Towards effective and efficient machine learning models for schistosomiasis diagnosis in microscopic images

Area of Science:

• Parasitology
• Medical Imaging
• Artificial Intelligence

Background:

• Schistosomiasis is a major health issue in tropical regions, primarily diagnosed through time-consuming and error-prone microscopic fecal examinations.
• Traditional diagnosis requires specialized training and is not always accurate, hindering effective public health interventions.

Purpose of the Study:

• To develop an automated system for detecting Schistosoma mansoni eggs using deep learning (DL) and machine learning (ML) techniques.
• To improve the speed and accuracy of schistosomiasis diagnosis, aiding healthcare professionals in endemic areas.

Main Methods:

• A dataset of 1100 images from the Kato-Katz technique was created and annotated by parasitologists.
• A deep learning object detection model (Faster R-CNN with ResNet-50) combined with HOG features and classical ML was employed.
• An integrated approach using a voting scheme of ML models was proposed to minimize false positives and negatives.

Main Results:

• The Faster R-CNN model achieved an Average Precision of 0.884 at IoU=0.50 on the annotated dataset.
• The proposed system demonstrated superior performance compared to other detection models and a commercial tool.
• Comparative analysis confirmed the system's promising results and applicability in public health settings.

Conclusions:

• The developed AI system offers a viable, automated solution for schistosomiasis diagnosis, potentially enhancing diagnostic speed and accuracy.
• This approach can significantly assist public health systems, like Brazil's SUS, in managing schistosomiasis in endemic regions.
• Further integration and validation of the AI system can lead to improved patient outcomes and disease control efforts.