Visual Representation of Tabular Electronic Health Records for Predicting Sudden Cardiac Arrest.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Machine Learning for Clinical Decision Support

Background:

• Electronic Health Records (EHRs) are crucial for computer-aided diagnosis, but their complex nature and extensive preprocessing requirements hinder model interpretability and transparency.
• Current deep learning models, like Convolutional Neural Networks (CNNs), offer scalable solutions but often struggle with the laborious feature engineering needed for EHR data.
• Lack of domain expertise in feature engineering can lead to loss of critical information, impacting predictive model reliability.

Purpose of the Study:

• To propose a novel method for representing tabular EHR data as 2D images, bypassing traditional preprocessing and feature engineering steps.
• To enhance the interpretability and transparency of machine learning models applied to EHR data.
• To evaluate the efficacy of this image-based representation for predicting cardiovascular diseases, specifically Sudden Cardiac Arrest (SCA).

Main Methods:

• Developed a technique to convert tabular EHR data into 2D image representations without requiring data cleaning or imputation.
• Utilized pre-trained deep CNN models to analyze the generated 2D EHR images for disease prediction.
• Focused on predicting Sudden Cardiac Arrest (SCA) as a case study for cardiovascular disease prediction.

Main Results:

• The proposed method successfully generated interpretable 2D visualizations of EHR data.
• The image-based EHR representation achieved high performance in predicting Sudden Cardiac Arrest (SCA).
• The approach demonstrated effectiveness without missing value imputation, offering greater transparency and reduced need for human expert intervention.

Conclusions:

• The novel method offers a generalized and interpretable approach to visualizing EHR data for machine learning applications.
• This technique significantly enhances the prediction of Sudden Cardiac Arrest (SCA), showcasing its clinical utility.
• The findings suggest a promising direction for improving transparency and reducing manual effort in AI-driven healthcare diagnostics.