Devices for accurately classifying cardiac activity

ES 2 425 345 T3

DESCRIPTION

Devices to accurately classify heart activity.

Field

The present invention relates generally to implantable medical device systems that capture and analyze cardiac signals. More particularly, the present invention relates to implantable medical devices that capture cardiac signals within the body of a patient receiving the implantable device in order to classify cardiac activity as probably benign or probably malignant.

Background

Implantable cardiac devices typically pick up cardiac electrical signals in an implantable device recipient patient and classify the implantable device recipient's heart rate as normal / benign or malignant. Illustrative malignant rhythms can include ventricular fibrillation and / or ventricular tachyarrhythmias. The precision with which an implantable medical device analyzes captured signals determines its ability to carry out a therapeutic intervention and to make other decisions.

US 2006/0116595 describes templating methods for use in heart rhythm management devices in order to create a robust template to be compared to captured cardiac complexes.

New and / or alternative methods and devices for the analysis of cardiac signals are desirable.

Summary

The invention is defined by the characteristics of the claims. Some illustrative embodiments relate to the use of correlation analysis to identify cardiac event oversensing. In one example, a High-Low-High correlation pattern is searched for relative to a template. The template can be a static template, it can be a representation of a recently captured event, or it can be an average of several recently captured events. In another example, multiple frontiers are defined for the “High” correlation, where a higher first frontier (which needs a higher correlation) allows the identification of oversensing based on a smaller set of detected events than with a second one. lower border. In one embodiment, a shorter High-Low-High sequence is sufficient in the case of the first boundary, whereas a longer sequence of 5 or more (e.g. 8) alternating events is needed in the case of the second. border. In another embodiment, the High "and" Low "correlation definitions are tailored to particular signals using average values ​​for subsets of detected event correlations to establish boundaries.

In another embodiment, a correlation analysis is performed multiple times for a given template and each detected event shifting the alignment of the template and the detected event to maximize the correlation score of the analysis. Such an offset can adjust the alignment in one or two samples beyond the reference points identified for analysis. In another embodiment, the stored templates are modified to accommodate changes in morphology for selected stretches of the signal. In yet another embodiment, multiple signature and / or signal signature features are identified and multiple correlation scores are calculated using several different signature features as alignment points.

When identified, an over-detection can be corrected by modifying stored data in order to affect rhythm analysis. In such an embodiment, data correction is inhibited if the intervals surrounding probable over-detection are greater than a predetermined threshold. In some embodiments, the over-detection correction is inhibited if analysis of ranges related to probable over-detection indicates that a particular type of over-detection is unlikely. In such an embodiment, the intervals surrounding a probable over-detection are analyzed to determine if an accepted mathematical formula for estimating the expected QT intervals is satisfied and, if not, the method determines that the probable over-detection is not. a T wave, and therefore no data correction occurs.

Brief Description of Drawings

Figure 1 is a block diagram of an illustrative method for identifying over-detections and taking corrective actions;

Figure 2 shows an illustrative implantable cardiac pacing system;

Figure 3A shows an example that uses correlation analysis to identify over-detection;

Figure 3B shows the steps of a method for an illustrative example that includes rhythm correction;

Figure 4 shows an example of correlation comparisons between events;

ES 2 425 345 T3

Figure 5 shows another example of correlation comparisons between events;

Figure 6 shows an analytical strategy for short series and long series correlation analysis;

Figures 7A-7B illustrate examples of applying the analytical strategy of Figure 6 to series of correlation analyzes;

Figures 8A-8B illustrate examples of fitting correlation analysis to observed template correlation levels;

Figure 9 illustrates another method of aligning a captured signal to correlation analysis templates;

Figure 10 shows another method for storing and applying a template for correlation analysis;

Figures 11-12 illustrate a method for inhibiting an over-detection identification correlation analysis;

Figure 13 illustrates more methods for inhibiting an over-detection identification correlation analysis;

Figures 14A-14B show the application of a method illustrated in Figure 13;

Figure 15 shows an electrical shock analysis method for identifying detected shock events and treatable rhythms; Y

Figure 16 illustrates a method for calculating the correlation between a captured signal and a template.

Detailed description

The detailed description that follows should be read with reference to the drawings. The drawings, which are not necessarily to scale, describe illustrative embodiments and are not intended to limit the scope of the invention. Some of the examples and explanations that follow include references to issued patents and pending patent applications. These references are made for illustrative purposes and are not intended to limit the present invention to the particular methods or structures of those referenced patents and patent applications.

Unless required implicitly or explicitly stated, the methods that follow do not require any particular order of steps. It should be understood that when the examples that follow refer to a current event, in some applications this means that the most recently detected cardiac event is being analyzed. However, this is not necessarily the case, and some embodiments carry out analyzes that are delayed by one or more detection and / or by a fixed period of time. The choices shown regarding the use of rectified / non-rectified signals are merely illustrative, and can be changed if desired.

The nomenclature used herein indicates that a signal is picked up by an implantable cardiac device system, events are detected in the sensed signal, and cardiac activity is classified using sensed events (detections). Rhythm classification includes the identification of malignant rhythms, such as, for example, ventricular fibrillation or certain tachyarrhythmias. Implantable therapy systems make pacing / therapy decisions based on heart rhythm classification.

In an illustrative example, a detected event is detected by comparing the received signals to a detection threshold, which is defined by a detection profile. Any appropriate detection profile can be used. The detected events are separated into intervals. Different intervals can be used to generate an average interval over a selected number of intervals, from which the heart rate can be calculated. For example, four, eight, or sixteen intervals can be used to estimate the rate of cardiac events as a function of the average interval.

A cardiac electrogram includes different parts (usually called waves) that, according to a well-known convention, are labeled with letters including P, Q, R, S, and T, each of which corresponds to specific physiological events. It is very typical to design detection algorithms to capture the R wave, although any part of the cardiac cycle, if detected repeatedly, can be used to calculate the beat rhythm. If morphological (shape) analysis is used in addition to heart rhythm, the system can capture and / or analyze the part of the cycle that includes the Q, R and S waves, called the QRS complex. Other parts of the patient's cardiac cycle, such as the P wave and the T wave, are often treated as artifacts that are not searched for for the purpose of estimating heart rate, although this is not necessarily the case.

Typically, for the purpose of rhythm determination, each cardiac cycle is counted only once. Over-detection (such as double or triple detection) can occur if the device declares more than one detected event within a single cardiac cycle. Oversensing can occur if more than one part of a single cardiac cycle is detected, or if noise causes an event to be declared when no cardiac event has occurred, for example, due to external therapy or noise, to pacemaker artifact, skeletal muscle noise, electrotherapy, etc.

ES 2 425 345 T3

If a cardiac cycle occurs and a detection algorithm declares a multiple number of detected events, an over-detection has occurred. If the heart rate is then calculated by counting each of these detections, an overcount occurs. The calculated heart rates can be used alone or in combination with other factors to classify heart rhythms as malignant or benign. Overcounting that is dependent on over-sensed events can lead to erroneously high heart rate calculation. An erroneous calculation of the heart rate can lead to incorrect rhythm classification and incorrect treatment decisions. Some of these concepts are discussed and further developed in US Patent Application Serial No. under Serial Number 12 / 399,914, entitled METHODS AND DEVICES TO ACCURATELY CLASSIFY CARDIAC ACTIVITY, and in US Patent Application Serial Number 12 / 399,901, entitled PRECISE DETECTION OF CARDIAC EVENTS IN A DEVICE OF IMPLANTABLE CARDIAC STIMULATION.

Figure 1 is a process flow diagram of an illustrative method for identifying over-detections and taking corrective actions. The illustrated method begins with an event detection 10, where a received cardiac signal is captured and compared to a detection threshold until the received signal crosses the detection threshold, resulting in a declaration of a detected event.

Next, the method carries out an over-detection identification step 12. This may include one or more different analysis methods including, as illustratively shown, a morphological analysis 14, an interval analysis 16, and a QRS width analysis 18. After over-detection identification 12, if one or more over-detections are identified, the method corrects the data, as shown in step 20. If no data correction is needed in step 20, this step can be skipped.

Finally, the method includes a therapeutic decision, as shown in step 22. A therapeutic decision 22 can classify a heart rate of the recipient of an implanted device and determine whether and when therapy should be delivered. The method then iterates and returns to event detection.

The therapeutic decision 22 may include one or more of different forms of analysis. In an illustrative example, the individual sensed events are marked as shockable or non-shockable and an XY counter is updated to determine whether the overall heart rate requires therapy. The marking of individual events as shockable or non-shockable can take different forms, including rhythm-based and / or morphology-based determinations, or combinations thereof. Figure 15, at the bottom, provides an illustrative example. Additional examples are also discussed in US Pat. No. 6,754,528, entitled APPARATUS AND METHOD FOR DETECTING ARRHYTHMIAS IN A SUBCUTANEOUS IMPLANTABLE DEFIBRILLATOR / CARDIOVERSOR, and in US Patent No. 7,330,757 entitled METHOD TO DISCRIMINATE BETWEEN VENTRICULAR AND SUPRAVENTRICULAR ARRHYTHMIAS.

The therapeutic decision 22 may also take into account the persistence of a malignant condition. Some illustrative examples are shown in US Patent Application Publication No. 2006/0167503 entitled METHOD OF ADAPTING THE INITIAL CHARGE FOR AN IMPLANTABLE CARDIOVERTORS-DEFIBRILLATOR. Other methods may be used as part of the therapeutic decision 22.

Figure 2 shows an illustrative implantable medical device and implantation location. More particularly, an illustrative subcutaneous-only system is shown in Figure 2. The subcutaneous system is shown relative to a heart 40, and includes a hermetic receptacle 42 coupled to a lead wire 46. The hermetic receptacle 42 preferably houses a set of operational electrical circuits to carry out analysis of cardiac activity and to provide a therapeutic outlet. The operational electrical circuitry may include batteries, input / output electrical and electronic circuitry, power capacitors, a high-voltage charging module, a controller, memory, telemetry components, etc., as known in The technique.

The electrodes are disposed at locations throughout the system, including, for example, an electrode 44 in the watertight receptacle 42, and electrodes 48, 50, 52 in the drop lead 46. Electrodes 44, 48, 50, 52 can have any suitable shape and can be made of any suitable material. For example, the electrode 44 of the hermetic receptacle may be an insulated button electrode or it may be a region or surface of the hermetic receptacle 42, and the electrodes 48, 50, 52 on the drop cable 46 may be wire electrodes, electrodes of ring, or other structures known in the art.

Electrodes 44, 48, 50, 52 define a plurality of uptake vectors such as V1, V2, V3, and V4. If desired, one or more vectors V1, V2, V3 and V4 may be chosen as the default capture vector, as discussed for example in US Patent Application Publication No. 2007-0276445 entitled SYSTEMS AND METHODS TO SELECT A CAPTATION VECTOR IN AN IMPLANTABLE MEDICAL DEVICE. Other uses of multiple vectors are shown, for example, in US Pat. No. 7,392,085 entitled MULTIPLE ELECTRODE VECTORS FOR IMPLANTABLE CARDIAC TREATMENT DEVICES. Another embodiment considers stance in vector analysis, as discussed for example in US Patent Application Publication No. 2008-0188901 entitled VECTOR SELECTION OF

ES 2 425 345 T3

CAPTATION IN A CARDIAC STIMULATION DEVICE WITH POSTURAL EVALUATION. Multiple uptake vectors can be analyzed, sequentially! or in combination, as desired.

A therapy can be applied using a chosen pair of electrodes. An illustrative example uses the hermetic receptacle electrode 44 and wire electrode 52 to deliver therapy. Other combinations of electrodes can be used. A therapy may include defibrillation, cardioversion, and / or monophasic or multiphasic resynchronization pacing.

The present invention is not limited to any particular hardware, implantation site, or configuration. Rather, it is intended to be an improvement over any implantable cardiac therapy system. Some embodiments may also be used in a monitoring system to either control monitoring functions (including alarms and / or data logging) and / or to test the suitability of data analysis for a particular setting, condition, or patient.

Some illustrative examples may be associated with an external programmer 54 configured to communicate with the implanted device for various purposes, including, for example and without limitation, one or more of the following actions: device testing; upload new / revised software; modify programmable parameters such as detection or therapy settings; determine the operating status of a device, the life of the battery, or the integrity of the drop cable; allow or disable a functionality; and / or download data related to the condition of the recipient patient of the implanted device, capture previous data, or apply a treatment. Any appropriate communication method can be used, such as certain protocols and hardware widely known in the art.

Figure 2 omits different anatomical landmarks. The illustrative system shown can be implanted under the skin, outside the rib cage of the recipient patient of the implanted device. The location shown illustratively would place the watertight receptacle 42 approximately in the left axilla of the recipient patient of the implanted device, in a position level with the apex of the heart, such that the bypass lead 46 extends centrally into the xiphoid and then towards the head of the patient receiving the implanted device along the left side of the sternum. An illustrative example uses a method / system as shown in US Co-owned Patent Application Publication No. 2006-0122676 entitled APPARATUS AND METHOD FOR SUBCUTANEOUS ELECTRODE INSERTION. Other illustrative subcutaneous locations and systems are shown in commonly owned US patents No.^you 6,647,292, 6,721,597 and 7,149,575.

The present invention can also be carried out in systems having various implantation configurations including, for example, other subcutaneous only, vascular only, and / or transvenous configurations / locations. The hermetic receptacle 42 can be positioned in anterior, lateral, and / or posterior positions including, without limitation, axillary, pectoral, and sub-pectoral positions, and can also be positioned on both the left and right side of the torso and / or abdomen of the recipient of the implanted device. A completely intravascular implantation of the system has also been proposed. The hermetic receptacle 42 and bypass cable 46 can be positioned in any of a number of appropriate configurations including anterior-posterior combinations, anterior-only combinations, transvenous placement, or other vascular placements. A unitary system may omit the drop lead 46 and instead include all of the electrodes in the sealed receptacle 42.

Figure 3A shows an example that uses correlation analysis to identify over-detections. The term "correlation analysis" as used herein can take different forms. An illustrative example is shown in Figure 16. Referring to Figure 16, a captured signal 500 undergoes an analog-to-digital conversion 502 to produce a series of samples {S1 ... S9} time-ordered forms that form a sampled (and usually digital) representation of the signal, as indicated in diagram 504. The example in Figure 16 is simplified for the purpose of illustration since the number of samples for a given signal it can be greater than nine. For example, in an illustrative embodiment, the captured signal 500 has a duration of approximately 160 milliseconds, covering 41 samples captured at a frequency of 256 Hz. Other durations and / or sampling frequencies can be selected. The signal can be windowed with a window width approximately equal to the width of the QRS complex, although this is not necessary.

The digital representation of the signal is compared to a template using correlation analysis 506. The displayed template comprises a series of sample {T1 ... T9} values. Before the comparison, or as part of the comparison, the digital representation of the signal or template is scaled such that the highest value peaks of the two data series are equal in amplitude. An example of a correlation analysis is waveform correlation analysis. Other examples are well known in the art.

A simple version of the correlation analysis is shown graphically in Figure 16: the highest valued sample or peak of the digital representation of the signal is aligned with the peak of the template and the surrounding samples are compared to each other as shown. shown in diagram 508. Because the peaks are already scaled to the same value, there is no difference in the peak, but the surrounding samples may be different. The hatched areas show the differences between the representation of the signal and the template.

ES 2 425 345 T3

A correlation score can then be calculated as shown in Table 510. The sum of the absolute values ​​of the differences between the samples (scaled) of the signal representation and the template samples is calculated and divided by the total area under the template. The quotient is subtracted from unity, providing a 512 correlation score. If the correlation score is close to unity, then the area of ​​the difference is small relative to the area under the template, indicating high correlation. Other methods for calculating the correlation are known in the art and can be surrogate; that shown in Figure 16 is simply an example. For example, a CWA scheme with weights can be used to apply a weight factor to the differences between individual samples as shown in co-pending US Patent Application Publication No. 2008-0077030.

Returning to Figure 3A, individual events are detected by applying a detection profile 70 to a signal 72. The detection profile 70 includes a refractory period 74 followed by a constant threshold period 76 and a decay period 78. Other shapes can be used for the detection profile 70.

Signal 72 has R waves and T waves shown in the figure. In the example shown, the T waves are large compared to the R waves. The refractory periods shown by the hatched rectangles on both the R waves and the T waves indicate that each R wave and each T wave is being treated as an event. detected. As a result, for each cardiac cycle, the detection profile 70 is detecting two events. This is an example of over-detection.

In the illustrative example, each of the individual detections is also being processed by correlation analysis relative to a template that is based on an R wave. The results of the correlation analysis are shown graphically in Table 80. The Chart 80 includes boundaries for a High ”correlation and for a“ Low ”correlation. In the example, each "X" indicates the correlation score for each detected event. A pattern of High-Low-High correlation scores is found as shown in rectangle 82. In the example, each High-Low-High sequence leads to the conclusion that the events detected with a Low score ”are over- detected. As a result, as shown, the “Low” score detected event will be discarded when a High-Low-High pattern is found. In one example, High "is defined as greater than a 52% correlation, while" Low "is defined as less than a 25% correlation, when calculated using the form shown in Table 510 of Figure 16. They can use other analytical values ​​and methods.

Figure 3B illustrates the steps of a method for an illustrative example that includes rhythm correction. Once a morphology over-detection pattern has been found, as indicated in step 90, one or more over-detections are identified, as shown in step 92. The event intervals are then recalculated. and / or rhythm, as shown in step 94.

For example, as shown in diagram 96, a series of R wave and T wave detections can result in a set of calculations of intervals of 225 ms (R to T) and 300 ms (T to R) , which provides an average interval of 263 ms. An average interval of 263 milliseconds results in a rate of approximately 229 beats per minute, which would constitute a treatable tachyarrhythmia in many patients. However, when the T waves are identified as oversensing and the intervals on both sides of the T waves are combined, as shown in diagram 98, the intervals average 525 milliseconds. The rhythm can be recalculated to a value of approximately 114 beats per minute, avoiding possible defibrillation, cardioversion, or resynchronization pacing that could occur if the data had not been corrected.

Figure 4 shows an example of correlation comparisons between events. A cross-event comparison is a comparison in which two individual detected events are compared to one another. The comparison can take the form of a correlation analysis, or it can make use of some other type of analysis such as a wavelet transform, a Principal Component Analysis (PCA), etc., to evaluate the similarity between two detected events. In the case of a comparison using a wavelet transform or PCA, the similarity of the data compression results to a wavelet or PCA output can be compared. For example, the similarity and / or the order of the PCA output eigenvalues, or the similarity of the wavelet coefficients that result from a wavelet transformation can be compared in a qualitative or quantitative way.

In the example shown in Figure 4, a correlation analysis is carried out. In the example, as shown in diagram 108, the correlation scores are characterized as Low, Medium, or High. The "High" scoring area indicates high confidence that the compared signals are of the same character (for example, if one event is an R wave, so will the other), while the "Low" scores indicate that the signals compared they are very different from each other. The “Middle” zone aims to capture those signals that are similar but do not generate high confidence that the two signals are of the same character. For example, in a patient experiencing a rhythm-dependent morphological change (such as a rhythm-induced bundle branch block), the captured R waves may not be strongly correlated with a stored static template but will likely fall into the range " Medium ”relative to the template. In another example, a monomorphic ventricular tachycardia probably has a High or Medium correlation between R waves, and a Medium correlation between T waves, whereas a polymorphic ventricular tachycardia would show

ES 2 425 345 T3 a Medium or Low correlation between R waves.

If so desired, fuzzy logic can be applied. The use of a Middle Zone suggests this possibility. For example, instead of considering simple High "or Low" characterizations, additional categories can be provided. Additionally, a previous measurement can be used to inform a subsequent characterization of a marginally similar or different signal.

As shown in diagram 100, a series of N, N-1, N-2 and N-3 events are considered as a group, where the N-th detection is compared to each of the N-1 events, N-2 and N-3 using correlation analysis. The results of the comparisons between events and the comparisons with a static template are shown in Table 102. The results of comparison between events are displayed in cells 104, which include ordered results of the comparisons of a given event with three previous events. Table 102 shows results for events N, N-1, N-2, and N-3. The results of the comparisons between events show that for any given event X, the correlation with event X-2 is higher than with events X-1 or X-3. This may indicate a pattern of double detections based on increased correlation between alternate events.

In the illustrated example, comparisons with a normal sinus rhythm static template can further be made. Illustrative results are shown in cells 106. The alternating results of the static template, Low-Medium-Low-Medium suggest a possible over-detection, but because the probable R waves are not strongly correlated, there is no strong confidence based exclusively on the static template. However, when taken in combination with inter-event comparison information, there is significant confidence that some events are over-detections. A set of applicable rules could be as follows:

1) Low-High-Low Alternation for N when compared to N-1, N-2 and N-3, and

2) Low-High-Low Alternation for N-2 when compared to N-3, N-4 and N-5.

Conclusion: Treat N-1 and N-3 as T waves.

An additional confirmation rule could be the following:

3) A correlation at least “Average” for N and N-2 with the static template.

Another strategy is to apply only rules 1) and 3), marking only event N-1 as an oversensing in response to compliance with the rule set. Once one or more events have been marked as over-detections, they can be handled in the manner shown in Figure 3B, above.

Figure 5 shows another example of correlation comparisons between events. Here the captured signal is detected three times, as shown in diagram 120. In this instance, the N-th detection is compared to each of the N-1, N-2, N-3 and N-4 detections. . The inclusion of four individual comparisons can further help to distinguish a triple detection from a double detection, although some embodiments stop the process at three comparisons.

The results are shown in Table 124. For each set of comparisons, there are three Low correlations, and one Medium correlation or one High correlation. It is likely that with a triple detection, some detections will have a low correlation in each comparison. An illustrative set of rules is as follows:

1. The N-th event has a High correlation with the N-3 event;

2. Events N-1 and N-2 have Low correlations with the N-th event; Y

3. Events N-1 and N-2 have Low correlations with the Static Template.

If these three conditions are met, then events N-1 and N-2 can be discarded. Additional conditions can be added. For example, the static template characteristics of event N and / or event N-3 can also be considered, for example:

Four. The N-th and N-3 events have a Medium or High Correlation with the Static Template.

Then, if all rules 1 through 4 are met, events N-1 and N-2 can be discarded and the interval between event N and event N-3 can be calculated and used in rhythm analysis.

In a further example, the widths of each event can also be considered, for example using this fifth condition:

5. Events N-1 and N-2 are wider than a Width Threshold.

The width threshold can be set as desired; In one example, the Width Threshold has a value in the range of 100 to 140 ms. This Threshold Width rule can be applied as a layer

ES 2 425 345 T3 added to any determination that an event should be discarded for being an over-detection. In another example, the polarities can be considered:

6. Events N-1 and N-2 have the same polarity.

Polarity can be defined, for example, by referring to most signal samples for an event, such as the polarity of the sample that has the highest value in the event, or by determining which extreme, the most positive, or the least positive, it occurs first in the event.

If desired, interval coupling can be added as another condition:

7. The combined range from N to N-3 is less than Duration.

Where Duration is in the range between 800 and 1200 ms. This condition, and variants thereof, are also explained in association with Figures 11-13 and 14A-B below.

Figure 6 shows an analytical strategy for short series and long series correlation analysis. Figure 6 shows a graph 140 to represent the correlation scores for a series of detected events. Correlation scores, shown as Xs, are plotted against lines 144 and 146 that define a wide band 148, and lines 150, 152 that define a narrow band 154.

The broad band 148 is applied to identify an over-detection when there are two detected events that have scores above line 144 separated by a single detected event with a score below line 146, for example as shown in Figure 7A. The narrow band is applied to identify over-detection (s) when a consecutive series of detections alternate above line 150 and below line 152, as shown for example in Figure 7B. The numbers are shown for each threshold for illustrative purposes; these numbers can quantify the correlation as a percentage.

The narrower band 154 is applied to a less stringent standard than the wider band 152 with respect to correlation scores, and therefore a greater number of events are analyzed before a decision is made to discard low scoring events. In an illustrative example, events are not discarded using narrow band 154 until the 8-event pattern shown in Figure 7B is encountered, at which point between one and four low-scoring events are discarded, correcting the intervals around each discarded event. Subsequent to finding the pattern in this initial step, only the most recent low scoring event would be discarded. For analytical purposes, previously discarded events are used to determine if the outer 8 consecutive events rule is met, even if those events are excluded from the pacing calculations. Another embodiment uses only 5 events, searching for a High-Low-High-Low-High sequence using the narrowest band 154 and, if such a sequence is found, one or both of the Low scoring events is discarded.

The examples in Figure 6 and Figures 7A-7B indicate numbers where correlations of 50% and 20% limit the wide band 148 and correlations of 40% and 25% limit the narrow band 154. These numbers are merely illustrative. In one example, these numbers are applied by scaling the formula shown in Table 510 of Figure 16 to a percentage basis.

Figures 8A-8B illustrate examples of how to tailor the correlation analysis to the observed levels of correlation with a template. Referring to Figure 8A, a correlation score plot is shown in diagram 158 for comparing a template to a series of events. For the purpose of identifying double detections, a mean correlation score is calculated for odd events. The grouping of the odd events is then analyzed to determine whether the odd events all fall within a preset distance from the mean, using for example the set standard deviation or using a fixed distance. If the odd events all fall within the preset distance from the mean, the separation of the mean from a Low boundary is calculated. If the separation is greater than a predetermined threshold, then the odd events are determined to show monotonicity, supporting the assumption that the odd events are QRS complex detections. If monotonicity is identified in the odd events, one or more of the even events that fall below the lower threshold are marked as over-detections.

In another embodiment, before any of the even-numbered events are marked as over-detections, they are all analyzed to determine if clustering of the even-numbered events has occurred, again using the mean of these events. Instead of the separation between the mean of the odd events and the lower boundary, the separation between the means of the even and odd events is calculated to establish groupings of the events. In this embodiment, oversensing markers are only applied when sufficient clustering of even-numbered events appears.

Figure 8B shows another example where overdetection marking is adapted to correlation scores with a static template. In this case, the average of the correlation score is calculated for a set of 10 events. A suppressed band is then established around the average value of the correlation score. For example, the suppressed band can be defined as +/- 15%. Other can be used

ES 2 425 345 T3 blanked band sizes.

In the example of Figure 8B, high scores are defined as those scores that are above the suppressed band, and low scores as those that are below the suppressed band. If a High-Low-High pattern appears around the suppressed band, then over-detection can be identified and one or more of the Low-scoring events is marked as over-detection.

Instead of a static template, the analysis shown in Figures 8A-8B can also be applied using a recently detected event as a template for comparison. The analysis seen in Figures 8A-8B can use averaging / mean calculation, or it can use some other predictor of a center point for signals including mode, median or other mathematical operation.

A further use of the comparisons between events shown here can be used in determining the occurrence of a Shock rhythm. Delivery of a stimulus is commonly used to treat polymorphic conditions, such as Polymorphic Ventricular Tachycardia and Ventricular Fibrillation. Monomorphic conditions such as a Monomorphic Ventricular Tachycardia (VMT) can be treated, but a VMT does not always require the most aggressive treatment. For example, a TVM can be treated using antitachycardia pacing (ATT) rather than defibrillation or cardioversion, as an EAT uses less energy and may be less traumatic to the patient. Correlation patterns can be used to distinguish between monomorphic arrhythmias and polymorphic arrhythmias. For example, an ongoing pattern such as that shown in Figures 7A or 7B, or even Figure 6, in which consistently high correlations are found, can be used to delay therapy, if desired.

In another example, a pattern such as that shown in Figure 8A can be further analyzed by determining the value of the standard deviation for the pooled high scores. If the pooled high scores are based on a static template and show a small standard deviation, this may indicate a monomorphic condition. In some embodiments, particularly if an EAT is not available, therapy can be inhibited until the monomorphic condition evolves to a more polymorphic condition.

In one example, a system uses stratified correlation analysis to identify treatable arrhythmias. In the example, a simple single-event correlation analysis using a static template is run until a pattern like the one shown in Figure 8A appears. Such a pattern then triggers multiple comparisons between events as shown in Figures 4-5. Then, if the comparisons between events show probable over-detection, the data ranges can be corrected. Additionally, if the comparisons between events show a monomorphic condition, the therapy can be inhibited.

Figure 9 illustrates methods for aligning and realigning a captured signal with a correlation analysis template. The correlation analysis template is shown on line 200, while a signal is shown on line 202. The correlation analysis template 200 can be a static template or it can represent a single detected event or the average of several recently detected events.

As seen in Figure 16, correlation analysis typically uses a reference point as an alignment guide for an ordered series of template values ​​and signal samples. In the example of Figure 9, a base alignment point is identified as the sample of both the template 200 and the signal 202 having a higher value. A series of comparisons are then carried out, beginning with a base alignment comparison, shown in box 210, and individual sample offsets to the right, shown in box 212, and to the left, shown in box 214. The right one shift correlation 212 turns out to be worse than the correlation score for base comparison 210, and therefore the result of the right one shift correlation 212 is discarded. Left one shift correlation 214 provides a correlation score higher than alignment correlation 210, and thus the result of base correlation 210 is discarded, and a new left shift correlation is calculated as as shown in frame 216, this time shifting the point alignment by two samples. The result in box 216 shows a correlation less than the left one shift correlation in box 214, and therefore the process stops and uses the correlation score calculated for the left one shift correlation 214 as a correlation score for signal 202.

When scrolling to the right and / or to the left, the scaling of the signal to the template can also be modified. For example, if scaling is carried out initially by comparing the signal peak and the template peak, then equating both, when the shift is carried out the signal peak can instead be scaled to the point with which is aligned on the template after the offset has occurred.

The method shown in Figure 9 can help correct for noise or misalignment due to artifact sampling, rate of rise, etc., which may cause the alignment point of the sample peak 202 to be suboptimal. The method includes calculating the correlation score when the reference points are aligned and also when the reference points are misaligned by one or more samples in each of the two directions until a maximum correlation score is found. Limits can be set, such

ES 2 425 345 T3 as desired, in the number of samples in which it shifts to the right or left. In another embodiment, different scores are automatically calculated (eg, one base, one, two, and three on the left, one, two, and three on the right) and the best of them is chosen.

In another embodiment highlighted in Figure 9, a plural number of alignment points can be defined for the template 200. Some examples include the start of the QRS complex, the maximum amplitude, the maximum amplitude in the opposite polarity of the maximum amplitude (note that the amplitudes are maximum are identified because they both correspond to a point of change in which dV / dt = 0), the point of maximum slope between the two maximum peaks (shown as dV / dt = MAX, etc.). By identifying the analogous points in the signal, the method can determine whether using different possible alignment points could provide different output values ​​from the correlation analysis. For example, the point of maximum amplitude of the entire signal could be used by default, but it could happen that some cardiac events are aligned in the opposite direction using the point of maximum slope in the monotonous segment that follows the point of maximum amplitude.

Figure 10 shows another method of storing and applying a template for correlation analysis. In this example, the signal that forms a basis for a template is shown in box 230. For the illustrative example, when the template is formed, an interpolation region is defined between the positive peak and the negative peak of signal 230. As a result, the stored template takes the form shown in frame 240: template 240 matches template signal 230 in regions before the positive peak and after the negative peak, but is flexible between the two peaks, as indicated by a dashed line in trace 242. The positive peak, in the example shown, is the highest magnitude peak in the template, and is therefore used to scale the template to a captured signal.

The alignment to a sample 232 is then performed as shown in box 244. The template is adjusted such that the positive and negative peaks align with the captured signal, with linear interpolation between them. Outside of the positive and negative peaks, the template continues to match the signal as shown in Table 230, but nevertheless the duration and slope between the positive and negative peaks are adjusted to match the captured event. The setting shown in Figure 10 can avoid the difficulty of the duration of a static template being set for a patient whose width of the QRS complex is affected by heart rhythm. The adjustment made can be limited in order to avoid an excessive increase in the width of the template.

In another example, more than two template points are identified and linear interpolation can be used between them. For example, a template can be made up of five values ​​where each has a relative width and a relative location. When a detected event is to be compared to the template, the width and amplitude of the peak of the detected event are used to scale each of the template values, using linear interpolation between the template points.

Figures 11 and 12 illustrate a method for inhibiting data correction after identification of probable over-detection.

As shown in Figure 11, a QRS complex occurs in line 260, followed by a premature ventricular contraction (PVC) shown in line 262, followed by another QRS complex in line 264. CVP is characterized, in this For example, due to a low correlation with the workforce. Consequently, a High-Low-High correlation pattern appears, similar to that shown in Figure 3A. Therefore, some examples would discard CVP 262. Analytically, however, discarding CVP 262 might be unnecessary as it is not really an over-detected event. Additionally, the intervals around CVP 262 are both greater than 500 milliseconds. Even without a data correction, averaging the two intervals would provide an event rate of approximately 103 beats per minute, a rate that would not threaten to lead to unnecessary therapy. Therefore, data correction would not improve the specificity of the rhythm in the device, while it would reduce the sensitivity to beats.

Figure 12 illustrates a method that would avoid discarding a CVP 262 as shown in Figure 11. Based on the detected events 270, the method determines, as shown in step 272, whether a sequence of correlation scores appears that would support the finding of a double detection (DD) or an over-detection. If not, the method terminates, as a data correction is not about to happen. If the result of step 272 is a "Yes," the method then includes determining whether the new interval that would result from a data correction would be greater than a predetermined threshold, as shown in step 274. In the illustrative example , the threshold is 1000 milliseconds (60 beats per minute), although this number is merely illustrative. Some likely thresholds are in the range 750 to 1200 milliseconds.

In another example, the order of the analysis is reversed, and the over-detection analysis does not take place unless the calculated rate is high (typically 150 beats per minute or more), or unless the intervals that could be affected are short enough to pass the applied test. In another embodiment, individual intervals are compared with a threshold (eg, in the interval between 400 and 600 ms) and, if both intervals exceed the threshold, no combination of intervals takes place. In yet another example, the threshold may be a programmable parameter of an implantable system. In another example, the threshold can be scaled based on

ES 2 425 345 T3 a programmable ventricular tachycardia parameter that is used to set a beat rhythm that the implantable system will treat as a ventricular tachycardia rhythm.

If the corrected interval is not greater than the threshold, the method continues to the step where the intervals are combined, as shown in step 276, to correct for the over-detected event (s). If the corrected interval were greater than the threshold in step 274, the method would simply terminate without combining intervals. In this way, unnecessary correction of the stored data can be avoided.

Figure 13 illustrates more methods for inhibiting correlation analysis after identification of oversensing. The methods in Figure 13 take advantage of known relationships between QT interval and RR interval in physiological cardiac cycles. The illustrative method begins again with the identification of a pattern suggesting over-detection, as indicated in step 300. As shown in step 302, the possibly over-detected event is then treated as a T wave (here, the presumption is that a pattern of three events is identified, where the intermediate event of the three is probably an over-detection ; other variants may be used) and, as shown in step 304, events on either side of the probable oversensing are treated as R waves.

These presumptive R and T “waves from steps 302 and 304 are then used to apply a formula that allows calculating the duration of the QT interval from the RR interval in step 306. In particular, in step 308 different plausible formulas are shown . Examples include Bazett's formula:

QT (Esp) = QT * JRR

Friderica's formula:

QT (Esp) = QT * ^ RR and the regression formula of Sagie and others:

QT (Esp) = QT + A * (RR -1)

Sagie and others found A = 0.154.

In each formula, the expected QT interval is displayed as QT (Esp), the RR interval value is in seconds, and the QT interval value is captured during a programming session between a programmer and a recipient of the implanted device. The QT interval is either captured or adjusted for a heart rate of 60 beats per minute. The RR interval is found in step 304, and the measured QT interval can be captured by adding the measured width of the presumed T wave to the interval between the first R wave and the presumed T wave.

The expectation is that if the probably over-detected event is an over-detected T wave, the measured QT period will coincide with the expected QT period for a given RR interval value, using whatever formula can be applied, with a tolerance band for allow a certain error.

If the formula applied in step 306 does not provide a matching value, no event is thrown away, as shown in step 310. Alternatively, if the formula applied in step 306 provides a matching value, then the probable over -detection is discarded as shown in step 312. When the probable over-detection is discarded in step 312, the gaps around the over-detection are combined, as shown in Figure 3B. Again, in other examples the order of analysis is reversed.

Figures 14A-14B show the application of a method illustrated in Figure 13. In the illustrative examples of Figures 14A and 14B, Friderica's cube root formula is applied. In each example, the previously measured QT interval has a value of 400 milliseconds. This value represents the estimated QT interval for the hypothetical patient that would occur at a heart rate of 60 beats per minute.

Referring to Figure 14A, given three events X, Y, and Z that have a correlation pattern indicating oversensing, the method is applied by assuming that event Y is a T wave. The QT interval is measured for the event X and for event Y, and the RR interval is measured from event X to event Z, as indicated. The measured QT interval is also referenced, and these values ​​are entered into the chosen formula. In the example shown, using an RR interval value of 0.8 seconds, the expected value for the QT interval is 371 milliseconds. Applying a +/- 10% error band to the calculation, the acceptable interval has a duration of approximately 334 to 408 milliseconds for the QT interval. However, as shown, the measured interval has a value of approximately 500 milliseconds, too long to be a QT interval for the given parameters. As a result, the calculation suggests that the Y detection is not an oversensed T wave, and therefore no data correction takes place. Smaller or larger error bands can be applied; for example, an error of +/- 5% is used in another illustrative embodiment.

Referring now to Figure 14B, this time the measured QT interval for event X and for event Y has a

ES 2 425 345 T3 duration of approximately 370 milliseconds. This value falls within the expected range, and therefore the calculation suggests that the Y detection is an over-detected T wave. Therefore, detection Y is discarded and the data for the interval between event X and event Z is corrected.

In the examples of Figures 11-13 and 14A-B, if a probable over-detection is not ruled out, leading to a data correction, the probable over-detection may instead be flagged as suspicious detection. In one example, suspicious detections are treated as unreliable, both when they are used as indicators of cardiac activity and when they are used as end points for intervals that can be used in rhythm analysis. If the probable over-detection is marked as a suspicious detection, the suspicious detection and each of the preceding and subsequent intervals around the suspicious detection are completely removed from the analysis.

Figure 15 shows an analysis method to identify detected shock events and treatable rhythms. Figure 15 shows the structure of an analysis method by including event detection steps 402, which is followed by waveform evaluation 404 and beat rating 406. In particular, event detection 402 will typically include monitoring a captured signal to detect signal amplitude changes that indicate cardiac events. Once cardiac events are captured in block 402, waveform evaluation 404 can take place. During waveform evaluation 404, the characteristics of the signal associated with a detected event are analyzed to identify and eliminate detected events that are likely caused by noise or artifacts.

The detected events that pass the waveform evaluation 404 are then subjected to a beat rating 406, during which the detected events are analyzed to determine whether they exhibit morphology or interval characteristics that indicate accurate detection. This may include the correlation analysis shown above, and / or the interval analysis or combinations of both; for example, analysis to eliminate wide complex double detections can use proximity criteria and shape characteristics of detected events to identify probable over-detections. Some additional discussion appears in US patent application. with Serial Number 12 / 399,914, entitled METHODS AND DEVICES TO CLASSIFY CARDIAC ACTIVITY ACCURATELY.

The architecture then proceeds to rate classification, which can begin by considering rate at block 408. If rate is low, then an individual detection is marked as "Not Shockable" as indicated in step 410. Alternatively, yes the rate is very high, the possibility of indicating ventricular fibrillation (VF) is considered and is therefore marked as Shockable, as shown in step 412. Between these VF and low rhythm bands is the ventricular tachycardia (VT) zone, and rhythms in the VT zone are analyzed using something we will refer to as Detection Enhancements, as shown in step 414. .

An example of a Detection Enhancement is as follows:

1. Compare to static template: if match, not crash; in another case

2. Compare with dynamic template: if no match, crash event; in another case

3. Compare with QRS complex width threshold: if wide, shocking, otherwise non-shocking.

Where the dynamic template can be any of the following:

a) An average of several previous detections that are correlated with each other;

b) A set of individual events, for example {N-1 ... Ni} in which the match of some or all of the individual events counts as the match of the dynamic template;

c) A continuously updated template.

The aforementioned QRS complex width threshold can be applied in different ways that can be tailored to the QRS complex width measurement method used in a given system and / or that can be tailored to an individual patient. In one example, the following rules apply for the width of the QRS complex:

x) During the analysis, the width of the QRS complex is calculated as the duration from the start of the longest monotonic segment captured during the refractory period before the reference point to the end of the longest monotonic segment captured during the refractory period after the point. reference;

y) The QRS complex width threshold is measured for the patient during a programming session, with a maximum allowed value of 113 ms; Y

z) The width of the QRS complex during the analysis is considered wide if it is at least 20 ms longer than the threshold of width of the QRS complex.

ES 2 425 345 T3

These rules x), y) and z) are adapted for a particular embodiment and may vary depending on the system used.

After the events are marked as Non-Shockable 410 or Shockable 412, an XY counter condition is applied as indicated in step 416. The XY counter condition analyzes the number of Shockable events, X, that are marked during a set above, Y, of detected events that pass both the waveform evaluation 404 and the beat score 406. The ratio applied, and the set size used, may vary. One embodiment applies an XY counter condition 18-24 in step 416. Other embodiments use ratios such as 8 or 9 over 12, 12 or 13 over 16, 24-32, and so on.

If the XY condition is not met, no shock will be delivered, as shown in step 418. If the XY condition is met, then the method can continue to the load confirmation block 420. For example, some embodiments require the XY ratio-set size condition to be met for a selected number of consecutive events, and this condition may be tested in the loading confirmation step 420. Another example condition is to determine if a set, N, of immediately preceding detected events are all Shockable, or if they all have intervals that are short enough to support the conclusion that the detected arrhythmia is occurring. Other factors may also be applied in confirming loading, for example, noting if an over-detection has recently been found (which might suggest that therapy should be delayed to ensure that the arrhythmia is not a manifestation of an over-count) , or by observing whether consistent long intervals have been detected (potentially suggesting a spontaneous conversion of the patient's heart to a normal rhythm). For example, the upload confirmation step 420 may also include methods such as those shown in co-pending U.S. Patent Application Publication No. 11 / 042,911, entitled METHOD FOR ADAPTING THE INITIATION OF THE LOAD FOR AN IMPLANTABLE CARDIOVERTER-DEFIBRILLATOR, the description of which is incorporated by reference herein.

Charge and Shock block 422 is reached if Charge Confirmation 420 is exceeded. Typically, the loading process lasts for a certain period of time, and therefore method 400 can be iterated several times before loading is complete. Some or all of the analyzes used to reach an initial determination that Charging should begin may be repeated during the process. Finally, if treatable conditions persist during loading, or are identified after loading, a stimulus can be delivered.

In relation to the implantable system, various hardware-specific features can be incorporated. For example, any battery with appropriate chemistry can be used, such as a lithium ion battery. The therapeutic output can be created using a capacitive system to store energy until a stimulus level is reached using one or more capacitors. A charging circuit such as a transformer reverse transfer (flyback) circuit can be used to generate therapeutic voltages. Therapy can be delivered using, for example, an H-bridge circuit or a modification thereof. General purpose or specialized circuitry can be used to carry out the analysis functions. For example, a specialized cardiac signal analog-to-digital circuit can be used, as well as a specialized correlation analysis block, as desired, while other functions can be carried out with a microcontroller. Static or dynamic memories can be provided and can be used for any appropriate functions. These elements may all be components of the operational electrical and electronic circuitry for the implantable cardiac pacing system.

Those skilled in the art will recognize that the present invention may manifest itself in a variety of ways other than the specific embodiments described and contemplated herein. Consequently, deviations in shape and detail may be allowed without departing from the scope of the present invention.