You have no items in your shopping cart.
ABSTRACT
Malaria is one of the deadliest diseases in the world and is contracted by humans through the bite of a female Anopheles mosquito carrying the Plasmodium parasite. The traditional method of diagnosing malaria involves a graphic examination of human blood smears under a microscope by laboratory or trained workers for parasite-infected red blood cells. This project aims to make a system that can recognize and precisely analyse malaria parasites in digital blood sample images and indicate if the blood sample is infected nor not. In this project, the Convolutional Neural Network (CNN), a deep learning algorithm, and Visual Geometry Group-19 (VGG-19), a type of CNN, were used for malaria detection. Images of red blood cells infected and uninfected with malaria were acquired, then split into training, validation, and testing, and pre-processed using data augmentation. Using the training and validation dataset, the CNN model, which I built from scratch, and the VGG-19 model, a 19-layer deep CNN model that had been pre-trained and which we fine-tuned by freezing all three convolution blocks of its model leaving the last two unfrozen so as to update their weights after each epoch and inserting our own fully connected layer, were trained and validated. Then a comparison of the models was done by computing how they performed on the test dataset using these performance metrics: accuracy, precision, recall, and f1 score.
