DCTL

DCTL applies deep learning-based cycle transfer learning to classify apicomplexan parasites in microscopic images by exploiting morphological similarity between parasite shapes and macroscopic objects.

Key Features:

• Deep Cycle Transfer Learning: Links microscopic parasite morphologies to macroscopic object images to transfer visual features for classification.
• Deep learning microscopic image analysis: Uses deep learning for automated classification of microscopy images containing parasites.
• Morphological mapping: Leverages shape correspondences for Toxoplasma (banana-shaped), Plasmodium (ring-shaped), Babesia (pear-shaped), and erythrocytes (apple) by mapping to bananas, rings, pears, and apples respectively.
• Empirical validation dataset: Evaluated on a dataset of 24,358 microscopic images containing various apicomplexan parasites.
• Performance metrics: Reported average classification accuracy of 95.7% and area under the curve (AUC) of 0.995 across parasite types.

Scientific Applications:

• Automated parasite detection: Detection and classification of apicomplexan parasites including Toxoplasma, Plasmodium, and Babesia in microscopy images.
• Diagnostic screening support: Support for high-throughput microscopy analysis to assist clinical and research workflows.
• Population-scale surveillance: Facilitation of large-scale screening and surveillance efforts, including use cases in resource-limited settings.

Methodology:

DCTL implements a deep learning-based cycle transfer learning approach that maps parasite microscopic morphologies to macroscopic object images and was trained and evaluated on 24,358 microscopic images, yielding reported accuracy and AUC metrics.