Dynamic Revision of Deep Learning Model

Dynamic revision of deep learning model (DLM) is a method that use personal pre-annotated electrocardiograms (ECGs) for enhancing the accuracy in patients with multiple visits. This repository contains de-identified data and code for the two parts of this method. The first part is to construct the linear mixed model (LMM) to obtain the parameters for the dynamic revision. The second part is the dynamic revision of DLM prediction, which is illustrated as below:

The DLM predictions via ECGs is revised with the personal best linear unbiased prediction (BLUP) on follow-up ECGs. The personal BLUP is calculated by the parameters from LMMs, the previous DLM predictions, and the previous ground truth. The black box indicated the result of the dynamic revision. The detail of method and results are presented in:

• YS Lou and C Lin, "Dynamic deep learning algorithm prediction for patients with multiple visits in electrocardiogram analysis", submitted to European Heart Journal – Digital Health in 2022.

Requirements

• R (version 3.4.4) and Rstudio (not necessary)
• Rtools (Rtools35.exe)

You need to have Rtools installed to compile the package. Use the above link for the installation of Rtools.

• lme4

You need to have lme4 to construct LMM. You can install lme4 by running the following line in your R console:

package_url <- "https://cran.r-project.org/src/contrib/Archive/lme4/lme4_1.1-21.tar.gz"
install.packages(package_url, repos = NULL, type = "source")

• pROC

You need to have pROC installed to compute the receiver operating characteristic curve (ROC). You can install pROC by running the following line in your R console:

package_url <- "https://cran.r-project.org/src/contrib/Archive/pROC/pROC_1.16.2.tar.gz"
install.packages(package_url, repos = NULL, type = "source")

• cli
• rlang
• pillar
• ggplot2
• cowplot

You need to have ggplot2, cowplot, and its dependencies installed to plot the summary of performance, and you can install these packages by running the following line in your R console:

package_url <- "https://cran.r-project.org/src/contrib/Archive/cli/cli_2.0.2.tar.gz"
install.packages(package_url, repos = NULL, type="source")
package_url <- "https://cran.r-project.org/src/contrib/Archive/rlang/rlang_0.4.11.tar.gz"
install.packages(package_url, repos = NULL, type="source")
package_url <- "https://cran.r-project.org/src/contrib/Archive/pillar/pillar_1.4.4.tar.gz"
install.packages(package_url, repos = NULL, type="source")
package_url <- "https://cran.r-project.org/src/contrib/Archive/ggplot2/ggplot2_3.3.3.tar.gz"
install.packages(package_url,  repos = NULL, type = "source")

package_url <- "https://cran.r-project.org/src/contrib/Archive/cowplot/cowplot_0.9.4.tar.gz"
install.packages(package_url,  repos = NULL, type = "source")

Usage

An example script can be found in 'code/dynamic_revision.R', and the performance is summarized as following:

The functions of constructing LMM and dynamic revision using BLUP can be found in 'code/personal_blup.R'.

Example data

We use de-identified clinical data in the hospital as the example data. There are two data files, 'data/serum_potassium_data.csv' and 'data/ejection_fraction_data.csv'. See here for more information about the data files.

• The two independent comma-separated values (csv) files containing the columns
  • "dataset": four subsets include development, tuning, internal validation, and external validation sets;
  • "exam_id": id used for internal usages;
  • "follow_up": the visit times of patients;
  • "diff_time": the difference hours between the first examinations and follow-up examinations;
  • "label": the corresponding annotations of serum potassium or ejection fraction with ECGs;
  • "DLM_direct_pred": the direct predictions of DLM via ECG;
  • "DLM_dynamic_pred": the dynamic revision of DLM predictions using personal BLUP;

How to cite

If you use this code in your work, please cite.