Ensemble-Based Early Detection of Malaria via Explainable ViT-CNN Feature Fusion and SHAP

Area of Science:

• Medical Diagnostics
• Artificial Intelligence in Medicine
• Parasitology

Background:

• Malaria diagnosis is crucial for disease control but current methods are slow and require expertise.
• Automated systems using artificial intelligence (AI) offer potential for rapid and reliable malaria diagnosis.
• Explainable AI (XAI) enhances trust and understanding in AI diagnostic tools.

Purpose of the Study:

• To develop and evaluate a hybrid AI model for accurate malaria diagnosis from microscopic blood smear images.
• To integrate explainable AI (XAI) features for improved transparency and interpretability.
• To assess the speed and efficiency of the proposed AI model compared to traditional methods.

Main Methods:

• A hybrid model combining Vision Transformer (ViT) and EfficientNet Convolutional Neural Network (CNN) features.
• Utilizing SHAP (SHapley Additive exPlanations) for feature importance filtering.
• Retraining significant features with an ensemble of CatBoost, XGBoost, and Logistic Regression.

Main Results:

• The hybrid model combining ViT_Large_16 and EfficientNetB0 achieved 95.61% accuracy.
• The system processed images in 1.6 seconds, approximately 204 times faster than traditional machine learning.
• ViT-based models demonstrated superior ability in capturing intricate details in blood smear images.

Conclusions:

• The proposed hybrid XAI model offers significant speed and explainability advantages for malaria diagnosis.
• The system's high performance suggests its suitability for integration into clinical applications for early infectious disease detection.
• This AI-driven approach can enhance the efficiency and accessibility of malaria diagnostics worldwide.