Hybrid Capsule Network for Precise and Interpretable Detection of Malaria Parasites in Blood Smear Images

Bader Alawfi^*

• Department of Medical Laboratory Technology, College of Applied Medical Sciences, Taibah University, Madinah, Saudi Arabia

An expedient and precise diagnosis is crucial in underprivileged regions to execute effective treatment options for malaria patients. Current blends of CNN and Capsule Networks frequently experience elevated computational expenses and constrained generalizability. This paper presents the Hybrid Capsule Network (Hybrid CapNet). This innovative lightweight architecture integrates CNN feature extraction with dynamic capsule routing, specifically engineered for the identification of malaria parasites and the classification of their life cycle stages. This study is distinguished by the implementation of a cohesive composite loss function that amalgamates margin, focal, reconstruction, and regression losses, thereby improving classification accuracy, spatial localization, and resilience to class imbalance and annotation noise.In contrast to conventional models, our approach achieves elevated accuracy with minimal processing demands (1.35M parameters, 0.26 GFLOPs), making it suitable for mobile diagnostic applications. The model attains exceptional performance on four malaria datasets (MP-IDB, MP-IDB2, IML-Malaria, MD-2019), achieving up to 100\% accuracy in multiclass classification and surpassing baseline CNN models in cross-dataset assessments. Interpretability is achieved via Grad-CAM displays, confirming the model's emphasis on biologically pertinent areas. Hybrid CapNet offers a pragmatic, accurate, and interpretable AI-based approach for malaria diagnosis in actual clinical settings.