DengueNet: Dengue Prediction using Spatiotemporal Satellite Imagery for Resource-Limited Countries

TL;DR

DengueNet tackles dengue outbreak forecasting in resource-limited countries by leveraging publicly available Sentinel-2 satellite imagery. It introduces a modular architecture that fuses Radiomics features and Vision Transformer representations with dual LSTMs to model spatiotemporal dynamics on an epi-week basis. The authors present a scalable Sentinel-2 data extraction framework via Sentinel Hub, include a cloud/shadow removal step and careful band selection, and demonstrate on five Colombian municipalities that the approach yields competitive accuracy (MAE $=43.92$) compared to case-only baselines. The work emphasizes public health impact and equity, offering a dockerized, low-resource solution that can be transferred to other LMIC contexts while highlighting areas for bias mitigation and fairness in deployment.

Abstract

Dengue fever presents a substantial challenge in developing countries where sanitation infrastructure is inadequate. The absence of comprehensive healthcare systems exacerbates the severity of dengue infections, potentially leading to life-threatening circumstances. Rapid response to dengue outbreaks is also challenging due to limited information exchange and integration. While timely dengue outbreak forecasts have the potential to prevent such outbreaks, the majority of dengue prediction studies have predominantly relied on data that impose significant burdens on individual countries for collection. In this study, our aim is to improve health equity in resource-constrained countries by exploring the effectiveness of high-resolution satellite imagery as a nontraditional and readily accessible data source. By leveraging the wealth of publicly available and easily obtainable satellite imagery, we present a scalable satellite extraction framework based on Sentinel Hub, a cloud-based computing platform. Furthermore, we introduce DengueNet, an innovative architecture that combines Vision Transformer, Radiomics, and Long Short-term Memory to extract and integrate spatiotemporal features from satellite images. This enables dengue predictions on an epi-week basis. To evaluate the effectiveness of our proposed method, we conducted experiments on five municipalities in Colombia. We utilized a dataset comprising 780 high-resolution Sentinel-2 satellite images for training and evaluation. The performance of DengueNet was assessed using the mean absolute error (MAE) metric. Across the five municipalities, DengueNet achieved an average MAE of 43.92. Our findings strongly support the efficacy of satellite imagery as a valuable resource for dengue prediction, particularly in informing public health policies within countries where manually collected data is scarce and dengue virus prevalence is severe.

Figures (7)

• Figure 1: DengueNet model architecture takes in weekly satellite imagery and dengue cases ${y}$ as input for predicting $\hat{y}$ (m/px: meters per pixel; RGB: red, green and blue bands; SWIR: short wave infrared spectrum band; ViT: Vision Transformer; LSTM: Long Short-Term memory; MLP: Multilayer Perceptron). The LSTM module consists of three stacked standard LSTM layers.
• Figure 2: Municipality-level dengue case numbers and geographic locations. (a) Dengue cases from 2016 to 2018 were obtained from the SIVIGILA database for the top five affected municipalities in Colombia. (b) Geographic locations from satellite imagery for each municipality.
• Figure 3: Gray-scale satellite band images captured by Sentinel-2 using different wavelengths.
• Figure 4: Average Pearson's correlation of the 12 bands for the Sentinel-2 satellite images across five Colombian municipalities in the training set from 2016 to 2018. The majority of correlations are statistically significant (p < 0.001).
• Figure 5: Stages involved in the cloud and cloud shadow removal module. The average tiles are generated using the normal tiles in the samples (CCS: cloud and cloud shadow).