From Detection to Motion-Based Classification: A Two-Stage Approach for T. cruzi Identification in Video Sequences.

Area of Science:

• Medical Diagnostics
• Computer Vision
• Parasitology

Background:

• Chagas disease, caused by Trypanosoma cruzi (T. cruzi), is a major Latin American health issue.
• Current manual microscopy diagnostics for T. cruzi are insensitive, subjective, and perform poorly in suboptimal conditions.

Purpose of the Study:

• To develop and validate a novel computer vision framework for automated T. cruzi detection in microscopic videos.
• To leverage parasite motion analysis and deep learning to overcome limitations of traditional diagnostic methods.

Main Methods:

• A motion-based detection pipeline using frame differencing, morphological processing, and DBSCAN clustering was applied to microscopic videos.
• Deep learning models (MobileNetV2, YOLOv5, YOLOv8) were trained on motion-identified patches for T. cruzi classification and detection.

Main Results:

• MobileNetV2 achieved 99.63% accuracy, 100% precision, and 99.12% recall.
• YOLOv5-Nano and YOLOv8-Nano demonstrated excellent detection performance on unseen data.
• The framework effectively handled noisy backgrounds, uneven illumination, and low contrast.

Conclusions:

• The dual-stage computer vision framework provides a practical and computationally efficient solution for automated Chagas disease diagnosis.
• This technology is particularly beneficial for resource-limited laboratories facing poor imaging quality and diagnostic challenges.