ENHANCING MALARIA PARASITE CLASSIFICATION USING THIN BLOOD SMEAR IMAGES THROUGH DEEP LEARNING OPTIMIZER SELECTION

Abstract

Malaria is still one of the most widespread vectors borne diseases in the world. Generally, the diagnosis is done manually by viewing thin blood smear images on a microscope, which is time-consuming, labor-intensive and human error prone. However, with recent developments in deep learning, automatic diagnosis systems can be developed to improve the accuracy and speed of diagnosis. In this research, an automatic Malaria Cell Detection Framework using EfficientNetB4 Convolutional Neural Network (CNN) Architecture is presented which is compared with the different optimization methods to enhance the classification accuracy. Pre-processing images, data augmentation, transfer learning and fine-tuning the EfficientNetB3 model are suggested to be used. A variety of optimization algorithms such as Adam, SGD with momentum and RMSprop are tested using important metrics like accuracy, precision, and sensitivity. Recall, F1 score and Area Under Receiver Operating Characteristic Curve (AUC-ROC). Experimental results show that the model convergence and stability as well as generalization performance can vary greatly depending on the type of the optimizer used. The Adamax optimizer outperforms the other optimization methods in terms of classification accuracy with 97.57%. The results reveal the efficacy of EfficientNetB4 model as a light-weight yet efficient network for malaria parasite detection in thin blood smear images and informative guidelines for the selection of suitable optimization techniques for medical image classification tasks. The findings of this study can be used for the development of robust computer aided diagnostic systems which provide early detection of malaria in a health care setting, especially in resource-limited environments.