Classification of Parasitized Cells for Malaria Detection with Help of Deep Learning

Millions of people worldwide suffer from malaria, a potentially fatal disease. Early and precise diagnosis is essential for the medical condition to be successfully treated and managed. This thesis employs three computer-aided methods to determine percentages of red blood cells that are either parasitic or uninfected given test set(s) obtained from the National Institutes of Health (NIH) dataset. The three methods employed are traditional image processing, Support Vector Machine (SVM), and Convolutional Neural Networks based Deep Learning (CNN-DL). The simulations are performed using a dataset that has 27,558 images of red blood cells. The traditional image processing method achieves an accuracy of 91.97%. SVM classifier using Histogram of Oriented Gradients (HOG) features has an accuracy of 88.6% and with features extracted using Local Binary Patterns (LBP) accuracy has improved to 92.5% using a smaller subset of 6,040 images. The two previous methods proved to be inferior when compared with the CNN- DL classification which gave an accuracy of 95.7% using AlexNet, and 96.32% using GoogLeNet. The accuracy of each of the three computer-aided methods was based on performance metrics calculated using confusion matrices and the Receiver Operating Characteristic (ROC) curves.