SYSTEMS AND METHODS OF ANALYTE MEASUREMENT ANALYSIS
Disclosed are techniques for training a machine learning model. A training data set comprising a plurality of electrocardiogram (ECG) measurements and corresponding analyte measurements for each of the plurality of ECG measurements is provided. For each ECG measurement of the training data set, an estimated analyte level at a time of the ECG measurement is determined based on the corresponding set of analyte measurements. The estimated analyte level may be determined using statistical estimation techniques. If it is determined that the estimated analyte level at the time of the ECG measurement meets a certainty threshold, the ECG measurement is labeled based on the estimated analyte level at the time of the ECG measurement. A machine learning is trained to predict a level of an analyte based on ECG data, where the training is done using each of the plurality of ECG measurements that are labeled.
1 . A method comprising:
providing a training data set comprising a plurality of electrocardiogram (ECG) measurements and a corresponding set of analyte measurements for each of the plurality of ECG measurements;
for each ECG measurement of the training data set:
determining using a statistical estimation technique, an estimated analyte level at a time of the ECG measurement based on the corresponding set of analyte measurements;
determining whether the estimated analyte level at the time of the ECG measurement meets a certainty threshold; and
in response to determining that the estimated analyte level at the time of the ECG measurement meets a certainty threshold, labeling the ECG measurement based on the estimated analyte level at the time of the ECG measurement; and
training, using each of the plurality of ECG measurements that are labeled, a machine learning model to predict a level of an analyte based on ECG data.
2 . The method of claim 1 , wherein determining the estimated analyte level at the time of the ECG measurement comprises:
determining a regression line that fits to each analyte measurement of the corresponding set of analyte measurements; and
determining the estimated analyte level at the time of the ECG measurement based on the regression line.
3 . The method of claim 1 , further comprising:
for each ECG measurement of the training data:
determining a statistical measure of a fit of the regression line to each of the corresponding analyte measurements of the ECG measurement; and
determining, based on the statistical measure, a confidence interval of the estimated analyte level at the time of the ECG measurement.
4 . The method of claim 3 , wherein determining whether the estimated analyte level at the time of the ECG measurement meets the certainty threshold comprises:
comparing the confidence interval of the estimated analyte level at a time of the ECG measurement to the certainty threshold.
5 . The method of claim 4 , wherein the statistical measure is an r-squared value.
6 . The method of claim 1 , wherein labeling the ECG measurement based on the estimated analyte level at the time of the ECG measurement comprises:
labeling the ECG measurement as one of a low analyte level, a medium analyte level and a high analyte level based on the estimated analyte level at the time of the ECG measurement.
7 . The method of claim 1 , wherein labeling the ECG measurement based on the estimated analyte level at the time of the ECG measurement comprises:
labeling the ECG measurement with the estimated analyte level at the time of the ECG measurement.
8 . The method of claim 3 , wherein training the machine learning model comprises:
analyzing, using the machine learning model, each of the plurality of ECG measurements that are labeled to generate a corresponding output; and
comparing the corresponding output for each of the plurality of ECG measurements that are labeled to the associated label for each of the plurality of ECG measurements that are labeled to update the machine learning model using backpropagation, wherein one or more weight matrices of the machine learning model are adjusted based on the confidence interval associated with each of the plurality of ECG measurements that are labeled.
9 . The method of claim 1 , wherein the machine learning model is one of a convolutional neural network, a recurrent neural network and a combination of a convolutional neural network and a recurrent neural network.
10 . The method of claim 1 , wherein the analyte is one of potassium, magnesium and calcium.
11 . A system comprising:
a memory; and
a processing device operatively coupled to the memory, the processing device to:
provide a training data set comprising a plurality of electrocardiogram (ECG) measurements and a corresponding set of analyte measurements for each of the plurality of ECG measurements;
for each ECG measurement of the training data set:
determine using a statistical estimation technique, an estimated analyte level at a time of the ECG measurement based on the corresponding set of analyte measurements;
determine whether the estimated analyte level at the time of the ECG measurement meets a certainty threshold; and
in response to determining that the estimated analyte level at the time of the ECG measurement meets a certainty threshold, label the ECG measurement based on the estimated analyte level at the time of the ECG measurement; and
train, using each of the plurality of ECG measurements that are labeled, a machine learning model to predict a level of an analyte based on ECG data.
12 . The system of claim 11 , wherein to determine the estimated analyte level at the time of the ECG measurement, the processing device is to:
determine a regression line that fits to each analyte measurement of the corresponding set of analyte measurements; and
determine the estimated analyte level at the time of the ECG measurement based on the regression line.
13 . The system of claim 11 , wherein the processing device is further to:
for each ECG measurement of the training data:
determine a statistical measure of a fit of the regression line to each of the corresponding analyte measurements of the ECG measurement; and
determine, based on the statistical measure, a confidence interval of the estimated analyte level at the time of the ECG measurement.
14 . The system of claim 13 , wherein to determine whether the estimated analyte level at the time of the ECG measurement meets the certainty threshold, the processing device is to:
compare the confidence interval of the estimated analyte level at a time of the ECG measurement to the certainty threshold.
15 . The system of claim 14 , wherein the statistical measure is an r-squared value.
16 . The system of claim 11 , wherein to label the ECG measurement based on the estimated analyte level at the time of the ECG measurement, the processing device is to:
label the ECG measurement as one of a low analyte level, a medium analyte level and a high analyte level based on the estimated analyte level at the time of the ECG measurement.
17 . The system of claim 11 , wherein to label the ECG measurement based on the estimated analyte level at the time of the ECG measurement, the processing device is to:
label the ECG measurement with the estimated analyte level at the time of the ECG measurement.
18 . The system of claim 13 , wherein to train the machine learning model, the processing device is to:
analyze, using the machine learning model, each of the plurality of ECG measurements that are labeled to generate a corresponding output; and
compare the corresponding output for each of the plurality of ECG measurements that are labeled to the associated label for each of the plurality of ECG measurements that are labeled to update the machine learning model using backpropagation, wherein one or more weight matrices of the machine learning model are adjusted based on the confidence interval associated with each of the plurality of ECG measurements that are labeled.
19 . The system of claim 11 , wherein the machine learning model is one of a convolutional neural network, a recurrent neural network and a combination of a convolutional neural network and a recurrent neural network.
20 . The system of claim 11 , wherein the analyte is one of potassium, magnesium and calcium.