Automated Malaria Ring Form Classification in Blood Smear Images Using Ensemble Parallel Neural Networks.

Area of Science:

• Medical Diagnostics
• Artificial Intelligence
• Computational Biology

Background:

• Manual microscopy for malaria diagnosis is time-consuming and subject to human error.
• Developing automated methods is crucial for efficient and accurate malaria screening.

Purpose of the Study:

• To develop and evaluate an automated binary classification approach for detecting malaria ring-form infections in single-cell images.
• To assess the performance of a parallel neural network framework integrating convolutional neural networks and self-attention mechanisms.

Main Methods:

• A balanced Kaggle dataset of 27,558 erythrocyte crops was used, with images standardized to 128x128 pixels and augmented.
• A dual-branch neural network architecture fused convolutional neural network features with multi-head self-attention.
• Performance was evaluated using 10-fold cross-validation and an independent test set.

Main Results:

• The primary model achieved an ROC-AUC of approximately 0.99 and a peak mean accuracy of 0.9567 during cross-validation.
• On the independent test set, the model reached 0.97 accuracy with a macro F1-score of 0.97.
• Increased model complexity led to a performance decrease, indicating the effectiveness of moderate feature fusion.

Conclusions:

• The proposed AI approach offers a robust and generalizable solution for automated malaria screening, outperforming more complex models.
• While promising, the model requires external validation and further development for direct clinical application due to its reliance on pre-segmented images.