<![CDATA[Malaria diagnostic automation faces critical challenges, including severe class imbalance with ratios of up to 54:1, limited datasets containing 200 to 500 images, and computational inefficiency resulting from the need to train separate models for each detection-classification combination. This study developed a multi-model framework with a shared classification architecture that trains classification models once on ground-truth crops and reuses them across all detectors. The framework systematically evaluated three YOLO Medium architectures for parasite detection and six CNN architectures for lifecycle and species classification across four complementary malaria datasets totaling 1,544 microscopy images. Detection achieved mAP@50 scores ranging from 70.84% to 96.27%, with high recall values of 71.05% to 93.12% minimizing missed parasite detections. Classification results demonstrated the importance of dataset-dependent model selection, with parameter-efficient EfficientNet models containing 5.3M to 9.2M parameters consistently outperforming ResNet variants with up to 44.5M parameters. EfficientNet-B1 achieved accuracies of 91.51% on the IML Lifecycle dataset and 98.28% on the MP-IDB Species dataset, while EfficientNet-B0 achieved 86.45% on the multi-patient MD-2019 dataset. ResNet50 achieved 96.13% accuracy on severely imbalanced MP-IDB Stages dataset. Focal Loss optimization with alpha = 1.0 and gamma = 1.5 enabled robust minority-class performance, achieving F1-scores between 0.44 and 1.00 on ultra-minority classes and demonstrating effective handling of class imbalance. The compact models, with sizes ranging from 46 MB to 89 MB, enable practical deployment on resource-constrained hardware.]]>
Copyrights © 2026
