Automating Malaria Diagnosis from Blood Smear Images Using Deep Transfer Learning

Abstract

Malaria is a disease that is found all throughout the world, primarily in tropical places. When a human is bitten by an infected female Anopheles mosquito, parasites enter the bloodstream and kill red blood cells (RBCs), which transport oxygen. The flu was the first symptom of malaria. After a few days/weeks, the symptoms generally arise. A lethal parasite may live in a person's body for nearly a year without generating any symptoms. As a result, postponing treatment might lead to complications and, in the worst-case scenario, death. As a result, early detection of malaria can save a lot of lives. In practise, radiologists look at a blood smear (thin/thick) to identify the disease and calculate parasitemia to diagnose malaria. One of the most effective methods for diagnosing malaria is microscopy. The microscope was often used since it was affordable yet time consuming. Quality blood smears and the presence of a trained expert in the differentiation between parasitized and non-parasitized blood cells are crucial for an accurate diagnosis. To automate the detection of malaria parasites in blood smear microscopy pictures, we employed Transfer learning, a deep learning approach, on pre-trained VGG16, VGG19, and ResNet50 models.