DANet a lightweight dilated attention network for malaria parasite detection

Area of Science:

• Medical Diagnostics
• Computer Vision
• Machine Learning

Background:

• Malaria diagnosis relies on blood smear microscopy, which is challenging in resource-limited settings.
• Existing deep learning models for malaria detection often have high computational demands, limiting their practical deployment.
• Accurate and efficient diagnostic tools are crucial for global malaria control efforts.

Purpose of the Study:

• To develop a lightweight and efficient deep learning model for robust malaria parasite detection in red blood cell (RBC) images.
• To address the challenges of subtle color variations, indistinct features, and diverse morphologies in malaria diagnosis.
• To enable practical deployment of automated malaria diagnosis tools in resource-constrained environments.

Main Methods:

• Proposed DANet (Dilated Attention Network), a lightweight model with approximately 2.3 million parameters.
• Utilized a novel dilated attention mechanism to capture contextual information and highlight critical features in low-contrast images.
• Evaluated performance on the NIH Malaria Dataset (27,558 images) and performed 5-fold cross-validation.

Main Results:

• Achieved high performance with an F1-score of 97.86%, accuracy of 97.95%, and AUC-PR of 0.98.
• DANet demonstrated superior efficiency compared to state-of-the-art models, suitable for edge device deployment (e.g., Raspberry Pi 4).
• Grad-CAM visualizations confirmed the model's interpretability and focus on relevant image features.

Conclusions:

• DANet offers a practical, high-performance solution for automated malaria diagnosis, particularly in resource-limited settings.
• The model's lightweight design facilitates deployment on edge devices, enhancing accessibility.
• This research contributes to improving malaria diagnostic capabilities globally.