Implantable medical device with pacing capture classification
This disclosure is directed to devices and techniques for classifying of pacing captures to evaluate effectiveness of pacing by a pacing device, such as an implantable medical device (IMD). An example system includes stimulation circuitry to generate a pacing stimulus, sensing circuitry to sense an evoked response after the pacing stimulus, and processing circuitry. The processing circuitry determines classification features from the evoked response and applies the classification features to a classification model, the classification model generated by a machine learning algorithm using one or more test sets comprising a plurality of sample evoked responses for each of a plurality of classifications. Based on the output of the model, the processing circuitry classifies the evoke response as one of the plurality of classifications.
1. A system comprising:
stimulation circuitry configured to generate a pacing stimulus;
sensing circuitry configured to sense an evoked response after the pacing stimulus, wherein the evoked response is indicative of capture of heart muscle in response to the pacing stimulus; and
processing circuitry configured to:
determine classification features from the evoked response;
apply the classification features to a classification model, the classification model generated by a machine learning algorithm using one or more test sets comprising a plurality of sample evoked responses for each of a plurality of classifications, wherein the plurality of classifications includes a selective capture, non-selective capture, and right ventricular capture;
based on the output of the model, classify the evoke response as a non-selective capture classification as a change from a selective capture classification for a previous evoked response; and
in response to determining the evoked response includes the change from the selective capture classification to the non-selective capture classification, change an amplitude level of pacing stimulus based on an amplitude level of the pacing stimulus causing an evoked response including a selective capture classification.
2. The system of claim 1 , wherein the classification features are a subset of possible features of the evoked response.
3. The system of claim 2 , wherein the classification features are determined by the machine learning algorithm from the possible features of the evoked response.
4. The system of claim 1 , wherein the classification model is a decision tree model.
5. The system of claim 1 , wherein the processing circuitry is further configured to, in response to classifying the evoked response as one of the plurality of classifications, generate an alert.
6. The system of claim 1 , wherein the processing circuitry is configured to, in response to determining the evoked response includes the change from the selective capture classification to the non-selective capture classification, change the amplitude level by changing a voltage level of the pacing stimulus.
7. The system of claim 1 , including an implantable medical device comprising the stimulation circuitry, the sensing circuitry, and the processing circuitry.
8. The system of claim 1 , wherein the sensing circuitry is configured to sense a near field (NF) electrogram (EGM) of the evoked response and a far-field (FF) EGM of the evoked response, and wherein the processing circuitry is configured to determine a differential far field (DFF) EGM.
9. The system of claim 8 , wherein the classification features comprise two or more of (a) a time (T1) between the pacing stimulus and a response of the FF potential when a negative deflection is below a predefined threshold, (b) a width (T2) of the FF potential at the negative deflection, (c) a time (T4) between a start of the pacing stimulus to a positive peak of the DFF potential, (d) a time (T6) between the start of the pacing stimulus to a positive peak of the NF potential, (e) a time (TMAX) between the start of the pacing stimulus to a positive peak of the FF potential, (f) a time (TMIN) between the start of ventricular pacing to a negative peak of the FF potential, (g) a negative peak amplitude (A1) of the FF potential from a zero line within a predefined window, (h) an amplitude (A2) of the FF potential from the negative peak to a positive peak within the predefined window 408 , (i) a positive peak amplitude (A3) of the DFF potential within the predefined window, (j) an absolute peak amplitude (AMAX) from a zero line of the NF potential, and (k) a negative slope (SP1) following a maximum positive peak of the FF potential.
10. A method comprising:
generating, by stimulation circuitry, a pacing stimulus;
sensing, by sensing circuitry, an evoked response after the pacing stimulus, wherein the evoked response is indicative of capture of heart muscle in response to the pacing stimulus;
determining, by processing circuitry, classification features from the evoked response;
applying, by the processing circuitry, the classification features to a classification model, the classification model generated by a machine learning algorithm using one or more test sets comprising a plurality of sample evoked responses for each of a plurality of classifications, wherein the plurality of classifications includes a selective capture, non-selective capture, and right ventricular capture;
based on output of the model, classifying, by the processing circuitry, the evoked response as a non-selective capture classification as a change from a selective capture classification for a previous evoked response; and
in response to determining the evoked response includes the change from the selective capture classification to the non-selective capture classification, changing an amplitude level of pacing stimulus based on an amplitude level of the pacing stimulus causing an evoked response including a selective capture classification.
11. The method of claim 10 , wherein the classification features are a subset of possible features of the evoked response.
12. The method of claim 11 , wherein the classification features are determined by the machine learning algorithm from the possible features of the evoked response.
13. The method of claim 10 , wherein the classification model is a decision tree model.
14. The method of claim 10 , comprising, in response to determining the evoked response includes the change from the selective capture classification to the non-selective capture classification, changing the amplitude level by changing a voltage level of the pacing stimulus.
15. The method of claim 10 , wherein the stimulation circuitry, the sensing circuitry, and the processing circuitry are within an implantable medical device.
16. The method of claim 10 , wherein the evoked response includes a far field (FF) potential, a near field (NF) potential and a differential far field (DFF) potential, and wherein the classification features comprise two or more of (a) a time (T1) between the pacing stimulus and a response of the FF potential when a negative deflection is below a predefined threshold, (b) a width (T2) of the FF potential at the negative deflection, (c) a time (T4) between a start of the pacing stimulus to a positive peak of the DFF potential, (d) a time (T6) between the start of the pacing stimulus to a positive peak of the NF potential, (e) a time (TMAX) between the start of the pacing stimulus to a positive peak of the FF potential, (f) a time (TMIN) between the start of ventricular pacing to a negative peak of the FF potential, (g) a negative peak amplitude (A1) of the FF potential from a zero line within a predefined window, (h) an amplitude (A2) of the FF potential from the negative peak to a positive peak within the predefined window, (i) a positive peak amplitude (A3) of the DFF potential within the predefined window, (j) an absolute peak amplitude (AMAX) from a zero line of the NF potential, and (k) a negative slope (SP1) following a maximum positive peak of the FF potential.
17. The method of claim 10 , wherein sensing the evoke response includes sensing a near field electrogram (EGM) of the evoked response and a far-field EGM of the evoked response, the method comprising determining, by the processing circuitry, a differential far field EGM based on the far field EGM.
18. A non-transitory computer readable medium comprising instructions, that when executed, cause an implantable medical device (IMD) to:
generate, by stimulation circuitry of the IMD, a pacing stimulus;
sense, by sensing circuitry of the IMD, an evoked response after the pacing stimulus, wherein the evoked response is indicative of capture of heart muscle in response to the pacing stimulus;
determine, by processing circuitry of the IMD, classification features from the evoked response;
apply, by the processing circuitry, the classification features to a classification model, the classification model generated by a machine learning algorithm using one or more text sets comprising a plurality of sample evoked responses for each of a plurality of classifications, wherein the plurality of classifications includes a selective capture, non-selective capture, and right ventricular capture;
based on output of the model, classify, by the processing circuitry, the evoked response as a non-selective capture classification as a change from a selective capture classification for a previous evoked response; and
in response to determining the evoked response includes the change from the selective capture classification to the non-selective capture classification, change an amplitude level of pacing stimulus base on an amplitude level of the pacing stimulus causing an evoked response including a selective capture classification.
19. The system of claim 1 , wherein to change the amplitude level of the pacing stimulus based on the amplitude level of the pacing stimulus having the evoked response including the selective capture classification, the processing circuitry is configured to:
change the amplitude level of pacing stimulus based on a lowest amplitude level of the pacing stimulus causing the evoked response including the selective capture classification.