AI-Based Detection of Neglected Tropical Diseases

The diagnosis of parasitic infectious diseases—often called “neglected tropical diseases”—usually requires highly trained medical personnel to analyze microscopy images for evidence of the disease. These skills are very scarce in the affected areas, and diagnosis can be costly and time-consuming. This creates significant challenges for promptly diagnosing, treating, controlling or even eliminating parasitic diseases for more than one billion people who live mostly in poor, vulnerable and marginalized communities in tropical and sub-tropical areas. Two such diseases with significant health impacts are onchocerciasis and schistosomiasis. This paper discusses a system based on artificial intelligence deep learning to support the automated identification of parasites in microscopic images of samples taken from infected individuals. The system highlights areas within an image that contain parasites using semantic segmentation and different visualizations. The convolutional neural network (CNN) architectures E-Net and U-Net are trained using available image data sets and evaluated in order to determine the optimal parameters for the final system. Finally, an inference is implemented that saves the final result as a digital image highlighting the areas containing parasites. The paper demonstrates that a basic implementation of microscope image analysis for parasitic disease detection is possible using the proposed CNN architectures and the available limited data sets.