Related Experiment Video
Updated: Jul 30, 2026
08:04
Fully Automated Leg Tracking in Freely Moving Insects using Feature Learning Leg Segmentation and Tracking FLLIT
Published on: April 23, 2020
6.7K
Deep learning model meets community-based surveillance of acute flaccid paralysis
Gelan Ayana1,2,3, Kokeb Dese1,2,3, Hundessa Daba Nemomssa1,2,3
1School of Biomedical Engineering, Jimma Institute of Technology, Jimma University, Jimma, 378, Ethiopia.
Infectious Disease Modelling
|December 25, 2024
Summary
A new deep learning model using transfer learning on mobile phone images significantly improves acute flaccid paralysis (AFP) surveillance in Ethiopia. This AI approach enhances early detection of potential poliovirus, aiding global disease eradication efforts.
Area of Science:
- Artificial Intelligence in Public Health
- Deep Learning for Disease Surveillance
- Image Recognition in Epidemiology
Background:
- Acute flaccid paralysis (AFP) case surveillance is crucial for early poliovirus detection, especially in endemic regions like Ethiopia.
- Existing community-based surveillance systems in Ethiopia face challenges with delayed detection and communication.
- There is a need for innovative, scalable solutions to enhance AFP surveillance in low-resource settings.
Purpose of the Study:
- To propose and evaluate a deep learning model for AFP surveillance using transfer learning on mobile phone-captured images.
- To assess the model's performance against traditional deep learning methods and vision transformers trained from scratch.
- To demonstrate a scalable AI-driven approach for improving community-based disease surveillance.
Main Methods:
- Implementation of a vision transformer model pretrained on ImageNet for transfer learning.
- Utilizing images collected from community key informants in Ethiopia via mobile phones.
- Comparative analysis against convolutional neural network (CNN) models and non-pretrained vision transformers.
Main Results:
- The proposed transfer learning model achieved superior performance metrics, including accuracy, F1-score, precision, recall, and AUC.
- The model demonstrated the highest average AUC of 0.870 ± 0.01, statistically outperforming alternative approaches (P < 0.001).
- The AI model effectively bridges community reporting with health system response for enhanced surveillance.
Conclusions:
- Deep learning with transfer learning offers a powerful and scalable tool for improving AFP surveillance using mobile-captured images in low-resource settings.
- The study highlights the potential of AI to strengthen global health initiatives and disease eradication strategies.
- Future work should focus on improving image data quality and establishing dedicated platforms for data management and continuous learning.
