Explainable Vision Transformer-Based Automated Malaria Detection from Blood Smear Images: Toward Intelligent Nanotechnology-Enabled Diagnostic Systems

Malaria remains one of the world's most significant parasitic diseases, requiring rapid and accurate diagnosis to reduce mortality. Recent advances in digital microscopy, artificial intelligence, and nanotechnology-enabled diagnostic platforms have created new opportunities for automated malaria detection. In particular, nano-enabled biosensors and lab-on-chip technologies increasingly rely on intelligent image analysis algorithms to improve diagnostic performance. In this study, convolutional neural network architectures (ResNet50, EfficientNetB0, InceptionResNetV2, and Xception) and Vision Transformer models (ViT-Base, Swin Transformer, DeiT, and Pyramid Vision Transformer) were evaluated for multiclass malaria detection using microscopic blood smear images. A five-fold cross-validation strategy was employed, and model interpretability was enhanced using Gradient-weighted Class Activation Mapping (Grad-CAM). Among the CNN models, Xception achieved an accuracy of 98.10%, while the Swin Transformer demonstrated comparable performance among transformer-based architectures. The results indicate that combining explainable artificial intelligence with transformer-based models improves both classification accuracy and model transparency, supporting their future integration into nano-enabled diagnostic systems, intelligent digital microscopy, and point-of-care biomedical devices.