AI and EO as Innovative Methods for Monitoring West Nile Virus Spread (AIDEO)

AI and EO as Innovative Methods for Monitoring West Nile Virus Spread (AIDEO) is being developed by the Istituto Zooprofilattico Sperimentale dell’Abruzzo e del Molise “G. Caporale”, a veterinary public health institution that has an established international track record in the surveillance, diagnosis, epidemiology, modelling, molecular epidemiology of Vector Borne Diseases (VBDs), AImageLab, that is a research laboratory of the Dipartimento di Ingegneria “Enzo Ferrari” at the University of Modena and Reggio Emilia with extensive experience in Computer Vision, Pattern Recognition, Machine Learning and Artificial Intelligence, Progressive Systems, that delivers solutions to simplify Earth Observation data exploitation and brings significant expertise and experience to the consortium based on years of collaboration with ESA and on-site presence at ESRIN, and REMEDIA Italia, that has relevant experience in designing and realising printed, web, multimedia and technology enhanced scientific communication projects, systems and tools developed inside the Earth Observation Department of ESA (ESRIN).

Aim of the project is to develop an innovative, scalable and accurate process to produce West Nile Disease (WND) risk maps, using EO data and specific AI algorithms.

Vector-borne diseases (VBDs) are an important threat with an increasing impact on public health due to wider geographic range of occurrence and higher incidences. West Nile virus (WNV) is one of the most spread zoonotic VBD in Italy and Europe. Identifying suitable environmental conditions across large areas containing multiple species of potential hosts and vectors can be difficult. The recent and massive availability of Earth Observation (EO) data and the continuous development of innovative Artificial Intelligence (AI) methods can be of great help to automatically identify patterns in big datasets and to make highly accurate predictions. Our project aims to develop an innovative, scalable and accurate process to produce West Nile Disease (WND) risk maps, using EO data and specific AI algorithms. Using historical ground truth data of WND cases and EO data derived from different sources (e.g. Sentinel-2, Sentinel-3, PROBA-V, etc.), a learning architecture, based on Convolutional Neural Network (CNN) and Graph Theory, will be applied on ground truth WND cases and satellite images and tested. This process will produce AI based risk maps that will be then compared with classical statistical methods to evaluate the degree of improvement in forecasting the disease occurrence and spread.

Knowledge acquired with this project can be potentially used to define intervention priorities within national diseases surveillance plans. Moreover, the definition and development of algorithms working on available and frequent satellite images could be applied in early warning systems not developed so far, and could be integrated into the Information Systems of the Italian Ministry of Health and made available to other interested stakeholders. This work will therefore lay the basis for a future early warning system that could alert public authorities when climatic and environmental conditions become favourable to the onset and spread of the disease.

This will be achieved in three key phases:

Phase 1: Definition of requirements

Information regarding EO data to be used, criteria to select ground truth data, temporal interval to be analysed and different Deep Neural Network models will be evaluated and defined. Selection criteria and preparation of remotely sensed products will then be investigated, considering data from multiple sources, various sensors, spectral bands, spatial resolutions and revisit times. WND and EO data will be selected to guarantee a correct spatial and temporal representation of the last ten-years epidemics.

Phase 2: Data retrieval and processing

WND cases will be extracted from the official repository of the Italian Ministry of Health (National Information System of Animal Disease Notification – SIMAN), integrated with laboratory data coming from the national veterinary laboratories, validated and selected, in space and time, according to the requirements defined in phase 1. WND ground truth outbreaks will be split in different datasets that will be used to train and test the DNN model, then fine-tune the model and hence make predictions and evaluate the overall accuracy. Selected EO data will be collected from different sources and stored in a centralised system where they will be organised and pre-processed according to the requirements defined in phase 1. Classical statistical models for WND spread (suitability analysis, logistic regression, etc.) will be developed to be compared with AI model performance.

Phase 3: Train, fine-tuning and validation of the AI model

AI models/algorithms for the analysis and prediction of WND “behaviour” will be developed and parameters estimated. Graph-based DNN models will be explored for merging geo-referenced local sites information with satellite images, the latter being processed through Convolutional Neural Networks (pre-trained or trained from scratch). Temporal deep models (e.g. RNN – Recurrent Neural Networks, LSTM – Long-short term memory) will then be employed for an effective forecasting of the behaviour based on EO data.

The accuracy of the chosen model will then be evaluated together with the need to include additional data or to change the train model hyper-parameters. We will hence produce the final model that will be compared with the classical statistical models developed in phase 2.

Dissemination of information and project results will last for the entire duration of the project and will be made available to stakeholders, relevant institutions, organisations and individuals through workshop and congress presentations, publications in peer reviewed journals, websites.