An Explaınable AI-Drıven Approach for Automated Malarıa Detectıon from Blood Smear Images

İngilizce (OpenAlex)

Malaria remains one of the world's most critical parasitic diseases, posing a serious threat to global public health, particularly in developing regions.Traditional microscopic diagnostic methods, considered the gold standard, are labor-intensive and time-consuming, as well as heavily dependent on human expertise.This study aims to improve the accuracy and automation level of malaria detection by using advanced deep learning models such as convolutional neural networks (ResNet50, EfficientNetB0, InceptionResNetV2, and Xception) and Vision Transformer-based architectures (ViT-Base, Swin Transformer, DeiT, and PVT).The models were trained and tested on a microscopic blood image dataset containing four different types of malaria using the 5-fold crossvalidation method.Among the CNN models, Xception achieved the highest accuracy rate of 98.10%, while Swin Transformer showed similar success among the Transformer-based models.Furthermore, the Gradient-Weighted Class Activation Mapping (Grad-CAM) technique was applied to visually highlight the most influential regions in the model's decision-making process and to increase model interpretability.The findings show that combining transformer-based architectures with explainable artificial intelligence significantly improves both classification performance and model transparency.This enhances the clinical reliability of automated malaria diagnosis systems, strengthening their potential for use in real-world applications