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

Computer-aided diagnosis in healthcare involves collecting and modeling patients' comprehensive health-related information from their Electronic Health Records (EHRs). Recently, the extensive acquisition of EHRs and scalable architectures of deep learning methods such as convolutional neural networks (CNNs) are generating robust healthcare assistance. Yet, the laborious pre-processing steps on EHR data and domain-specific feature engineering hinder the results of the model's interpretability and transparency. Moreover, conducting obscure feature engineering tasks without domain knowledge may lead to the loss of relevant features, limiting the model's ability to make reliable predictions. To alleviate these challenges, we proposes a method to represent the tabular EHR data in 2D images without leveraging any pre-processing or data cleaning tasks, which results in a generalized and more interpretable visualization of EHR tabular data. In addition, we evaluate the proposed method by predicting a cardiovascular disease, Sudden Cardiac Arrest (SCA), using pre-trained deep CNN models. The results demonstrate that the proposed method can perform highly in the SCA prediction without any missing value imputation and allow more transparency with less human expert intervention.