# GIS-based neural network framework for zoonotic cutaneous leishmaniasis risk mapping in Western Iran

## TL;DR

This paper introduces a GeoAI framework using 3D-CNNs to map and predict zoonotic cutaneous leishmaniasis risk in Western Iran, showing how temperature affects disease spread and projecting future risk shifts.

## Contribution

The novel use of 3D-CNNs to model spatiotemporal transmission dynamics of ZCL is a key advancement.

## Key findings

- 3D-CNNs outperformed other models in capturing spatial and temporal patterns of ZCL risk.
- Warmer western and southern regions showed higher ZCL risk compared to cooler northern and eastern areas.
- By 2030, ZCL risk is projected to decrease in the west and increase in the south.

## Abstract

This study presents a Geospatial Artificial Intelligence (GeoAI) framework for high-resolution Zoonotic Cutaneous Leishmaniasis (ZCL) risk mapping, correlation analysis, and scenario-based projection, integrating geographic information systems (GIS), remote sensing, and neural network architecture. Historical disease maps and multi-temporal satellite-derived environmental layers were jointly modeled using a multilayer perceptron (MLP), two-dimensional convolutional neural networks (2D-CNNs), and three-dimensional CNNs (3D-CNNs). The principal methodological contribution is the implementation of a 3D-CNN, which enables explicit learning of spatiotemporal transmission dynamics. Environmental–disease relationship analyses, based on Pearson coefficients and regression models, identified temperature as the dominant positive environmental driver of ZCL risk. Model performance assessment using root mean square error (RMSE), mean absolute error (MAE), and the area under the receiver operating characteristic curve (AUC) indicates that the 3D-CNN consistently outperforms alternative architectures in capturing complex spatial and temporal patterns. Elevated risk was concentrated in warmer western and southern regions, whereas cooler northern and eastern mountainous areas exhibited lower susceptibility. By 2030, ZCL risk is projected to undergo a spatial shift, with risk decreasing in western regions and intensifying in southern areas, which has direct implications for targeted surveillance and intervention efforts.