Parasitisation prediction in malarial cell images using transfer learning- driven vision transformers
S. Sinha, S. Sinha · IET conference proceedings. · 2023
The World Malaria Report 2022 stated that nearly 247 million cases of malaria and 619,000 deaths were reported annually as of 2021. This emphasizes on the tremendous impact the disease has, especially in third-world countries. Through efficient and timely diagnosis, the disease could be accurately identified and suitable lines-of-treatment could be initiated. The requirement for systematic diagnosis benchmarked the arrival of the computer-aided diagnosis landscape through imaging and Machine Learning (ML) techniques. Deep learning-based backbones are particularly efficient for image classification and disease detection. Through this paper, we propose a Deep Neural Network (DNN) backbone VGG16 for extracting the local and global features of blood smear cell images marked with the Giemsa stain, followed by a Vision Transformer (ViT) neural network architecture for classifying the coloured malarial cell images into the parasitized and uninfected classes of images based on the extracted features. This novel approach returned 97.72% classification accuracy with leading sensitivity, specificity, precision, recall, and area under curve metrics.