Malaria Diagnosis with Dilated Convolutional Neural Network Based Image Analysis
Alif Bin Abdul Qayyum, Tanveerul Islam, Md. Aynal Haque · 2019
Malaria is a worldwide health issue. Traditionally, microscopic visual examination of blood smears to detect parasite infected red blood cells has been the standard procedure for diagnosing malaria. This manual method is prone to human errors and time consuming too. To automate the diagnosis, machine learning based image recognition has been applied in the past. But the performance was below par specially in case of big datasets. This paper proposes a kernel dilation based new and robust convolutional neural network (CNN) to automatically classify infected and uninfected red blood cells. Three different dilation approaches were used among which Fibonacci series-wise dilated CNN model performed best in all metrics such as accuracy (96.05%), precision (95.80%), recall (96.33%) and F1 score (96.06%) while working with a dataset of 27,558 cell images.