Iranian Journal of Biomedical Engineering (IJBME)
نشریه علمی مهندسی پزشکی زیستی

Predicting Heart Failure Mortality Using Registry Data: A Novel Data-to-Image Transformation and Ensemble of CNNs

Document Type : Full Research Paper

Authors

Hadi Sabahi 1
mansour vali 1
Davood Shafie 2

1 Department of Biomedical Engineering, K. N. Toosi University of Technology, Tehran, Iran

2 Heart Failure Research Center, Cardiovascular Research Institute, Isfahan University of Medical Sciences, Isfahan, Iran

https://doi.org/10.22041/ijbme.2025.2047242.1936
Abstract
Heart failure (HF) is a cardiovascular disorder with a substantial mortality rate, which can be assessed using registry data. This work aims to predict 6-month mortality (6-MM) and 12-month mortality (12-MM) of HF patients using registry data from before hospitalization, during hospitalization, and discharge phase. The data comprise 3968 HF records sourced from Persian Registry Of cardioVascular diseasE (PROVE)/HF registry.

We proposed an ensemble model employing popular Convolutional Neural Networks (CNNs) to predict the patients’ survival status. However, the HF registry, being tabular data, isn’t suitable for leveraging the benefits of CNNs. To address this challenge, we introduced a novel data-to-image transformation using self-organizing maps (SOMs) to convert HF data samples into images. We compared our proposed model with an existing data-to-image transformation method, as well as with multiple baseline classifiers that utilize the HF tabular data along with down-sampling.

The proposed model, using SOM for data-to-image transformation, outperformed all others, achieving the highest area under receiver operating characteristic (ROC) curve for both 6-MM and 12-MM predictions, with rates of 74.22% and 75.2%, respectively.

In conclusion, the proposed model, along with SOM for data-to-image transformation, effectively predicts the survival status of HF patients utilizing registry data.

Keywords

Heart Failure Mortality prediction
Imbalanced data
Ensemble model of CNNs
Data-to-image transformation
Self-Organizing Maps (SOM)

Subjects

Iranian Journal of Biomedical Engineering (IJBME)
Volume 18, Issue 2
Summer 2024
Pages 139-155

  • Receive Date 02 December 2024
  • Revise Date 03 February 2025
  • Accept Date 06 March 2025

Home

Contact Us