative of obstructive breathing, such as described in U.S. Pat. No. 5,704,345, invented by Berthon-Jones and published on January 6, 1998. Berthon-Jones describes a method based on detecting events such as apnea, snoring, and respiratory flow flattening. Therapeutic pressure is automatically adjusted based on the detected conditions.
[0009] Flow flattening may further be determined based on different weighting factors as described in International Patent Application PCT / AU01 / 01948 (published as WO0218002) by Wickham.Weighting factors are applied to the partial flows to improve sensitivity to various types of respiratory obstruction.
[0010] Other methods for detecting obstruction have also been used. For example, in U.S. Pat. Nos. 5,490,502 and 5,803,066, Rapport et al. disclose methods and apparatus for optimizing controlled positive pressure to minimize airflow from a flow generator while attempting to ensure that flow limitation does not occur in the patient's airway. The controlled positive pressure to the patient's airway is adjusted by detecting flow limitation based on the shape of the inspiratory flow waveform. The pressure setting is increased, decreased, or maintained based on whether flow limitation has been detected and previous actions taken by the system.
[0011] In U.S. Pat. No. 5,645,053, Remmers describes a system for automatically and continuously adjusting the level of nasal pressure relative to an optimal value during the treatment of OSA (obstructive sleep apnea). Parameters related to the shape of the time curve of the inspiratory flow are determined, including the roundness and flatness of the inspiratory curve. OSA treatment is then performed by automatically re-evaluating the applied pressure and continuously searching for the minimum pressure required to adequately dilate the patient's pharyngeal airway.
[0012] Although such devices have some effect on treating OSA, some sleep obstruction events still cannot be treated using some devices. Therefore, it is desirable to obtain new methods for automatic detection and treatment of obstruction events.
[0013] RESMED 3.4-001CN D5(2635) Summary of the invention
[0014] In one aspect of the present technology, apparatus and methods provide for improved automatic detection and / or automatic treatment of sleep disordered breathing or flow limitation.
[0015] In another aspect of the present technology, improved detection and / or treatment of flow waveforms indicative of partial occlusion is provided.
[0016] In another aspect of the present technology, the apparatus and methods provide for a faster response to indications of partial obstruction or flow limitation.
[0017] In another aspect of the present technology, apparatus and methods are provided for quantifying a measure of flow limitation or partial obstruction by measuring or detecting a second condition of the airway to more accurately detect obstructive events, such as a waveform of flow limitation or its shape index.
[0018] Aspects of the present technology include a method for detecting flow limitation, the method comprising determining a measurement of respiratory flow, determining a shape index indicating a flow limitation pattern based on the respiratory flow measurement, determining a ventilation measurement or a respiratory duty cycle measurement based on the respiratory flow measurement, and deriving the flow limitation measurement and either or both of the determined ventilation measurement and duty cycle measurement based on the determined shape index function.
[0019] In some embodiments, the shape index can be a flatness index, an "M" shape index, and a chair shape index ("chairshaping") and / or a roundness index, etc., or other indices that can indicate partial obstruction. The ventilation measurement value can be a tidal volume measurement value (such as the ratio of the current tidal volume to the previous tidal volume). The respiratory duty cycle measurement value can be a ratio such as the ratio of the inspiratory time of the current breath to the respiratory cycle time and the ratio of the inspiratory time of the previous average breath to the respiratory cycle time.
[0020] The method may be performed by a flow limitation detector and / or a flow limitation pressure therapy device. For example, the method is performed to adjust a therapy pressure of the pressure therapy device, such as by increasing the pressure to satisfy a condition of a shape index indicating the presence of an M-shaped breath in the respiratory airflow and a sufficient decrease in the ventilation measurement to indicate less than normal ventilation. Optionally, the therapy pressure value may be adjusted or increased based on the condition that the shape index indicates the presence of an M-shaped breath in the respiratory airflow and an increase in the duty cycle measurement.
[0021] In one embodiment of the present technology, an apparatus for detecting flow limitation may include a patient interface for delivering breathable gas, a flow sensor coupled to the patient interface to generate a flow signal representative of the flow of breathable gas through the patient interface, and a controller coupled to the flow sensor to process the flow signal, wherein the controller is configured to control a method of detection to derive a flow limitation measurement based on a shape index, and either or both of a ventilation measurement and a respiratory duty cycle measurement. The apparatus also includes a flow generator coupled to the controller and the patient interface, whereby the controller may be configured to calculate a pressure requirement based on the flow limitation measurement, and to set the flow generator based on an adjusted pressure requirement such as described herein.
[0022] In one embodiment of the present technology, a system for flow limitation detection may include an interface device that delivers a flow of breathable gas, a flow measurement device coupled to the interface device to generate a flow signal indicative of the flow of breathable gas through the interface device, and a processing device coupled to the flow measurement device to process the flow signal, wherein the processing device is configured to process the detection method to derive a flow limitation measurement based on a shape index and either a ventilation measurement or a respiratory duty cycle measurement. The system also includes a flow device coupled to the processing device and the interface device to generate a controlled flow of breathable gas through the interface device, such that the processing device is configured to calculate a pressure requirement as a function of the flow limitation measurement and set a flow generator based on the pressure requirement.
[0023] In another embodiment, the technology can be an information bearing medium having processor-readable information thereon, whereby the processor-readable information can control an apparatus for detecting respiratory flow limitation. The processor-readable information can include control instructions for determining a respiratory flow measurement, determining a shape index indicative of a flow limitation pattern based on the respiratory flow measurement, determining a ventilation measurement and / or a respiratory duty cycle measurement based on the respiratory flow measurement, and deriving a flow limitation measurement based on the determined shape index, and either or both of the determined ventilation measurement and respiratory duty cycle measurement. The processor-readable information also includes instructions for calculating a pressure requirement based on the flow limitation measurement and adjusting pressure based on the description herein.
[0024] Another aspect of the present technology includes the generation of pressure, flow control information, or signals that are derived proportionally as a function of a measurement of flow limitation. For example, a controller or processor is configured to control or determine a pressure setting or flow rate setting for a respiratory therapy device. Based on the measurement of respiratory flow, an obstruction measurement or a shape index representing the degree of obstruction or flow limitation can be determined or calculated by the device. The device further optionally determines a ventilation measurement representing the degree of variability in ventilation from the respiratory flow measurement. The treatment pressure setting or flow rate setting can be derived by the device based on a proportional function of either or both of (1) the degree of obstruction or flow limitation and (2) the degree of variability in ventilation.
[0025] Other embodiments of the present technology will become apparent from the following disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] The present technology is illustrated by way of example, but not limitation, and like reference numerals in the accompanying drawings represent similar components, including:
[0027] Figure 1The patient is shown first breathing normally with airflow, then experiencing airway collapse and finally awakening;
[0028] RESMED 3.4-001CN D5(2635)
[0029] Figure 2 A flat index is shown for a patient experiencing obstruction, indicated by an "M" shaped breathing pattern;
[0030] Figure 3 shows a flatness index with and without multiple breath averaging for a patient experiencing obstruction, an "M" shaped breathing pattern indicating that the patient is experiencing obstruction, thereby indicating a breathing pattern of the patient arousing from sleep;
[0031] Figure 4 Example components of an apparatus for detection and / or treatment of flow restriction or partial obstruction are shown;
[0032] Figure 5 Suitable steps for testing or treating the device in detecting flow restriction or partial obstruction are shown;
[0033] Figure 6 An example fuzzy membership function for a variable based on a single-breath flat shape index is shown;
[0034] Figure 7 An example fuzzy membership function for a variable based on an m-shaped breathing shape index is shown;
[0035] Figure 8 An example fuzzy membership function for a variable based on a ventilation ratio index is shown;
[0036] Fig. 9 An example fuzzy membership function for a variable based on breathing duty cycle is shown;
[0037] Fig.10 An example output membership function for flow limit measurement is shown;
[0038] Fig.11 An example defuzzification operation in determining flow restriction or partial blockage measurements using a centroid method is shown;
[0039] Fig.12 An example function for modifying the detection performance of flow limitation or partial occlusion measurement based on ventilation measurement is shown;
[0040] Fig.13 is a graphical comparison of a conventional flatness index with the present technology's calculated flow restriction measurement FFL determined from the same flow signal as described herein;
[0041] Fig.14 A flow signal derived from nasal flow and output measurements of a continuous RERA detector quantifying the likelihood of respiratory arousals are shown;
[0042] Fig.15 The ratio of peak to mean expiratory flow is shown, which can be an index of sleep-wake-awake detection;
[0043] Fig.16 It is a recording of two overlapping filtered and unfiltered flow signals during a series of obstructive breaths in the presence of low-frequency snoring;
[0044] Fig.17 shows the response of an example FIR filter useful for filtering an airflow signal;
[0045] Fig.18 Example functions acting as basis vectors B1 & B2 for detecting M-shaped breathing are plotted;
[0046] Fig.19 It is a graph of the flow signal for a typical M-shaped breathing;
[0047] Fig. 20 is a graph showing a flow signal of an M-shaped breathing with a simple chair shape augmentation;
[0048] Fig.21 is a graph of membership functions for variables based on the ratio of peak to mean inspiratory flow, which is useful for detecting M-shaped breathing with a chair shape;
[0049] Fig. 22 is a graph of membership functions for variables based on the peak inspiratory position, which is useful for detecting M-shaped breathing with a chair shape;
[0050] Fig.23 showing a graph of a flow signal, a graph of determined flat measurements, and a graph of determined chair measurements;
[0051] Fig.24 shows a graph of the flow signal for a patient experiencing upper airway resistance with increased breathing effort terminated at t=420 sleep arousal and a corresponding graph of the single breath flatness index;
[0052] Fig.25 Shown based on Fig.24 Example graphs of flow limitation measurements, ventilation measurements, and duty cycle measurements of a flow signal;
[0053] Fig.26 is based on Fig.24 a graph of a continuous occluded measurement or a continuous flat measurement of the respiratory sequence shown in the flow signal;
[0054] Fig. 27 A function for a de-weighting factor is shown for modification of the flow limit measurement;
[0055] Fig.28 a graph showing a respiratory flow signal, a graph showing an instantaneous snoring signal, and graphs showing three snoring indices;
[0056] Fig.29 A classifier for treating snoring based on various snoring indices is shown;
[0057] Fig.30 is a graph of M-shaped breathing with corrected and uncorrected trimming results indicated thereon;
[0058] Fig.31 including a graph of flow signal proportional to time with m-shaped breathing and a graph of inspiratory time determined by two methods;
[0059] Fig.32 A graph showing a function useful for de-weighting flow limitation measurements according to leak conditions is shown;
[0060] Fig.33 is another function useful for de-weighting flow limitation measurements or partial occlusion measurements (such as FFL) based on leak conditions;
[0061] Fig.34 is a graph of a function suitable for de-weighting flow limitation or partial occlusion measurements by mask pressure level;
[0062] Fig.35 is a graph of a function suitable for de-weighting flow limitation or partial blockage measurements by interference measurements;
[0063] Fig.36 is a graph of a function suitable for de-weighting snoring measurements according to treatment pressure;
[0064] Fig.37 is a graph of a function suitable for de-weighting a snore measure based on a ventilation measure indicative of large breathing;
[0065] Fig.38 is a histogram of normalized peak expiratory positions for various groups of patients; and
[0066] Fig.39 is a graph of a function of the de-weighting factor applied to flow limitation or partial occlusion measurements based on normalized expiratory peak position values. DETAILED DESCRIPTION
[0067] Patients with OSA have periodic apnea or hypopnea during sleep, which are terminated only by patient awakening. These periodic events stimulate sleep segmentation and the sympathetic nervous system. This has serious consequences for patients, including daytime sleepiness (with the incidental possibility of motor vehicle accidents), lack of energy, memory problems, depression and excessive tension. Patients with OSA may also snore loudly, so they can also affect the sleep of their companions. The best form of treatment for patients with OSA is continuous positive airway pressure (CPAP) applied by a hair dryer (compressor) via a connecting hose and a mask. During inspiration, positive pressure prevents the collapse of the patient's airway, therefore, preventing periodic apnea or hypopnea and their sequelae.
[0068] Typically, patients undergo two sleep studies in which they are monitored using multiple sensors, which record what is known as a polysomnogram. The first study (without treatment) confirms OSA, while the second is used to titrate the patient to correct the treatment pressure. Pressure requirements vary throughout the night due to changes in position, posture, and sleep state. The physician will recommend a pressure that may cover any event (titrated pressure). An alternative to this is an automatic machine that adjusts the pressure according to the patient's needs; this treatment is called automatic positive airway pressure or APAP. The advantage of APAP treatment is that it is adaptable to pressure requirements, which can vary over multiple time scales. For example, pressure requirements vary in the following situations, that is, in nights with posture and sleep state, perhaps on weekends when drinking alcohol, in months due to the beneficial effects of the treatment itself, and in years due to weight loss or gain. Also, because APAP machines typically increase pressure when needed, patients can go to sleep at a comfortable pressure far less than any treatment maximum. Finally, because APAP machines can work at low pressures, the effect of mask leaks is improved to some extent.
[0069] A known algorithm for automatically setting patient pressure in an APAP machine is called ResMed AutoSet. In summary, the AutoSet device and its algorithm are very good for treating OSA patients. The ResMed AutoSet algorithm responds to three things: flow limitation, snoring (audible noise), and apnea. Setting automatic pressure is also described in U.S. Patent No. 5,704,345, so its content is specifically cross-referenced here. Flow limitation is the fluid dynamic property of a so-called "collapsible tube" that transports fluid. The pharynx of a patient with OSA is an example of a collapsible tube (a muscular tube rather than a simple passive tube studied on the bench). Basically, fluid limitation is a situation in which the flow is delivered in a collapsed tube, where (assuming that the upstream pressure remains constant) the flow can no longer be increased by reducing the downstream pressure (i.e., the flow drives an increase in differential pressure). In a patient with upper airway collapse, this is equivalent to a situation in which the patient is no longer receiving adequate ventilation and is still increasing in respiratory effort, but no longer increasing the suction flow rate. Figure 1 An example of such a patient is shown, typically breathing first at 102, then experiencing airway collapse at 104, without control to increase tidal volume or flow rate even with a more negative shift in esophageal pressure (a measure of respiratory effort). Finally, the patient awakens at 106, thereby opening the airway and taking a few large breaths to restore blood gas homeostasis.
[0070] The ResMed AutoSet algorithm monitors patient flow and raises pressure when it detects flow limitation or snoring. Apnea is rarely encountered because it is usually preceded by a period of flow limitation (also called partial obstruction) or snoring. As a backup measurement, pressure may also be raised if an apnea is detected. In the absence of any measured flow disturbances, pressure is allowed to decay slowly and hopefully achieve a sleep where equalized pressure allows the patient to wake naturally. The AutoSet algorithm responds proportionally and therefore a metric is used for each situation to which it responds. The metrics used are: Flatness Index for flow limitation, RMS measurement corrected for sound averaged on inspiration for snoring, and the length of any apnea detected.
[0071] The Flatness Index is a dimensionless characteristic (i.e., a real number) calculated using the patient's inspiratory waveform. It attempts to measure, essentially, how flat the top of the waveform is. A characteristic of flow limitation is that when the downstream pressure is low enough to keep the tube collapsed, the flow rate will remain more or less constant, regardless of changes in driving pressure. In a patient with flow-limited breathing, this equates to an inspiratory waveform with a flat top (i.e., a constant inspiratory flow rate).
[0072] For example, depending on the chosen scale of the index, a normal breath may produce a flatness index of about 0.2, while a strictly flat waveform may produce a flatness index of about 0.1 or less. For APAP therapy, a limit is typically determined (e.g., 0.19 in some machines) and the pressure is increased in proportion to how far below the threshold the flatness index is. To reduce the effects of noise and increase specificity, a typical pressure setting algorithm may also use a five-breath point-wise moving average. A possible disadvantage of a five-breath average is that it reduces the speed of flow limitation detection, since five abnormal breaths need to be averaged before the flatness drops to its lowest point. Three heuristic weights may be applied to the following thresholds at which flatness will cause a pressure rise, thereby requiring a greater reduction in flatness: when leaks increase, when therapy pressure increases, and when signs of valve-like breathing increase. These heuristics serve to prevent potential pressure from being released in the face of degraded information (flow signal).
[0073] While the flatness index is an excellent measure of flow limitation, it is designed to detect certain situations. However, in some special implementations, it has been observed that certain rare situations are not encountered. The following lists the areas where we have made such observations.
[0074] 1. The five-breath moving average reduces the speed of detecting flow limitation. Figure 2 In Figure 2 , the top trace shows a graph of a traditional five breath moving average flatness index. The bottom trace shows a measurement of respiratory flow. The patient starts out moderately obstructed and the five breath average flatness trace steps down at 202. To the right of the figure, the obstruction becomes more severe and increasingly more "M" shapes appear. As shown, the flatness index eventually begins to reverse direction and increases rather than decreases as the obstruction worsens.
[0075] 2. Because different inhalation shapes can be averaged to give a completely new shape, a five-breath moving average can have results. Figure 3 Shown in. Figure 3 Shown is a series of so-called M-shaped blocked breaths at 302, which terminate with an awakening at 304, followed by some appropriately normal recovery breaths at 306. As shown in the figure, both the single breath flat and the traditional flat are high at the end of the series of M-breaths (the former reaches the maximum faster) and the traditional flat actually falls below 0.1 after the awakening, not because the breaths are flattened, but because the M-breaths averaged by the normal breaths create a false flat shape.
[0076] 3. The Flatness Index is not designed to detect M-shaped breathing. In fact, the Flatness Index becomes higher when M-shaped breathing occurs. Figure 2 and3 Shown in.
[0077] 4. The Flatness Index may cause a pressure increase regardless of the current ventilation or sleeping state of the patient-user of the device.
[0078] 5. The heuristics applied to weight flattening may also lead to treatment in some patients.
[0079] 6. The flatness index is susceptible to regular random variables which have consequences for the sensitivity and specificity of any algorithm that uses it to detect flow limitations.
[0080] Although M-shaped breathing is rare, it is desirable to further develop methods and devices for detecting flow limitations and / or improve existing methods and devices.
[0081] Reference Figure 4 , the present technology includes a pressure delivery and / or flow limitation detection device, which may include a flow generator, such as a servo-controlled blower 402. The device typically also includes a patient interface, such as a mask 406, and an air delivery conduit 408 to allow air flow or breathable gas to be delivered to and / or from the patient. The blower 402 can be coupled to the air delivery conduit 408 and the mask 406. The exhausted gas is exhausted via an exhaust device 413. Optionally, a flow sensor 404f and / or a pressure sensor 404p can also be used. For example, a pneumotachogram and a differential pressure transducer or a similar device such as one that uses a bundle of tubes or pipes is used to measure the mask flow to obtain a flow signal F(t). Similarly, a pressure transducer can be used to measure the mask pressure at a pressure tap to obtain a pressure signal P 面罩 (t). Pressure sensor 404f and flow sensor 404p are only Figure 4 , because it can be understood that other configurations and other devices can be used to measure flow and pressure. 面罩 The (t) signal may be sent to a controller or microprocessor 415 via one or more analog-to-digital (A / D) converters / samplers (not shown) to obtain a pressure demand signal P 需求 (t).
[0082] Alternatively, for a blower motor, the flow signal f(t) and / or the pressure signal P may be monitored by monitoring current (I) applied to the motor, speed (S) and / or motor rotation (w) with or without a flow supply and pressure sensor as described above. 面罩(t) is estimated or calculated. Optionally, the blower motor speed is kept substantially constant and the pressure change in the mask is achieved by controlling the opening of a servo-valve, which can variably divert / exhaust or deliver the airflow to the mask. In addition, other types of patient interfaces can be used in place of the mask, and flow and / or pressure sensors can measure flow and pressure in alternative locations with respect to the patient interface.
[0083] The controller or processor 415 is configured and adapted to execute the methods or algorithms described in more detail herein and may include an integrated chip, memory, and / or other control instruction, data, or information storage medium. For example, programming instructions having control methods may be encoded on an integrated chip in the memory of the device or such instructions may be downloaded as software or hardware using a suitable medium.
[0084] With such a controller, the device can be used for treatment at a variety of different pressures by adjusting the pressure delivery formula used to set the speed of the blower or to manipulate the exhaust with the release valve, and can be based on the detection methods included in the pressure delivery formula as shown in the example exemplary embodiments described in detail herein. Alternatively, they can be implemented in a device without a pressure treatment component, thereby using the device to diagnose, detect and / or quantify the presence of flow restriction.
[0085] The methods and algorithms presented herein for detecting and / or treating breathing patterns indicative of partial obstruction can be implemented by a controller that is a fuzzy logic control system. In general, fuzzy logic is a way to convert opinions about the behavior of a system into computer code that is based on real numbers for parameters used to control the system. However, the problem at hand is essentially one of statistical pattern recognition and is implemented using other techniques, such as alternative techniques discussed in more detail herein. Therefore, while the following techniques are presented in terms of a fuzzy logic control system, it should be understood that there are other ways of implementing the methods in a control system based on the desired control inputs and example exemplary output functions.
[0086] To this end, embodiments of the control system of the present technology are implemented using a partial obstruction or flow limitation measure, such as a fuzzy flow limitation (FFL) measure, according to the following general method. Using such a measure is a method that uses a traditional flat index as a system control variable to modify some of the observations described above. The flow limitation measure FFL uses multiple characteristic pattern recognition to improve the sensitivity and specificity of the algorithm to flow limitation (or more characteristics: partial upper airway obstruction). In an embodiment, the general method with the flow limitation measure FFL includes some or all of the following steps:
[0087] 1. Monitor the patient's respiratory flow.
[0088] 2. Obtain the inhalation and exhalation waveforms from the flow signal.
[0089] 3. Compute features from each portion of the waveform or, where appropriate, from the combined waveforms. A variety of features may be considered to construct a pattern.
[0090] 4. Calculate the fuzzy input variables based on the original variables. This step constructs the mapping of the input variables to the new space. Data mining of appropriate patient groups allows good estimation of the parameter space.
[0091] 5. Combine the derived fuzzy variables using fuzzy logic to produce a fuzzy output. This step is formulated using a matrix that allows laymen to intuitively understand.
[0092] 6. Defuzzify the fuzzy output to produce a crisp value (real number) used by the pressure control algorithm.
[0093] 7. Combine fuzzy outputs from multiple sources.
[0094] 8. Adjust treatment pressure based on FFL results.
[0095] In accordance with the present technology, an example system maps input parameters that vary widely over a range of variables that can be easily interpreted by fuzzy logic principles. For example, a shape index such as a flatness index can be mapped to a fuzzy variable such as high flatness (HIGH_FLATTENING), normal flatness (NORMAL_FLATTENING), and low flatness (LOW_FLATTENING). Each of these variables takes values between 0 and 1, which is based on the flatness index value determined by a numerical function associated with the variable, and based on input data related to flatness determination such as respiratory airflow. Similarly, based on appropriate numerical functions and input data related to determining ventilation, other fuzzy variables can map ventilation measurements such as HIGH_FLATTENING, NORMAL_FLATTENING, and LOW_FLATTENING. Nevertheless, other fuzzy variables can also map respiratory duty cycle measurements such as inspiratory time (T i ) and the overall duration of the respiratory cycle (T tot ) ratio), such as T i -on-T tot _LOW, T i -on-T tot _NORMAL or T i -on-T tot_HIGH, etc. These variables may be further mapped to fuzzy outputs, which in turn may be used to generate control variables. The fuzzy outputs associated with the flow restriction measurement may be, for example, MILD, MODERATE, and SEVERE and may be based on other fuzzy variables, such as the examples previously mentioned. For example, the fuzzy logic controller may execute the scenarios represented by the examples, such as:
[0096] (a)IF(LOW_FLATTENING&LOW_VENTILATION)then flow-limitation is SEVERE.
[0097] (b)IF(HIGH_FLATTENING&T i -on-T tot _HIGH)then flow-limitation is MILD.
[0098] The output of the defuzzification, which can be used as a flow limitation measure, can be enforced in a range between 0 (non-flow limiting) and 1 (severe). Thanks to such a system, the input space is classified in non-linear sensing using reliable and explainable rules without any loss of real number precision.
[0099] In fact, the accuracy of such a flow limitation measurement, which represents the degree of obstruction on a continuous range from 0 to 1, can be appropriately performed to obtain a response pressure, or a flow control setting that is better suited to treat the situation detected by such a measurement. For example, without the need to utilize a threshold value, the threshold value is additionally compared to such a measurement value to determine whether a pressure adjustment should be made, and the measurement is more directly completed in the setting of a respiratory therapy device or the determination of a therapeutic setting for such a device. For example, the obstruction measurement value acts as a proportional function, which can be directly applied to the pressure adjustment amount or flow rate adjustment amount (such as measuring the * adjustment amount). The resulting adjustment amount can then be applied to the therapeutic setting of the respiratory measurement device. The "FFL" measurement value discussed in more detail with respect to the technology described in this article can be performed to act as such a proportional therapeutic setting function.
[0100] As in Figure 5As further shown in , in such a system, flow limitation can be detected and / or quantified by a method comprising (a) determining a shape index indicating a flow limitation mode from a measurement of respiratory flow, as shown in step 502; (b) determining a ventilation measurement or a respiratory duty cycle measurement from the respiratory flow measurement, as shown in step 504; and (c) deriving a flow limitation measurement as a function of the determined shape index and one or both of the determined ventilation measurement or the determined respiratory cycle measurement, as shown in step 506. In this way, fine adjustments to the treatment pressure can be made. For example, an increase in treatment flow limitation can be based on a situation in which a pressure change occurs if it is determined that both conditions are satisfied: (a) the shape index indicates the presence of an M-shaped breath in the respiratory airflow or the presence of a flat breath in the respiratory flow and (b) the ventilation measurement is sufficiently reduced to indicate less than normal ventilation. An additional increase in therapeutic flow limitation may be based on another scenario in which a pressure change occurs if it is determined that both (a) the shape index indicates a pressure at which an M-shaped breath is present in the respiratory airflow and (b) an increase in the duty cycle measurement is detected. In such a scenario, an appropriate therapeutic response may be provided without unnecessary therapeutic changes, which may occur when the shape index, ventilation measurements, or respiratory duty cycle measurements taken alone do not accurately predict the level of flow limitation.
[0101] Accordingly, in one embodiment, the system features the use of multiple input features (e.g., 4) for deriving the flow limitation measure FFL. The input features may include (1) a single breath flat (SBF) shape index, (2) an M-shape index (MS), (3) a ventilation measure such as ventilation ratio (VR), and / or (4) a ventilation measure such as inspiratory time to total breath time (T i -on-T tot ) ratio of the breathing duty cycle measurement.
[0102] The embodiment implementing the shape index as a single breath flatness index SBF is calculated similarly to the method disclosed in U.S. Patent No. 5,704,345, except that it is preferably performed without using a five breath moving average. The details of the implementation of the single breath flat shape index determination are disclosed in Section A herein.
[0103] Here, an example exemplary execution of the determination of the M-shape shape index is described in Section B. The calculation of the illustrated M-shape feature can vary from 0 (i.e., the inspiratory flow waveform is not M-shaped) to 1 (i.e., the inspiratory flow waveform is determined to be M-shaped). The M-shape index can optionally be augmented by another shape index such as a simple "chair shape" index. An exemplary determination of a chair shape index (e.g., "Fuzzy Chairness") is also described in detail in Section B.
[0104] Embodiments of determination of suitable ventilation measurements may be based on tidal volume calculations. For example, the ventilation measurement may be a ratio that includes the current tidal volume and the previous tidal volume, such that the measurement may indicate a change in ventilation or tidal volume. In one suitable embodiment, the measurement may be a ventilation ratio VR, which may be calculated as follows:
[0105] 1. Take the absolute value of the respiratory flow and filter it with a single low pass filter (time - duration = 3 minutes).
[0106] 2. Divide the filter output by 2 to produce "3 minute ventilation".
[0107] 3. Calculate the tidal volume of the breath (basically the average of the inspiratory and expiratory volumes).
[0108] 4. Divide the tidal volume by the length of the breath, giving the average respiratory flow rate, and then divide this value by the 3 minutes of ventilation, giving a measure of how the breath compares to recent ventilation.
[0109] Typically, VR will oscillate over a small range around 1. In severe upper airway obstruction events, it will drop to values much less than 1 and can drop to zero for total obstruction. Large breaths (recovery breaths on awakening) will have VRs much greater than 1 (e.g., 1.5-2). Additional details on the calculation of VR for specific ventilation ratios are included herein in Section G.
[0110] As previously described, the system may optionally utilize a respiratory duty cycle measurement derived from the flow limit measurement FFL. The measurement may be based on a ratio of the current and previous duty cycles, and thus may indicate a change in the respiratory duty cycle. Exemplary details for the calculation of the respiratory duty cycle ratio TTR are further described herein in Section G. For example, a suitable respiratory duty cycle measurement may be based on T i -on-T tot Calculation of variables. The variables chosen have the (unbelievable) limits of zero (non-inhalation) and one (non-exhalation). Because the cardinality T i -on-T tot It can vary between patients, T i -on-T totThe ratio to its most recent average value can be calculated, which has been designed here as T i -on-T tot T i -on-T tot The most recent average value of T can be obtained by adding each calculated i -on-T tot The value is sent to a simple low-pass filter to calculate. i -on-T tot As the relative recent history increases, the TTR will exceed 1. i -on-T tot has a value of approximately 0.4, while TTR will be a normal value and greater than 1 when the patient attempts to "compensate", ie increase their tidal volume in the face of an obstructed upper airway.
[0111] As mentioned above, these input features are mapped into multiple variables by numerical functions. Thus, the shape index SBF can be mapped into, for example, six fuzzy variables: B_LOW_FLATTENING, EXTRA_LOW_FLATTENING, VERY_LOW_FLATTENING, LOW_FLATTENING, NORMAL_FLATTENING, and HIGH_FLATTENING. These fuzzy memberships are shown by sample functions, which have been Figure 6 , and are shown as being joined to each of the variables listed above in order from left to right. As an example application of the membership functions, if the shape index SBF is calculated and produces a value of 0.05, when applied to each membership function, the variable B_LOW_FLATTENING produces a value of 1.0, and for Figure 6 All other fuzzy variables in are zero. Other suitable functions and variables can be selected as required.
[0112] Similarly, the exemplary membership functions and variables calculated based on the M-shape index are Figure 7 . The graph represents the appropriate functions for LOW_M-SHAPE, NORMAL_M-SHAPE, HIGH_M-SHAPE, VERY_HIGH_M-SHAPE, and B_HIGH_M-SHAPE. Exemplary membership functions and variables calculated based on ventilation ratio index are shown in Figure 8. This figure shows the appropriate functions for VERY_LOW_VENTILATION, LOW_VENTILATION, NORMAL_VENTILATION, HIGH_VENTILATION, VERY_HIGH_VENTILATION. Finally, the example membership functions and variables calculated based on the respiratory duty cycle ratio are shown in Fig. 9 This graph shows the i -on-T tot Low (T i -on-T tot _LOW)、T i -on-T tot Normal (T i -on-T tot _NORMAL) or T i -on-T tot _High(T i -on-T tot _HIGH) and T i -on-T tot _Very_High(T i -on-T tot _VERY_HIGH) and T i -on-T tot Special_High(TT i -on-T tot _EXTRA_HIGH).
[0113] Based on some or all of these mapped variables, the system then applies or estimates rules based on the combination of variables in the flow limit measurement FFL derivation. For example, based on the fuzzy variables identified previously, the fuzzy rules have been shown in the table below. These matrices also express a view of what constitutes a partial blockage in natural language.
[0114] Table A
[0115] Flat and Ventilated Measurements
[0116]
[0117] Table A summarizes a set of fuzzy rules that are applied by fuzzy variables related to single-breath flat shape index and ventilation ratio measurement. For example, the rules shown in bold ("M to M", i.e., mild to moderate) in the "Very High" column of the fuzzy variable for the flat shape index and in the "Low" row of the fuzzy variable for the ventilation ratio measurement represent fuzzy rules:
[0118] if(VERY_HIGH_FLATTENING AND LOW_VENTILATION)then FFL is "Mild-to-Moderate".
[0119] This and other rules represented in the table can be evaluated, where "AND" is fuzzy-AND. The usual output results for each output response (e.g., "mild to moderate") can then be fuzzy-ORed together. For example, two rules with a "moderate" output (e.g., a measure of moderate flow restriction) can be fuzzy-ORed to give a final "moderate" output. This is equivalent to the following formula:
[0120] moderate=(VERY_HIGH_FLATTENING AND VERY_LOW_VENTILATION)OR(EXTRA_HIGH_FLATTENING AND LOW_VENTILATION)
[0121] Based on the application of these rules in the derivation of flow limitation measures, at least three things are apparent in their implementation: 1) as flatness intensity increases (e.g., shape index decreases), the FFL measure increases, 2) where flatness is occurring and ventilation is reduced, the FFL measure is more severe, and 3) where ventilation is high (e.g., recovery (large) breaths, etc. are occurring), the response to the flatness shape index is tempered.
[0122] There are situations where flow limitation (at least initially) does not include a reduction in tidal volume (e.g., ventilation ratio). In these situations, the patient maintains his or her tidal volume by increasing the duty cycle, i.e., by extending the inspiratory time as a percentage of the total inspiratory-expiratory time. This trend can be measured or determined using a duty cycle measurement (e.g., TTR), which in this embodiment will increase beyond 1 as the inspiratory time lengthens.
[0123] The system further combines the Flat Shape Index information with a respiratory duty cycle measurement (eg, TTR) in a manner similar to that in which the Flat Shape Index is combined with a ventilation measurement (eg, VR) using the rules represented in Table B below.
[0124] Table B
[0125] Flat and breathing duty cycle measurements
[0126]
[0127] Like Table A, Table B represents fuzzy rules and corresponding natural language descriptions of flow obstruction based on respiratory duty cycle measurements and flat shape index. For example, the rules shown in bold ("M to S", i.e., moderate to severe) in the "Extremely High" column and "Very High" row are fuzzy rules:
[0128] if(EXTRA_HIGH_FLATTENING AND T i -on-T tot _VERY_HIGH)then FFL isModerate-to-Severe.
[0129] This rule represents the consideration that if the waveform of the inspiratory breathing pattern tends to be severely flat and the inspiration is considered to be moderately prolonged by the duty cycle measurement, then the flow limitation measurement FFL is moderate to severe. For those involved in Table A, the rule results for the general output function of this table can be combined using a fuzzy-OR operation.
[0130] In a manner similar to the method of deriving a flow limitation measure using flat shape index information and one or both of a ventilation measure and a respiratory duty cycle measure, flow limitation may also be derived by using both an M-shape and / or an augmented M-shape detection index and one or both of a ventilation measure or a respiratory duty cycle measure. This derivation aspect is illustrated in Tables C and D.
[0131] Table C
[0132] M-shape or enlarged M-shape and ventilation measurements
[0133]
[0134] Table C represents rules similar to the previous chart rules, except that they involve fuzzy variables based on the M-shape index and / or the chair shape index. As the nature of the output function selection in the table makes clear, some of the rules will prevent response to a detected M-shaped breathing pattern where the patient has normal ventilation or above average ventilation. Therefore, flow limitation measurements are derived so that the system will not respond to certain types of awake breathing and "behavioral" M-shaped breathing patterns, such as those that occur in REM sleep but do not indicate current flow limitation. For example, the rule represented by the entry in the top row of the last column of Table C indicates that the output function is "zero" for a bad "very high" M-shape index related measurement when there is a good ventilation measurement (e.g., "very high").
[0135] Furthermore, most M-shaped breaths represent moderate to severe obstruction and will indicate decreased ventilation. Therefore, the other rules of Table C allow for flow limitation detection for system pressure changes and thus querying the location of detected M-shaped breaths and hypopnea measures such as decreases in ventilation.
[0136] For those involved in Table A, the rule results for the general output functions of this table can be combined using a fuzzy-OR operation.
[0137] Table D shows the rules involving breathing duty cycle measurements.
[0138] Table D
[0139] M-shape or increased M-shape and duty cycle measurements
[0140]
[0141] Table D represents specific fuzzy rules and the corresponding natural language description of flow obstruction or flow restriction derived from respiratory duty cycle measurements and M-shaped and / or augmented M-shaped determinations. In this example, the matrix "combines" the M-shaped information and T i -on-T tot (rather than ventilation ratio). In the case of an increasing trend in the breath duty cycle, the matrix can be considered to relate to breaths that exhibit an M-family, or M-shaped pattern with an angle of presence. This combination identifies flow limitation in breaths where the patient's ventilation is approximately "normal" and the patient "compensates" by taking longer inspirations.
[0142] For those involved in Table A, the rule results for the general output functions of this table can be combined using a fuzzy-OR operation.
[0143] After applying the rule results based on tables A, B, C, and D, the results can be applied to the output function. Fig.10 The appropriate fuzzy output membership functions applied by the results are shown. The output membership functions include negative (NEGATIVE), zero (ZERO), mild (MILD), mild to moderate (MILD_TO_MODERATE), moderate (MODERATE), moderate to severe (MODERATE_TO_SEVERE) and severe (SEVERE).
[0144] As an example, consider Table D. Once a single fuzzy rule in the table is evaluated (e.g., if (EXTRA_HIGH_MSHAPE AND T i -on-T tot_VERY_HIGH) then FFL is Moderate-to-Severe), collect the outputs and fuzzy-OR them together. For example, Table D has three Moderate-to-Severe outputs that need to be fuzzy-ORed. Once this is done for all outputs, they are provided to Fig.10 The defuzzification function in Fig.11 The centroid method can be used to provide an unambiguous output. This produces a real number between 0 and 1.25. Once a single output from all tables is available, the maximum output can be used for the FFL value.
[0145] In determining a single "clear" measurement value in the defuzzification step just mentioned, a centroid method or other such defuzzification operation may be performed. In the centroid method, the clear value of the output variable is determined by finding the value associated with the centroid of the values for the output function. An example of this method is Fig.11 Shown in. Fig.11 The application of the centroid method for calculating fuzzy input is shown, where MILD = 0.1, MILD_TO_MODERATE = 0.5, MODERATE = 0.8. The calculated result is 0.61064. In view of this description, those skilled in the art will understand that other ways can calculate the unambiguous measurement value.
[0146] In one embodiment of the system, a flow limitation measure may be derived so as to avoid treatment of other conditions that may be detected. For example, it is possible for an awake breath to be flat in shape or flat in M-shape, thus potentially indicating flow limitation as determined by the shape index, but the breath is actually a prolonged inspiratory time and thus would not indicate actual flow limitation in the patient. Therefore, a flow limitation measure may be derived so as not to treat such flow limited breaths. One way to do this is to modify the fuzzy outputs of the fuzzy rules from Tables A, B, C, and D to adjust the following combination, which uses a formula having Fig.12 Essentially, the fuzzy outputs generated by the rules based on the combination of duty cycle measurements TTR (such as the rules in Tables B and D) are combined with Fig.12 The output of the function is multiplied by . Therefore, if the ventilation ratio is approximately 1 or less, the result of each special rule remains unchanged. As the ventilation ratio measure VR increases, it indicates the possibility of increased arousal breathing. The output of the rules in Tables B and D is based on Fig.12 The example function of gradually unweights and multiplies until the ventilation ratio measurement VR reaches 1.5. At this point, the result of the multiplication will cause the output of the affected rule to be zero.
[0147] Based on the establishment of the previous components (e.g., shape index, fuzzy membership functions and variables, fuzzy rules, etc.), the calculation of the flow limitation measurement FFL embodiment in the flow limitation detection system can be further summarized by the following exemplary steps:
[0148] 1. Respiration can be framed in a typical manner from a patient flow signal to distinguish and extract data representing the inspiratory and expiratory waveforms from the flow signal.
[0149] 2. The inspiratory waveform can be selectively adjusted to any lead-in pause using an adjustment method that takes into account the special M-shape. This method is described in Section F herein.
[0150] 3. M-shape and fuzzy-chair determination can be performed based on the exemplary calculations shown in Part B.
[0151] 4. Optionally calculate the augmented M-shape as the fuzzy-OR of the M-shape shape index and the chair shape index (eg, fuzzy chair).
[0152] 5. The filtered single breath flat (SBF) shape index can be determined based on the exemplary calculations shown in Part A.
[0153] 6. The ventilation ratio (VR) and T can be determined based on the exemplary calculations shown in Section G. i -on-T tot Ratio (TTR).
[0154] 7. Fuzzify SBF, VR, TTR and augmented M-shaped into variables according to the exemplary membership functions described previously.
[0155] 8. Fuzzy rules are then applied according to the matrices of Tables A, B, C and / or D described previously.
[0156] 9. You can Fig.12 The previously described "limited awake breathing" can optionally modify the fuzzy rules from Tables B and D, which use the respiratory duty cycle measurement TTR.
[0157] 10. Each fuzzy rule matrix is collected and defuzzified as described in detail above.
[0158] 11. This produces the following fuzzy outputs which are all fuzzy-ORed to give a flow limit measurement FFL such as:
[0159] FFL=fuzzy-OR(SBF-VR,SBF-TTR,M-shape-VR,M-shape-TTR)
[0160] Where:
[0161] SBF-VR is the deffuzzified result based on
[0162] the rules of table A;
[0163] SBF-TTR is the deffuzzified result based on
[0164] the rules of table B;
[0165] M-shape-VR is the deffuzzified result based
[0166] on the rules of table C; and
[0167] M-shape-TTR is the deffuzzified result based
[0168] on the rules of table D;
[0169] 12. Next, the fuzzy continuous flat measurement (FPF) is arbitrarily calculated and fuzzy-ORed from the flow limit measurement FFL by the following equation. An exemplary calculation of the measurement FPF is described in Section C herein. The adjustment of the FFL measurement may be performed by the following equation:
[0170] FFL=fuzzy-OR(FFL,FPF)
[0171] 13. The FFL is now modified based on the degree to which the breathing frame (e.g., detection of an inspiratory or expiratory waveform) may be erroneous, such as using the calculation described herein in Section D. The factor causing the incorrect breathing frame can therefore be multiplied by the FFL by the following equation:
[0172] FFL=FFL*Bad-breath-framing-factor
[0173] 14. The final FFL value may optionally be used for a circular buffer of a certain length, such as 3, and the last FFL value used by the pressure setting algorithm is a running average based on the buffer's most current FFL values. A suitable equation for this operation is:
[0174]
[0175] Fig.13A graphical comparison between a conventional flatness index and a derived flow limitation measure FFL according to the steps outlined above is provided. The index and the measure are based on the same flow signal shown in the bottom trace. The bottom trace flow signal includes a series of obstructed breaths at the end of arousal, followed by a certain amount of normal recovery breathing, followed by more obstruction. The breathing is initially "flat" and then gradually becomes M-shaped. In the top trace, the conventional flatness index is initially falling, then rises and finally drops sharply during the recovery breathing process. Due to both the flatness of the breathing and the M-shaped breathing, the flow limitation measure FFL of the middle trace rises steadily and then drops to zero as the breathing in the flow signal becomes more normal.
[0176] The flow restriction measurement may also be implemented in the control of the pressure therapy device. In one such embodiment, the measurement may be implemented using any one or more of the following steps:
[0177] 1. Measure the flow rate at the flow generator (FG).
[0178] 2. Measure the pressure at FG.
[0179] 3. Calculate the pressure drop between the FG and the mask using the flow rate from step 1.
[0180] 4. Calculate the pressure at the mask by subtracting the pressure drop from the pressure at FG calculated in step 3.
[0181] 5. Using the pressure at the mask, calculate the flow through the mask outlet (sometimes called the intentional leak).
[0182] 6. Subtract the outlet flow from the flow measured at FG to give the total patient flow (respiratory flow) plus any unintentional (mask or mouth) leaks.
[0183] 7. Filter the signal from step 6 to extract the DC (unintentional leakage) component.
[0184] 8. Subtract the DC component calculated in the last step from the flow calculated in step 6 to give the patient flow (respiratory flow).
[0185] 9. Patient flow is easily filtered here to remove unwanted higher frequencies, such as that shown in Section H.
[0186] 10. Construct the breath in the usual way using the patient flow.
[0187] 11. Adjust any lead pause here from before inspiration using a method such as that shown in Section F.
[0188] 12. Once a complete breath (inhale + exhale) has been completed calculate the following characteristics:
[0189] a. The length of any apnea preceding a respiration.
[0190] b. Inspiratory snoring index (such as the average value of the snoring signal during the currently analyzed inspiratory process).
[0191] c. Here, such as by the method shown in Part A, the value of the filtered single-breath flatness factor (SBF) for the current analysis breath.
[0192] d. The value of the M-shape index for the current inhalation and the value of the blurred chair index for the current inhalation, such as by the method described in Section B. The values of the M-shape and the blurred chair index are blurred or processed to give an enhanced M-shape feature.
[0193] e. Here, for example, by the method described in Section G, the ventilation ratio (VR) and T of the current analyzed breath are i -on-T tot Ratio (TTR) value.
[0194] f. Calculate FFL using SBF, VR, TTR and increased M as described in detail above.
[0195] g. Using the exhaled breath currently being analyzed, calculate the valve-like leak ratio.
[0196] h. Unintentional leak is measured at the end of the current analyzed exhalation.
[0197] i. Find the current setting for mask pressure (ie EPAP).
[0198] j. Here, the normalized peak exhalation position NEPL is calculated using the current analyzed exhalation, such as by the method described in Section K.
[0199] k. Find the recent peak interference value.
[0200] 13. Calculate the de-weighting factors according to the pseudo code below. Section J describes the example individual de-weighting functions utilized.
[0201] a.deweight=1.0
[0202] b.deweight*=FFL_function_of_Leak(leak)
[0203] c.valve-like leak*=
[0204] valve_like_leak_function_of_leak(leak)
[0205] d.deweight*=FFL_function_of_valve_like_leak (valve-like leak)
[0206] e.deweight*=FFL_function_of_pressure(EPAP)
[0207] f.deweight*=FFL_function_of_NEPL(NEPL)
[0208] g.deweight*=FFL_function_of_Jamming(jamming)
[0209] 14. The current threshold value (current_crit_FFL) required for pressure increase is then calculated using the following formula:
[0210] Current_crit_FFL=1.0-deweight*(1.0-crit_FFL);
[0211] 15. The normalized value of unmodified crit_FFL is 0.05. Therefore, if all deweighting is applied, deweight = 1.0 and current_crit_FFL = 0.05. Alternatively, if deweight = 0.0, then current_crit_FFL = 1.0.
[0212] 16. The pressure rise values associated with the FFL (FFL "module") can now be calculated and "prescribed":
[0213]
[0214] The final step is a simple decay of the pressure exponentially over the time of the breath, which preferably has a duration of approximately 20 minutes.
[0215] 17. The snoring-dependent pressure rise can now be calculated and "prescribed" and the individual de-weighting functions used as shown in this section J:
[0216]
[0217] 18. The pressure rise due to apnea can now be calculated:
[0218]
[0219]
[0220] 19. The new EPAP setting can then be calculated:
[0221] a.EPAP= EPAP_range_min +apnoea_prescription+snore_prescription+FFL-prescription
[0222] b. The new EPAP setting is then accomplished by increasing the FG pressure in such a way that the mask pressure approaches the new EPAP value at the maximum flow rate of 1.0 cmH2O per second. Again, the therapy pressure is preferably increased only when the patient is inhaling.
[0223] In another embodiment of the present technology, the flow limit measurement FFL can be optionally performed as part of a respiratory effort related arousal (RERA) detector. In 1999, the AASM Task Force defined RERAs as:
[0224] "A series of breaths characterized by an increase in respiratory effort that results in arousal from sleep but that does not meet the criteria for apnea or hypopnea. These events must meet both of the following criteria:
[0225] 1. A pattern of increasing esophageal negative pressure, terminated by a sudden change in pressure to a less negative level and arousal
[0226] 2. The event lasts 10 seconds or longer. "
[0227] In 2000, a study conducted at the NYU School of Medicine and published in Sleep, vol. 23, No. 6, p / 763-771, “Non-Invasive Detection of Respiratory Effort-Related Arousals (RERAs) by a Nasal Cannula / Pressure Transducer System” demonstrated that the Nasal Cannula / Pressure Transducer System was appropriate and reliable in detecting RERAs.
[0228] By utilizing the techniques described herein, a RERA detector can be based on a true flow signal originating from a flow generator. For example, a flow limitation measure by any of the methods previously described can be determined based on the flow signal. An arousal measure is then derived as a function of the flow limitation measure and another function of a measure of increase in ventilation.
[0229] Therefore, in one embodiment, the RERA detector may be based on the following techniques:
[0230] If there has been recent flow limitation (eg, FFL greater than (>) 0) followed by a change in ventilation step (eg, large breaths), then RERA is tested.
[0231] Preferably, the measurement is performed as a continuous variable, so the threshold is adjusted based on experimental data. This is an alternative to Boolean thresholds on each input parameter that causes loss of information. The following algorithm can be used:
[0232] 1. Keep track of the three most recent FFL values in the rotation buffer.
[0233] 2. Track the three most recent ventilation ratios (VR is the ratio of the average tidal volume to the current three-minute ventilation).
[0234] 3. Sum the three most recent FFL values and clamp the result to the range [0.0:1.0].
[0235] 4. Calculate the difference between the two most recent VRs (i.e. VR n -VR n-1 and VR n -VR n-2 ).
[0236] 5. Extract the maximum value from step 4 and limit it to a value greater than or equal to zero.
[0237] 6. Multiply the result of step 3 by the result of step 5 to give the RERA result.
[0238] 7. Normalize the result of step 6 by taking the square root of the result.
[0239] 8. A threshold may be set, wherein if the result of step 7 exceeds the threshold, RERA is counted.
[0240] 9. Between breaths, there should be breath(s) with a RERA count below the threshold, which are counted as RERAs.
[0241] 10. The buffer size is chosen to be three because in the extreme case of UARS there are only three detectable breaths between arousals.
[0242] like Fig.14 As shown, the bottom signal represents continuous RERA detector measurements based on this method. The bottom trace shows the detection of respiratory effort related micro-arousals based on the arousals of the top trace representing the flow signal.
[0243] As part of the respiratory disturbance index, a RERA detector is further implemented. The RERA detector is further used in the diagnostic mode to calculate the RDI (Respiratory Disturbance Index), such as by the equation:
[0244] RDI=RERAs+Apnoeas+Hypopnoeas per hour
[0245] Alternatively, RERAs may be reported by the treatment device as an indication of treatment effectiveness. Finally, the RERA index described herein may be used as an input into a treatment algorithm. For example, the RERA index may be determined over a certain time frame, such as every hour, and the result may be used in an "external loop controller" to set a threshold (or gain) for increasing pressure for a flow restriction index. An example of an external loop controller is described in WO2005 / 051470 (PCT / AU2004 / 001652) filed by ResMed Ltd., the disclosure of which is incorporated herein by reference.
[0246] The ratio of peak to mean expiratory flow may be used as another type of sleep-wake-awake detection index.The ratio of peak expiratory flow to mean expiratory flow may be part of sleep-wake / awake detection. Fig.15 Included is a histogram comparing the awake normality to the two sleep data sets. Awake breathing tends to be "flatter", i.e. the peak is closer to the mean. By comparing a threshold value such as 0.5 to the ratio, a RERA can be considered detected.
[0247] Part A - Single Breath Flat
[0248] An embodiment of a single breath flatness index may be calculated by the following method, as follows:
[0249] 1. Structure your breathing.
[0250] 2. Extract the inhalation portion of your breath.
[0251] 3. Adjust any tracking or lead pauses needed.
[0252] 4. The resulting aspiration is interpolated onto a standard grid of N points (usually N=65).
[0253] 5. Divide the point y value by a factor to normalize the breathing area to 1 with unit baseline length.
[0254] 6. Calculate the rms (root mean square) value of the difference between the middle half of the points starting from 1.
[0255] The flow signal used to calculate the flatness may be referred to as (among other things) "patient flow". The flow signal may be the result of filtering the raw flow signal with a filter, such as a "10 Hz" filter, however the filter type may vary. Preferably the filter reduces unwanted signal inclusions, such as those caused by eddy currents and snoring. The filter may slightly attenuate cardiac inclusions. The filter is alternating between accurately detecting zero crossings on the one hand, and rejecting unwanted signals on the other hand. For example, if the filter were more aggressive in rejecting all cardiac oscillations, then breathing detection would be discarded as inaccurate. However, once breathing detection is complete, further filtering may occur as desired.
[0256] Arbitrarily, the filtered single breath flat can be determined by:
[0257] 1. Filter the raw flow in the normal way to give the patient flow.
[0258] 2. Patient flow is used to construct a normal breath.
[0259] 3. The patient flow is continuously fed to the FIR filter to further reduce the unwanted signal components.
[0260] 4. Because FIR filters have a constant phase delay with frequency, we can pick the point required for the flat calculation by simply counting the filter delay.
[0261] 5. Calculate flatness in the normal way.
[0262] Fig.16 Two overlapping flow signals are shown recording a series of obstructive breaths, where the usual calculation of flatness may be corrupted by (in this case) low frequency snoring. One signal has been filtered (and delay compensated) so that snoring is not apparent. The filtered signal is given a suitably low flatness value calculated on the basis of the signal's respirations, in which snoring is apparent when compared to the calculation based on the other signal.
[0263] exist Fig.16 In the illustrated signal, the last breath has a conventional flat value calculated as follows: 0.21 (based on the non-filtered signal) and 0.13 (based on the filtered signal). The latter value indicates a desired pressure rise, while the former does not. As an alternative to this filtering, a conventional five-breath point-wise average can also achieve this filtered result, but at the expense of a major response delay.
[0264] The typical (and not computationally intensive) filter used is the boxcar FIR filter. A boxcar filter of length 12 has Fig.17 The impulse response of the filter is not critical if the signal in the middle half of each inhalation is predominantly involved.
[0265] Part B - M-Shape Index and Enlarged M-Shape
[0266] A suitable method for determining an obstruction measure such as an M-shape index may be accomplished using the following algorithm. The index detects the presence of an "M" shaped breathing pattern. Such an index is also generally considered to be an indicator of a "u" shaped breathing pattern. The obstruction measure may also be augmented or modified by additional obstruction measures. For example, as described in more detail herein, an index of obstruction is derived as a function of a first obstruction measure, such as an "M" shape index, and as a function of a second obstruction measure, such as an "h" shape index or a chair index.
[0267] For example, in some embodiments, each breath may be interpolated on an N-point grid, such as N = 65. In this embodiment, the two basis functions are calculated as follows:
[0268] t=i / (N-1) where i takes values from 0 to N-1.
[0269] B1=sin(π t)
[0270] B2=sin(3π t)
[0271] These basis functions can then be stored for use in all subsequent calculations of M-shape indices.
[0272] Each inhalation is then extracted and interpolated on an N-point grid. Two factors are then calculated as follows:
[0273] F1= sum (B1·fs)
[0274] F2= sum(B2·fs)
[0275] where fs represents the interpolated inspiration point and · is the dot product operation.
[0276] The final shape value is obtained by normalizing as follows:
[0277]
[0278] The shape factor is then constrained to vary between 0 (purely sinusoidal) to 1 (perfectly M-shaped). Fig.18 Appropriate functions are plotted to serve as basis vectors B1 & B2.
[0279] exist Fig.19 The flow signal for a typical M-shaped breath is plotted in . Based on the above method, the calculation of the breath used for plotting is as follows:
[0280] F1=4.6082
[0281] F2=2.6538
[0282] shape index = 0.50
[0283] Typical non-flow limited breaths have an M-shape index of only about 0.2.
[0284] As mentioned earlier, you can use the simple chair pose to increase the M-shape breathing. Fig. 20 This is illustrated in , which plots a flow signal with a flow restriction that produces an enlarged shape. Fig. 20 The flow-limited inspiration shown in (from t=0 to about t=2 seconds) has a typical "chair" shape. This inspiration can be characterized by two features: 1) a high ratio of peak to average inspiratory flow and 2) a flow rate close to zero (e.g. Fig. 20 In one embodiment, these indices can be calculated by:
[0285] Normalized inspiratory peak position (NormPeakLoc):
[0286]
[0287] in:
[0288] t0 is the time when inspiration begins;
[0289] t end It is the time at the end of the inhalation;
[0290] t peak is the time at peak suction flow rate;
[0291] Ratio of peak to mean inspiratory flow (RPMIF):
[0292] RPMIF=Q peak / Q'
[0293] in:
[0294] Q peak is the maximum flow rate during inspiration; and
[0295] Q' is the average flow rate during inspiration.
[0296] During quiet sleep, normal inspiration has a normalized peak position of approximately 0.5 and a peak-to-mean inspiratory flow ratio of 1.35. To measure "chair degree", based on Fig.21 and 22 The results of the above equations are used as mathematical functions for the respective graphs.
[0297] Once these fuzzy variables are calculated, the final fuzzy chair index is calculated as follows:
[0298] Fuzzy_Chairness=Fuzzy-AND(Fuzzy_Peak_to_Mean, Fuzzy_Peak_Loc).
[0299] Based on this determination, Fig. 20 The inhalation has a fuzzy chair degree of 0.82. The scaling of the fuzzy chair degree feature is such that it is fuzzy-ORed directly with the M-shape feature for the combined shape index. A fuzzy chair degree approximately greater than 0.3 suggests flow limitation. The fuzzy chair degree is described in English as: "If the ratio of peak inhalation to average inhalation is high and the location of the inhalation flow peak is close to the beginning or end of the inhalation, the inhalation is chair-shaped". Fig.23 , a typical flow sequence is shown, where the inspiratory shape is initially flat, so the SBF index is approximately less than 0.2. However, the shape becomes chair-shaped, so that the SBF index (SB flat) is approximately greater than 0.2, but the fuzzy chair or chair index is approximately greater than 0.3.
[0300] part C - Continuous Flat
[0301] While the example flow limitation measure FFL outlined above is designed to react in a real-time manner to flow limitations with both high sensitivity and specificity, there are situations in which it may not be able to prevent arousals due to increased upper airway resistance. Consider a patient with Fig.24 The trace shown, where the patient experiences upper airway resistance and increased respiratory effort, ends with arousal from sleep at t=420 (large breaths in the upper panel). Fig.24 The lower panel shows the single-breath flatness index.
[0302] Fig.25 This shows why the FFL flow limitation measure does not cause pressure to rise in this sequence. The patient maintains their tidal volume (as VR is approximately 1) without extending their inspiratory time as a function of total breath time (as TTR is approximately 1). Therefore, even though the single breath flatness index achieves a low value, the FFL flow limitation measure cannot reach the level required to adequately raise pressure.
[0303] To handle this situation, the system may optionally perform another obstruction measure. For example, a first obstruction measure is derived from filtering a second obstruction measure such as a flatness index. This may provide the instrument with some historical values associated with previous obstructions. Optionally, the historical values may be reset if a normal breath is completed (e.g., without obstruction), so that the historical obstruction measure or the filtered obstruction measure becomes an indicator of continuous or persistent past obstruction.
[0304] For example, a "fuzzy sustained flatness" may be performed. As the name implies, the fuzzy sustained flatness measurement essentially responds to consistently low values of the flatness index. The measurement may also be performed to respond slowly relative to the FFL flow limitation measurement so that it does not interfere with the flow limitation measurement FFL and does not over-treat the patient. Therefore, preferably the system filters the single breath flatness index in the following manner:
[0305] A simple first order autoregressive digital filter can be used as follows:
[0306] Y n =y n-1 +G(x n -y n-1 )
[0307] in:
[0308] G is the gain of the filter.
[0309] A time constant may be established that is twice as fast in length as five breaths. Thus, for example, initially if a breath is considered to be 4 seconds long:
[0310]
[0311] Taking into account the fact that a breath may not be four seconds long, a breath detection algorithm may be used to derive the current breathing rate (RR, breaths per minute), which may be determined as the average of the five most recent breaths detected. This may be performed as an adaptation time constant given below:
[0312]
[0313] Therefore, if the RR is 15 breaths per minute, it can be a normal RR and the time constant can be calculated as 10 as before.
[0314] The appropriate gain of the filter is simply given below:
[0315]
[0316] The value to be filtered is the (filtered) single breath flatness index. To prevent spurious values from interrupting the filtering process, the system may arbitrarily "head-limit" the incoming values. Thus, a value may be determined as the first-limited flatness (HLF) index as follows:
[0317] if(SBF index>0.3)then
[0318] HLF=0.3
[0319] else HLF=SBF index.
[0320] The output of the filter is a persistence flatness factor (PF), which can be initialized to an arbitrarily high value so that it initially has no therapeutic effect, such as: PF = 1
[0321] Another feature of the filter is that different gains are used for inputs smaller than the current PF value than for values larger than the current PF value, as follows:
[0322] if HLF <PF then
[0323] PF=PF+G(HLF-PF)
[0324] else
[0325] PF=PF+3*G(HLF-PF)
[0326] This asymmetry means that the filter will slow down to any sustained low input, but will reset relatively quickly. This helps improve the noise threshold so that only flat sustained low values are responded to in a time frame shorter than the flow limit measurement FFL.
[0327] Next, prevent the PF value from deviating from the reasonable value as follows:
[0328] if(PF>0.2)then
[0329] PF=0.2
[0330] else PF=PF
[0331] Finally, PF can be mapped to a fuzzy variable - Fuzzy Persistent Flatness (FPF) using, for example, the following equation:
[0332]
[0333] Fig.26 The FPF is shown Fig.24 The same sequence of breathing responses shown in .
[0334] Section D - Bad Breathing Structure
[0335] Sometimes, the breath detection algorithm may frame a breath inaccurately or the patient will cough or swallow causing a breath that provides little information about the current state of the airway. Optional heuristics may be used to de-weight breaths that are less likely to be relevant based on inspiratory time. Therefore, if a breathing pattern is detected that indicates an inaccurately framed breath, the derived obstruction measure may be adjusted. For example, Fig. 27 illustratively shows a function for a de-weighting factor that can be multiplied by a flow limit measurement (eg, FFL). i >2.5 seconds) or short (T i <0.7 seconds) were subjected to a progressive de-weighting operation.
[0336] Section E - Snoring Entropy
[0337] Traditionally, snoring has been measured using the Corrected Inspiratory Snoring Index, which can be considered a measure of obstruction. Fig.28 A typical snoring sequence captured while a patient is asleep is shown. The upper panel shows the respiratory flow rate, the middle panel shows the instantaneous snoring (as a measure of the acoustic power in the frequency range of interest) and the lower panel shows three snoring indices. The corrected inspiratory snoring index (in Fig.28 Snoring index) is calculated as follows:
[0338] Use the respiratory flow signal to frame inspiration
[0339] • Adjust the instantaneous snore signal taking into account the background noise due to the current situation (such as set pressure, turbine speed, etc.).
[0340] • Calculate the snore index as the average of the instantaneous snore signal over each inspiration (ie approximate the average acoustic power per inspiration).
[0341] This technique is reliable when used to treat snoring by increasing the mask pressure in proportion to the measured snoring index. However, the technique requires calibration constants to be measured for each flow generator mask combination and stored in the device's permanent memory. This calibration is time-consuming and therefore expensive, and is susceptible to time variations.
[0342] Matthew Alder et al. in PCT application number PCT / AU2007 / 000002 teach a method Fig.28The alternative, labeled delta snore shown in . Delta snore is calculated as the difference between the average of the instantaneous snore signal during inspiration and the average of the instantaneous snore signal during expiration. This is effectively a self-calibrated measurement, which assumes that background noise sources vary little between inspiration and expiration. However, this is generally not true.
[0343] In another alternative, an inspiratory snoring entropy method is performed. In one example, the obstruction measure is determined by filtering the respiratory flow measure within a frequency range associated with snoring (e.g., 30 to 300 Hz). The power or energy level of the filtered signal within the frequency range is then examined as a function of time to assess whether the power signal represents some inspiratory shape or is simply random noise. For example, the Shannon entropy function can be used. If the function indicates that the entropy signal is simply random information or noise, then the snoring index is adjusted or de-weighted since true snoring is not occurring.
[0344] Such methods use the information contained in the instantaneous snore signal during inspiration. Such indices also do not require calibration. For example, the obstruction measure is calculated as follows:
[0345] Collect a vector S of length n containing the instantaneous snoring value for the inspiration in question
[0346] If n < 2 or n > arbitrarily large, give up and return to zero
[0347] Subtract the floor of the snore vector and add 1:
[0348] S=S-min(S)+1
[0349] Calculation area A
[0350]
[0351] If A < any very small value, give up and return to zero
[0352] Normalized snoring vector
[0353]
[0354] Calculate the Shannon entropy (se) of the normalized snoring vector:
[0355]
[0356] Arbitrarily, the results were scaled to approximate the inspiratory snoring index as:
[0357] Inspiratory snoring entropy = 10.0*se
[0358] exist Fig.28 Both delta snore and snoring entropy are plotted in the lower panel of . As illustrated, the "snoring entropy" trace tends to obtain sharp snoring with low average snoring values (t=100). A judicious combination of both delta snore and snoring entropy is found to correctly classify snoring in almost all cases. Fig.29 The graph of shows one such arrangement. The classifier represented by the plotted line can be specified as follows: "If delta snore is positive, treat, and if delta snore is negative, treat with progressively higher snoring entropies that accompany negative delta snore".
[0359] Section F - Improved Tuning Guide Pause
[0360] Some pressure therapy devices contain a known "trim_leading_pause" function. This function is designed to trim ahead of the current framed inspiration to remove any wobbly expiratory pauses from the previous exhalation by detecting the peak and extrapolating backwards to the appropriate zero crossing that indicates the start of inspiration. Fig.30 As shown, this function fails for M-shaped breathing. Fig.30 Incorrect trimming (on the second peak) and correct trimming (on the leading peak) are shown.
[0361] The method can be modified so that a boundary is set above the found inspiratory peak, which contributes to M-shaped breathing. So, in the case of M-shaped breathing, the start of inspiration is determined based on a boundary within the inspiration, by inferring from the first peak of the inspiratory part of the breath instead of the second peak. For example, the boundary is calculated as follows:
[0362] 1. Get the sum of the inhaled air:
[0363]
[0364] 2. Find the inhalation t lim The points in are as follows:
[0365]
[0366] 3. Find the value in [t0:t lim ] Peak in inhalation
[0367] 4. Process the current trim_leading_pause algorithm
[0368] Fig.31 The plot shows a sequence of flow-limited breaths in a flow signal having a Fig.30 Lots of breathing shown. Fig.31The upper panel shows the flow rate plotted against time, and the lower panel shows the calculated inspiratory time (T) with trim_leading_pause applied. i ). The current algorithm and the new algorithm (“revised” T i ) The results of the two. It can be seen that the current algorithm cuts some breaths in half, forming false and incorrect low T i The new algorithm enforces 'trim_leading_pause' in a consistent manner.
[0369] part G - Ventilation and duty cycle measurements
[0370] (1) Calculation of ventilation ratio (VR)
[0371] A ventilation measure, such as the ventilation ratio (VR), can be determined as the ratio of the current breathing mode ventilation to the recent medium-term ventilation (V3). In this example, the medium-term can be a ventilation measure filtered using a filter with a three-minute time constant τ. However, other time constants are also suitable. The filter used can be a simple first-order autoregressive filter. Because the time constant of the filter is moderately large, the filter will take some time to rise from zero. Thus, over the time [t0:3×τ], the measurement will slowly switch between the moderate ventilation value and the filter output. VR can be calculated as follows:
[0372] 1. Set the filter gain:
[0373]
[0374] in:
[0375] f s is the sampling frequency;
[0376] τ is the time constant of the filter for a very short time.
[0377] For example, if the sampling frequency is 50 Hz and the time constant is 3 minutes, G can be calculated as follows:
[0378]
[0379] 2. Initialize the filter to a moderate ventilation value, such as 0.2 L / sec.
[0380] 3. Calculate the patient's respiratory flow (Q p ), which is the total flow minus the ventilation flow minus any mask leak, properly filtered.
[0381] 4. Calculate mid-term ventilation:
[0382] V3=V3+G(Q p -V3)
[0383] 5. During the conversion period 0 < t ≤ 3τ:
[0384]
[0385] 6. Frame breathing in the usual way.
[0386] 7. Once inspiration is determined, calculate as follows:
[0387] Inspiratory volume: V i
[0388] Inspiratory time: T i
[0389] Average inspiratory flow rate:
[0390]
[0391] Inspiratory component of the ventilation ratio:
[0392]
[0393] 8. Once expiration is determined, calculate as follows:
[0394] Expiratory volume: V e
[0395] Expiratory time: T e
[0396] Average expiratory flow rate:
[0397]
[0398] Expiratory component of the ventilation ratio:
[0399]
[0400] 9. Calculate VR as follows:
[0401]
[0402] (2) Calculate the duty cycle measurement value (such as TTR)
[0403] As described above, perform the duty cycle measurement value to derive the occlusion measurement value. For example, from the respiratory flow measurement value, a second ratio of the duration of the inspiratory part of the respiratory cycle to the duration of the respiratory cycle, as a function of the respiratory flow measurement value. Similarly, a second such measurement value can be determined, and the determination of the second such measurement value follows immediately after the first measurement value. Then derive the occlusion measurement value as a function of the first ratio and the second ratio.
[0404] It is possible to use such as T i-on-T tot The appropriate duty cycle measurement value of the ratio (TTR) is determined as the current (respiration) T i -on-T tot Value and recent medium-term T i -on-T tot The medium term can be determined by filtering at a five minute time constant or other suitable time constant. The filter can be a simple first order autoregressive filter. TTR can be calculated as follows:
[0405] 1. Set the filter gain to:
[0406]
[0407] in:
[0408] f s is the sampling frequency;
[0409] τ is the time constant of the filter for a very short time.
[0410] For example, if the sampling frequency is 0.25 Hz (a suitable breathing frequency) and the time constant is 5 minutes, G can be calculated as follows:
[0411]
[0412] 2. Initialize the filter to a moderate value such as 0.4.
[0413] 3. Calculate the patient's respiratory flow (Q p ), which is the total flow minus the ventilation flow minus any mask leak, properly filtered.
[0414] 4. Calculate mid-term T i -on-T tot as follows:
[0415] T i T tot(5) =T i T tot(5) +G(T i T tot -T i T tot(5) )
[0416] 5. Calculate TTR as follows:
[0417]
[0418] Partial H-deweighting function
[0419] Fig.32 and 33The function plotted in can be used to de-weight the effect of a flow limitation (such as FFL) measurement or a snoring measurement. The function can be applied to a threshold of an increase measurement (such as FFL or snoring) that needs to be exceeded for a pressure rise to occur. For example, the de-weighting function can include a valve-like leakage ratio. The valve-like leakage measurement can be calculated in a conventional manner so that its value varies between 0 and 5. In order to prevent a false valve-like leakage from preventing a pressure rise when in fact no leakage has occurred, the system can be executed to de-weight the valve-like leakage value to the extent that absolutely no leakage has occurred. Therefore, the algorithm for determining the de-weighting according to the valve-like leakage is as follows:
[0420] 1. Calculate the valve leakage from the most recent inspiration.
[0421] 2. Determine the absolute value of the leak (value at end exhalation).
[0422] 3. Multiply the value of valve leakage by Fig.32 Output of the absolute leakage function for the function illustrated in the left graph. For example, if the absolute leakage is less than (<) 0.025, the valve-like leakage is set to zero. Alternatively, if the absolute leakage is greater than (>) 0.05, the value of the valve-like leakage remains unchanged and between 0.025 and 0.05, the value of the valve-like leakage decreases linearly.
[0423] 4. Output the de-weighting factor using the value of the valve-like leakage with a leakage function such as Fig.32 The function shown in the figure on the right. For example, for values less than (<) 4, the output is 1, and for values greater than (>) 5, the output is 0. Between 4 and 5, the output decreases linearly from 1 to 0.
[0424] De-weighting of flow limit measurements (such as FFL) based on leakage (such as L / s) can be done using Fig.33 The function shown in the graph of is implemented. For leakage values less than (<) 0.5, there is no de-weighting and the output is 1. For leakage values greater than (>) 0.7, there is complete de-weighting and the output is 0. For leakage values between 0.5 and 0.6, the output decreases linearly from 1 to 0.
[0425] De-weighting of flow limitation measurements to pressure, such as mask CPAP levels in cmH2O, can also be done by using Fig.34 The function shown in the example curve of is completed. In the example, for pressure levels less than 10 cmH2O column, there is no de-weighting, and for pressures greater than 15 cmH2O column, there is partial de-weighting and the output is 0.8. For pressures between 10 and 15, the output decreases linearly from 1 to 0.8. The output is then multiplied by the flow limit measurement.
[0426] Finally, the effect of flow limitation or obstruction measurements with increased noise can be de-weighted. Noise is a measure of ambiguity such as how long the current inhalation or exhalation has lasted. For example, interference can be determined by the method described in U.S. Patent No. 6,484,719, the disclosure of which is incorporated herein by reference. High interference indicates brief changes in leakage, such as when a patient opens their mouth or when they move in bed and change the position of their mask on their face. At high levels of noise, it is possible that flow estimates are inaccurate and that reducing the leak constant can help improve the flow estimate. However, it will be prudent to prevent pressure increases until things have calmed down. This can be accomplished by Fig.35 The example function of is completed. When the interference reaches 0.25, the system starts to de-weight until the measurement is completely de-weighted at the 0.5 interference level. The output of the function can be multiplied by the measured value of the flow limit to complete the de-weighting.
[0427] Similarly, the snoring measurements are de-weighted based on various conditions of the system. For example, Fig.36 The snoring value is de-weighted by the pressure function shown. This makes the system less sensitive to the snoring measurements used to generate the pressure increase. Therefore, the snoring value required for the increase in treatment pressure increases as the pressure increases.
[0428] The inspiratory snoring index can also be de-weighted by ventilation measures such as the "big breath" measure. Big breaths usually only cause noise due to the high peak flows achieved. Fig.37 An example function of , ventilation can be used to generate a de-weighting output factor depending on whether the ventilation measurement is considered to be large breathing. For example, a ventilation measurement such as ventilation ratio (VR) can be used as a large breathing measurement. VR values greater than (>) 1 can be considered to be indicative of breathing, which is widely compared to mid-term ventilation. For VR values less than (<) 1.2, the system can refrain from de-weighting any snoring measurements, and for VR values greater than (>) 0.5, the system can completely de-weight the snoring effect. For VR values between 1.2 and 1.5, the output de-weighting factor can be linearly reduced from 1 to 0.
[0429] Other calculations may also be utilized to correct for the effects of flow limitation or snoring based on leak values, pressure, interference, and / or ventilation measurements.
[0430] Part I Normalized expiratory peak position
[0431] Normalized expiratory peak position (NEPL) is a better indicator of the conversion from a sleep or obstructed sleep flow waveform to indicate wakefulness, arousal or other unnatural, unusual or unrecognized events, such as severe mouth leaks. Therefore, the instrument of the present technology can perform a measurement of arousal based on the expiratory flow peak. For example, arousal is estimated based on the position of the expiratory flow peak or the normalized expiratory flow peak within the expiratory portion of the breathing cycle. In such an embodiment, the time of the expiratory portion can be limited by a range and the occurrence of the peak within the limited range defines an index representing arousal. If the expiratory peak occurs in the later time portion or within the expiratory portion of the breathing cycle, the index represents arousal.
[0432] Such an index can be calculated as follows:
[0433] 1. Structure your breathing.
[0434] 2. Isolate the exhalation portion of each breath.
[0435] 3. Locate the time when peak expiratory flow occurs.
[0436] 4. Divide the time from the start of exhalation to the peak by the total exhalation time, if the index is in the range [0.0:1.0].
[0437] Typically, when a patient is asleep, the NEPL is in the range of 0 to 0.3. For example, consider Fig.38 A histogram showing the comparison results of the three data sets corresponding to NEPL:
[0438] (1) Essen - OSA patients under treatment
[0439] (2) Coordination - Patient titration, and
[0440] (3) Awake - People breathing with the AutoSetSpirit airway pressure device from ResMed.
[0441] Fig.38 The right hand side graph shows that awake breathers have more breaths with values greater than (>) 0.5 than sleeping patients, who have fewer such breaths. Thus, a system can be implemented with a de-weighting function for unusual, unnatural or unexpected events related to sleep based on flow limitation measurements. Fig.39 A suitable function based on the calculated or determined normalized inspiratory peak position values is illustrated in the graph of The function is used to increase the strength of the flow limitation measure, which is required to cause an increase in treatment pressure as the breathing becomes more unnatural, unusual or unrecognized (e.g., bizarre).
[0442] In the foregoing description and the accompanying drawings, specific terms and figure marks are set forth to provide a thorough understanding of the present technology. In some instances, the terms and marks imply specific details that are not required to practice the technology. However, although the technology has been described herein with reference to specific embodiments, it is understood that those embodiments merely illustrate the principles and applications of the present technology. It is therefore understood that the illustrative embodiments should be extensively modified, and other arrangements may be made without departing from the spirit and scope of the present technology.
[0443] For example, the desired flow limitation detection and / or treatment system may be based on other techniques for pattern recognition, such as collecting each inspiration and interpolating it on a grid of multiple points needed to retain important frequency information. For example, each inspiration may be interpolated on a grid of 65 points. A large set of training data may be developed using different types of obstruction waveforms, which may be recorded based on clinical assessments. The waveforms may be pre-classified into types such as mild, moderate, severe, or other desired types. The 65 points of each waveform are then input into a classifier for training (such as a neural network, support vector machine or other) and using a suitable algorithm, such as one based on a genetic algorithm, arbitrary boundaries may be defined between the various obstruction classes RESMED 3.4-001CN D5 (2635)
[0444] Such a system would then be able to classify any inspiratory waveform that has been measured in a patient by the classifier metric and used in a control system for treating the patient.
[0445] Some potential issues with the approach are as follows:
[0446] 1. The “curse of dimensionality” refers to the exponential growth of hypervolume as a function of dimensionality. In other words, as the number of dimensions of our input space grows, the number of training vectors needed to “cover” that space grows exponentially. For the speed of this technique, covering 65 dimensions becomes a problem.
[0447] 2. Pre-classifying a waveform as obstructive (and to what degree) is difficult and can have high inter-observer variability. In the absence of increasing CPAP pressure to observe waveform changes such as "bunching", it is difficult to determine whether obstruction is truly present. Reliable effort measures, such as esophageal pressure, can help detect problems, but the problem of determining classification of such data is still intractable. For example, REM sleep can produce different and unexpected waveforms.
[0448] 3. The training phase requires a "valid" digital source and a large number of processor cycles.
[0449] 4. It would be difficult for anyone to interpret and numerically intensive to generate a classifier to run on an embedded system (depending on, for example, how many neurons we end up with if using a neural network).
[0450] 5. To test the generated classifier, we will need to feed it sleep studies of any human on Earth.
[0451] However, one way to reduce the complexity of the system is to limit the information fed into the system to only that which is "interesting." The calculation of specific features such as the flatness index does this; it acts as a form of signal compression.
Claims
1. An apparatus for automatically determining a measure of partial obstruction of a patient's airway for use in treating sleep disordered breathing, comprising: a sensor that determines a respiratory flow measurement; as well as A processor configured to: determining a first partial occlusion measurement based on the respiratory flow measurement; determining a second partial obstruction measure based on the respiratory flow measure; and A third partial blockage measure is determined as a function of the first partial blockage measure and the second partial blockage measure.
2. The apparatus according to claim 1, wherein: The first partial occlusion measure is an index indicative of an M-shaped breathing pattern.
3. The apparatus according to any one of claims 1 to 2, wherein: The second partial occlusion measure is an index indicative of a chair breathing pattern.
4. The apparatus according to any one of claims 1 to 2, wherein: The apparatus further comprises a flow generator, and wherein the processor is configured to control the flow generator to provide respiratory therapy.
5. A processor-readable storage medium comprising processor-executable instructions that, when executed, cause a processor to determine a measure of partial obstruction of an airway of a patient for use in treating sleep disordered breathing, the processor-executable instructions comprising: instructions for determining a first partial occlusion measurement based on a respiratory flow measurement from a sensor; as well as instructions to determine a second partial occlusion measurement based on the respiratory flow measurement; as well as Instructions for determining a third partial blockage measure as a function of the first partial blockage measure and the second partial blockage measure.
6. The processor-readable storage medium of claim 5, wherein: The first partial occlusion measure is an index indicative of an M-shaped breathing pattern.
7. The processor-readable storage medium according to any one of claims 5 to 6, wherein: The second partial occlusion measure is an index indicative of a chair breathing pattern.
8. The processor-readable storage medium of any one of claims 5 to 6, further comprising instructions for controlling a flow generator to provide respiratory therapy.
9. An apparatus for automatically determining the onset of inspiration for treating sleep-disordered breathing, comprising: a sensor that determines a respiratory flow measurement; as well as a processor coupled to the sensor, the processor being configured to: detecting a first peak according to the respiratory flow; and determining the onset of a breathing cycle as a function of said first peak; Therein, detection of the first peak is limited to avoid detection of a second peak in the breathing cycle.
10. The apparatus according to claim 9, wherein: The apparatus further comprises a flow generator, and wherein the processor is configured to control the flow generator to provide respiratory therapy in accordance with the determined initiation of a breathing cycle.
11. An apparatus for automatically determining a measure of partial obstruction of a patient's airway for use in treating sleep disordered breathing, comprising: a sensor that determines a respiratory flow measurement; as well as a processor coupled to the sensor, the processor being configured to: determining a partial obstruction measure based on the respiratory flow measure; determining an interference measurement from the respiratory flow, the interference measurement being a measurement indicative of temporal changes in leakage; and The partial blockage measure is adjusted as a function of the interference measure.
12. The apparatus according to claim 11, wherein: The partial occlusion measure is a function of the detection of M-shaped pattern breathing.
13. The apparatus according to any one of claims 11 to 12, wherein: The apparatus further comprises a flow generator, and wherein the processor is configured to control the flow generator to provide respiratory therapy.
