Akademik Veri Yönetim Sistemi
33rd Telecommunications Forum (TELFOR), Belgrade, Sırbistan, 25 - 26 Kasım 2025, ss.1-4, (Tam Metin Bildiri)
Abstract—The deadly malaria parasite, transmitted by female Anopheles mosquitoes, causes approximately 200 million infections and 400,000 deaths annually. Diagnosis through microscopic examination of blood samples by microbiologists is time-consuming and faces significant challenges in some regions due to economic constraints, lack of equipment, and the need for specialized laboratories and trained personnel. Artificial intelligence (AI) models play a critical role in early and accurate detection of malaria parasites, which is essential for effective treatment. In this study, Convolutional Neural Network (CNN) and Vision Transformer (ViT)-based models are proposed for the detection of Plasmodium falciparum, one of the five species causing malaria in humans. Additionally, a VGG-16 model with weight fine-tuning (Wft) was implemented using transfer learning, and its performance was compared with the proposed models. The accuracy performances of the developed models were tested on three different datasets. In the first dataset consisting of Giemsa stained microscopic images, the VGG-16 Wft model achieved the highest accuracy of 97.06%, while in the second dataset, CNN, CNN-ViT and VGG-16 models achieved the best accuracy of 98.18%. Finally, in the third dataset extended with gametocyte and trophozoite cell images, the VGG-16 Wft model showed the highest accuracy of 96.85%. These models aim to reduce the workload of healthcare professionals and offer a more accurate and cost-effective alternative to traditional diagnostic methods