Optimized Malaria Identification through Transfer Learning Appoach
Al Jubayer Pial, Md. Ashiq Ul Islam Sajid, Abdullah Mubin, Muhammad Zawad Mahmud, M. F. Mridha, Md. Abdullah-Al-Jubair · 2024
Plasmodium parasites invade red blood cells and replicate inside; they cause malaria, a deadly disease transmitted by female Anopheles mosquitoes.Untreated, this has dire health implications.The classic way to diagnose malaria is to collect blood samples and examine them under a microscope for parasite and red blood cell counts, a lengthy and often unreliable process.Recently, new machine learning and deep learning approaches emerged as promising alternatives to standard microscopy for enhancing the diagnosis of pathological imaging at lower costs and with improved discrimination accuracy.Deep learning is a three-part process, consisting of training, feature extraction, and testing & validation.The data group itself is diverse with different kinds of images so that these image systems learn and predict the malaria infection accurately.The aim is to create a reliable, rapid, and inexpensive malaria diagnostic method, which is necessary for its treatment and prevention.We compared results from different studies for data on the classification of malaria vs. healthy blood samples.We used Convolutional Neural Networks (CNNs), specifically pre-trained deep learning models, to extract features and identified the best layers for this purpose.The images captured were of various facets of malaria, and key features of those images, such as size, color, shape, and cell size, helped improve the accuracy of classification.Statistical analysis was performed for the feature extraction outputs produced by pre-trained CNNs.Improvements in microscopy, when combined with automation, enable detectives to make more arrests in the case of malaria identification.Our transfer learning model VGG16 achieved a testing accuracy of 97% and a training accuracy of 96%, with a training loss of 10% and a testing loss of 9% indicating its reliability and efficiency in identifying malaria.InceptionV3 achieved a testing accuracy of 95%.