Back to results
US 2023/0420132 A1Published application
Noninvasive heart failure detection
AssigneeUniversity of Southern California
InventorsNiema M. PAHLEVAN, Rashid ALAVI DEHKORDI, Ray V. MATTHEWS
Life of this patent
Estimating the term…
Abstract
Some embodiments relate to non-invasive techniques for determining whether a patient has experienced heart failure. In some embodiments, a machine learning model may be trained to determine whether the patient has experienced heart failure using blood pressure waveforms and ECGs taken concurrently. Using the intrinsic frequency methodology and the cardiac triangle mapping methodology, features about the patient's cardiac cycle can be extracted and provided to the trained model as input.
Claims15
A system for non-invasively determining whether a patient has experienced heart failure based on an evaluation of a left ventricular end-diastolic pressure (LVEDP) of the patient, the system comprising:
- at least one sensor configured to capture femoral blood pressure measurements of a patient and generate a femoral blood pressure waveform representing the femoral blood pressure measurements;
- an electrocardiograph (ECG) device configured to capture an ECG of the patient, wherein the ECG and the femoral blood pressure measurements are captured concurrently;
- a computing system comprising one or more processors executing computer program instructions to effectuate operations comprising:
- obtaining, from the at least one sensor, first data representing the femoral blood pressure waveform of the patient;
- obtaining, from the ECG device, second data representing the ECG of the patient;
- computing, based on the first data, a first intrinsic frequency of a systolic phase of a cardiac cycle of the patient, a second intrinsic frequency of a diastolic phase of the cardiac cycle of the patient, a systolic intrinsic phase angle describing the systolic phase of the cardiac cycle of the patient, a diastolic intrinsic phase angle describing the diastolic phase of the cardiac cycle of the patient, a systolic intrinsic envelope of the systolic phase of the cardiac cycle of the patient, a diastolic intrinsic envelope of the diastolic phase of the cardiac cycle of the patient, and an envelope ratio based on the systolic intrinsic envelope and the diastolic intrinsic envelope;
- extracting, based on the first data, a first set of features comprising:
- a first temporal value corresponding to a beginning of the femoral blood pressure waveform, a second temporal value corresponding to a dicrotic notch of the femoral blood pressure waveform, and a third temporal value corresponding to an end of the femoral blood pressure waveform;
- extracting, based on the second data, a second set of features comprising:
- a fourth temporal value corresponding to a beginning of a QRS complex and a fifth temporal value corresponding to a peak of an R-wave of the ECG of the patient;
- estimating, based on a first amount of time between the first temporal value and the fourth temporal value or a second amount of time between the first temporal value and the fifth temporal value, a maximum rate of a rising portion of the femoral blood pressure waveform;
- selecting a subset of features from the first set of features, the second set of features, the first amount of time between the first temporal value and the fourth temporal value, the second amount of time between the first temporal value and the fifth temporal value, and the maximum rate of the rising portion of the femoral blood pressure waveform;
- providing the subset of features to a trained random forest model to obtain a first value representing a left ventricular end-diastolic pressure (LVEDP) of the patient, wherein the trained random forest model is trained to estimate an LVEDP based on instances of the first set of features and the second set of features extracted from training data, wherein the training data comprises femoral blood pressure waveforms and ECGs of patients that have experienced heart failure and patients that have not experienced heart failure;
- determining whether the first value representing the LVEDP of the patient satisfies a threshold condition, wherein the threshold condition being satisfied comprising the first value being greater than or equal to a threshold LVEDP value;
- and
- storing, in memory, based on the first value, a result indicating whether the patient has an elevated LVEDP or a normal LVEDP.
13 more elements
A non-transitory computer-readable medium storing computer program instructions that, when executed by one or more processors of a computing system, effectuate operations comprising:
- obtaining first data representing a blood pressure waveform of a patient;
- obtaining second data representing an ECG of the patient;
- computing, based on the first data, cardiac parameters of a cardiac cycle of the patient, the cardiac parameters comprising intrinsic frequencies, intrinsic phase angles, and an envelope ratio (ER);
- extracting, based on the first data, a first set of features of the cardiac cycle, the first set of features comprising temporal values associated with a beginning, dicrotic notch, and end of the blood pressure waveform;
- extracting, based on the second data, a second set of features of the cardiac cycle, the second set of features comprising temporal values associated with a beginning of a QRS complex and a peak of an R-wave of the ECG;
- determining, from the first set of features and the second set of features, a third set of features comprising at least one of:
- an amount of time between the beginning of the blood pressure waveform and the beginning of the QRS complex, an amount of time between the beginning of the blood pressure waveform and the peak of the R-wave, and a rate of change of the blood pressure waveform;
- selecting a subset of features from the first set of features, the second set of features, and the third set of features;
- providing the subset of features to a trained machine learning model to obtain a first value representing a left ventricular end-diastolic pressure (LVEDP) of the patient;
- determining whether the first value satisfies a threshold condition;
- and
- storing, in memory, based on the first value, a result indicating whether the patient has an elevated LVEDP or a normal LVEDP.
9 more elements
A method implemented by a computing system configured to execute computer program instructions, the method comprising:
- obtaining, with the computing system, first data representing a blood pressure waveform of a patient;
- obtaining, with the computing system, second data representing an ECG of the patient;
- computing, with the computing system, based on the first data, cardiac parameters of a cardiac cycle of the patient, the cardiac parameters comprising intrinsic frequencies, intrinsic phase angles, and an envelope ratio (ER);
- extracting, with the computing system, based on the first data, a first set of features of the cardiac cycle, the first set of features comprising temporal values associated with a beginning, dicrotic notch, and end of the blood pressure waveform;
- extracting, with the computing system, based on the second data, a second set of features of the cardiac cycle, the second set of features comprising temporal values associated with a beginning of a QRS complex and a peak of an R-wave of the ECG;
- determining, with the computing system, from the first set of features and the second set of features, a third set of features comprising at least one of:
- an amount of time between the beginning of the blood pressure waveform and the beginning of the QRS complex, an amount of time between the beginning of the blood pressure waveform and the peak of the R-wave, and a rate of change of the blood pressure waveform;
- selecting, with the computing system, a subset of features from the first set of features, the second set of features, and the third set of features;
- providing, with the computing system, the subset of features to a trained machine learning model to obtain a first value representing a left ventricular end-diastolic pressure (LVEDP) of the patient;
- determining, with the computing system, whether the first value satisfies a threshold condition;
- and
- storing, with the computing system, in memory, based on the first value, a result indicating whether the patient has an elevated LVEDP or a normal LVEDP.
9 more elements
Dependencies are inferred from the claim text.