ROENet: A ResNet-Based Output Ensemble for Malaria Parasite Classification.

Area of Science:

• Medical diagnostics
• Computational biology
• Parasitology

Background:

• Malaria diagnosis is time-consuming and can be inaccurate.
• Current methods, including CNNs, show potential but room for improvement exists.
• Automated classification of malaria parasites is crucial for efficient diagnosis.

Purpose of the Study:

• To develop an improved automated method for malaria parasite classification.
• To enhance diagnostic accuracy and efficiency in malaria detection.

Main Methods:

• Proposed ROENet, a novel method for malaria parasite classification on blood smears.
• Utilized a pretrained ResNet-18 backbone.
• Employed an ensemble of three randomized neural networks (RNNs): random vector functional link (RVFL), Schmidt neural network (SNN), and extreme learning machine (ELM).

Main Results:

• Achieved high performance metrics: 96.68% specificity, 95.69% F1 score, 94.79% sensitivity, and 95.73% accuracy via five-fold cross-validation.
• Demonstrated superior performance compared to other state-of-the-art methods.

Conclusions:

• ROENet offers a significant advancement in automated malaria parasite classification.
• The proposed method provides highly accurate and reliable results, surpassing existing techniques.
• ROENet holds promise for improving malaria diagnosis and patient outcomes.