Malaria Parasite Detection From Peripheral Blood Smear Images Using Deep Belief Networks

Abstract

In this paper, we propose a novel method to identify the presence of malaria parasites in human peripheral blood smear images using a deep belief network (DBN). This paper introduces a trained model based on a DBN to classify 4100 peripheral blood smear images into the parasite or non-parasite class. The proposed DBN is pre-trained by stacking restricted Boltzmann machines using the contrastive divergence method for pre-training. To train the DBN, we extract features from the images and initialize the visible variables of the DBN. A concatenated feature of color and texture is used as a feature vector in this paper. Finally, the DBN is discriminatively fine-tuned using a backpropagation algorithm that computes the probability of class labels. The optimum size of the DBN architecture used in this paper is 484-600-600-600-600-2, in which the visible layer has 484 nodes and the output layer has two nodes with four hidden layers containing 600 hidden nodes in every layer. The proposed method has performed significantly better than the other state-of-the-art methods with an F-score of 89.66%, a sensitivity of 97.60%, and specificity of 95.92%. This paper is the first application of a DBN for malaria parasite detection in human peripheral blood smear images.

Keywords

• Deep learning
• deep belief network
• malaria parasite detection
• restricted Boltzmann machine
• contrastive divergence
• discriminative training