HealthMap: Transforming Clinical Time-series into Visual Representation for Predictive Modeling

Accurate mortality prediction in intensive care units (ICUs) is crucial for timely medical interventions and improved patient outcomes. Traditional approaches often rely on recurrent neural networks (RNNs) to model temporal dependencies in time series data. However, RNNs suffer from limitations such as difficulty in capturing complex temporal patterns over extended heterogeneous ICU data. In this study, we propose a vision-based deep learning model that transforms ICU time series data into heatmap representations and utilizes a convolutional neural network (CNN) for mortality estimation. By leveraging spatial feature extraction, our approach captures both local and global temporal trends. Experimental results demonstrate that our vision-based model outperforms traditional RNN-based methods, achieving superior performance, AUROC of 0.82, AUPRC of 0.39, and overall classification accuracy. These findings suggest that heatmap-based representations can effectively encode time series dynamics which makes visual models a promising alternative to sequence models for ICU risk stratification.Clinical relevance- A mortality prediction model employing EHR data in the form of heatmaps, rather than traditional time-series representations, provides clinicians with an intuitive visual mechanism for discerning critical patterns of patient deterioration.