Enhanced YOLO-based framework and benchmarking for automated Plasmodium vivax detection.

Area of Science:

• Medical Diagnostics
• Parasitology
• Artificial Intelligence in Healthcare

Background:

• Malaria remains a significant global health threat, particularly in sub-Saharan Africa and Southeast Asia, despite advances in life expectancy.
• Traditional malaria diagnostics (microscopy, RDTs) have limitations like reduced sensitivity and operator dependency, necessitating advanced automated solutions.
• Artificial intelligence (AI) and deep learning (DL) offer potential for automated parasite identification, addressing current diagnostic challenges.

Purpose of the Study:

• To comparatively evaluate the performance of various YOLO (You Only Look Once) deep learning variants for malaria detection.
• To identify the most suitable AI architecture for precise and efficient identification of malaria parasites, specifically Plasmodium vivax.
• To propose a novel DL framework for enhanced malaria diagnosis in diverse clinical settings.

Main Methods:

• Comparative analysis of YOLOv3, cascade v3, scaled v4, v5, and v8 object detection models.
• Evaluation using key performance metrics: precision, accuracy, F1-score, recall, and mean Average Precision (mAP).
• Development of a novel DL framework integrating YOLOv3 with a MobileNetV2 backbone and a Transformed Convolutional Layer (TCL).

Main Results:

• The novel DL framework combining YOLOv3, MobileNetV2, and TCL demonstrated high efficacy, particularly for detecting Plasmodium vivax in dense smear images.
• The chosen AI architecture exhibited multi-scale texture sensitivity and efficient feature extraction, crucial for analyzing complex parasitic samples.
• Performance metrics indicated the superiority of the proposed framework over standard YOLO variants for malaria parasite detection.

Conclusions:

• The developed AI-driven diagnostic framework offers a robust and scalable solution for early malaria detection.
• This approach provides valuable guidance for implementing AI-based tools in varied clinical and field settings for improved malaria management.
• The study highlights the potential of deep learning, specifically the proposed YOLOv3-based model, to enhance malaria diagnosis accuracy and efficiency.