JMIR Publications

ABSTRACT

Background:

Hyperkalemia is a critical condition especially in the intensive care unit. There was no accurate and noninvasive method for recognizing hyperkalemia event from ambulatory electrocardiogram monitor so far.

Objective:

This study aims at improving the accuracy on hyperkalemia predictions from ambulatory electrocardiogram (ECG) monitors through a personalized transfer learning method by training a generic model and fine tuning it with personal data.

Methods:

This is a retrospective cohort study using open-source data from Waveform Database Matched Subset from Medical information Mart from Intensive Care III (MIMIC-III). We included patients with multiple serum potassium test results, and matched ECG data from MIMIC-III database. A one-dimensional convolution neural network based deep learning model is first developed to predict hyperkalemia in a generic population. Once the model achieved a state-of-art performance, it was then utilized in an active transfer learning process to perform patient-adaptive heartbeat classification tasks.

Results:

The results show that by acquiring a few data from each new patient, the personalized model can improve the accuracy of hyperkalemia detection significantly from an average of 0.604 ± 0.211 to 0.980 ± 0.078 when compared with the generic model. The Area Under the receiver operating characteristic Curve level also improved from 0.729 ± 0.240 to 0.945 ± 0.094.

Conclusions:

By utilizing deep transfer learning method, we were able to build a clinical standard model for hyperkalemia detection from ambulatory ECG monitor. These findings could potentially be extended to applications that continuously monitor one's ECG for early alerts of hyperkalemia and avoiding unnecessary blood tests. Clinical Trial: This study were approved by the Institutional Review Board of Chang Gung Medical Foundation (number: 202001217B0, date of approval: 21/07/2020).