Enhancing malaria detection and classification using convolutional neural networks-vision transformer architecture

Abstract

Malaria remains a significant global health challenge, particularly in sub-Saharan Africa. Despite advancements in treatment and prevention, malaria continues to cause substantial morbidity and mortality, particularly among vulnerable populations such as children and pregnant women. Although effective, traditional diagnostic methods, such as microscopy, are time-consuming and require skilled personnel prone to human error, leading to delays in diagnosis and treatment. More so, existing machine learning models used in malaria detection and classification have low performance and overfitting issues. This study presents an enhanced malaria detection and classification model using an ensemble of Convolutional Neural Networks (CNN) and Vision Transformers (ViT). The proposed ensemble model, which combines CNN and ViT, outperforms each individual model, achieving an accuracy of 99.64%, precision of 99.23%, recall of 99.75%, F1 score of 99.51%, and a cross-entropy loss of 0.01. The proposed model demonstrated superior performance compared to those reported in the literature. These results highlight the potential of the CNN-ViT ensemble model for accurate and reliable malaria detection, offering a significant improvement over existing methods.