[Paper Review] DengueNet: Dengue Prediction using Spatiotemporal Satellite Imagery for Resource-Limited Countries

Kuan-Ting Kuo, Dana Moukheiber+10 more|arXiv (Cornell University)|Jan 20, 2024

Mosquito-borne diseases and controlMedicine3 citations

TL;DR

DengueNet is a novel deep learning model that leverages spatiotemporal satellite imagery from Sentinel-2 to predict dengue outbreaks in resource-limited countries. By combining Vision Transformers, radiomics-based feature extraction, and stacked LSTMs, it achieves an average MAE of 43.92 ± 42.19 across five Colombian municipalities, demonstrating the feasibility of satellite data for scalable, equitable dengue forecasting in low-income settings.

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.

Motivation & Objective

• To address the challenge of limited access to high-quality epidemiological data in low- and middle-income countries (LMICs) by leveraging publicly available satellite imagery.
• To develop a scalable, low-resource satellite data extraction and processing framework for epidemiological forecasting.
• To improve health equity by enabling early dengue outbreak prediction without reliance on costly or inconsistent field-collected data.
• To evaluate the effectiveness of a hybrid model combining Vision Transformers and radiomics for spatiotemporal feature extraction from satellite images.
• To demonstrate that satellite-based forecasting can achieve performance comparable to case-only models, supporting early public health intervention.

Proposed method

• A cloud- and shadow-removal (CSR) preprocessing pipeline is applied to Sentinel-2 satellite images using Sentinel Hub’s cloud-based platform for scalable data access.
• Band selection is performed using Pearson correlation to reduce redundancy and retain informative spectral bands (e.g., RGB, SWIR).
• Two parallel feature extractors are used: a Vision Transformer (ViT) for self-attention-based spatial feature learning and a radiomics pipeline for handcrafted texture and intensity features.
• Extracted features from both branches are concatenated and fed into a stacked Long Short-Term Memory (LSTM) network to model temporal dependencies across weekly satellite images.
• The model is trained end-to-end to predict weekly dengue case counts using mean absolute error (MAE) as the loss function.
• A Dockerized version of the satellite extraction pipeline ensures reproducibility and scalability for deployment in LMICs.

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

Experimental results

Research questions

• RQ1 Can high-resolution, publicly available satellite imagery be effectively used to predict dengue outbreaks in resource-limited countries?
• RQ2 How does a hybrid architecture combining Vision Transformers and radiomics improve spatiotemporal feature extraction for dengue forecasting compared to traditional methods?
• RQ3 What is the predictive performance of DengueNet across diverse urban environments in Colombia, and how does it vary by location?
• RQ4 To what extent can satellite-derived features alone support accurate dengue forecasting without relying on ground-truth meteorological or clinical data?
• RQ5 Can the proposed framework be deployed with minimal computational resources in low-income settings to support equitable public health decision-making?

Key findings

• DengueNet achieved an average mean absolute error (MAE) of 43.92 ± 42.19 across five municipalities in Colombia, demonstrating robust predictive performance.
• The lowest MAE was recorded in Ibagué (5.67 ± 0.18), indicating strong model generalization in certain urban environments.
• The highest MAE was observed in Cali (113.65 ± 0.08), highlighting challenges in high-urban-density or complex terrain areas.
• The model’s performance using only satellite imagery was comparable to that of a baseline LSTM model trained exclusively on dengue case data.
• The integration of cloud and shadow removal (CSR) significantly improved feature quality, as shown by performance gains in ablation studies.
• The Dockerized satellite extraction pipeline enables reproducible, scalable deployment in low-resource settings, supporting long-term public health monitoring.

(a) Dengue cases