Research On Self-detection And Classification Algorithm Of Malaria Based On Deep Learning Methods

Malaria is a life-threatening infectious disease caused by malaria parasites.According to the data released by the World Health Organization in 2017,there are 219 million cases of malaria worldwide covering more than 90 countries,among which the number of deaths caused by malaria was more than 40 thousand.A precise diagnosis of malaria is the premise of malaria treatment,prevention and management.Optical Microscope Screening diagnosis is considered to be the golden standard method of malaria diagnosis.However,it's a time-consuming approach and whose precision totally depends on the professional level of technicians.The lack of expert technicians will be the biggest obstacle of malaria diagnosis in some low-source areas currently.Focus on the issue mentioned above,in this paper,we study on some automated methods based on deep learning for rapid and precise classification and detection of malaria parasites.(1)A novel deep neural network(VGG+SVM)is introduced for identifying infected malaria parasite using transfer learning method.This proposed model can be achieved by combining existing Visual Geometry Group(VGG)Network and Support Vector Machine(SVM).In this model,the initial k layers of the pre-trained VGG are retained and(n-k)layers are replaced with SVM.The pre-train VGG network facilitates the role of expert learning model and SVM as domain specific learning model to classify.We tested the proposed model on the open,free dataset MAMIC provided by the Institute for Molecular Medicine Finland(FIMM),which contained 2550 images samples taken from 50 thin blood smears,including 1030 infected images and 1520 non-infected images.The experimental results show that this method achieves better results than other CNN-based models.(2)For small target in images taken from thick blood smear,we use the Faster-RCNN deep learning model to detect malaria parasites.The model is mainly composed of two parts.The first part is the deep neural network for extracting the feature region,and the second part is a Fast-RCNN model for detecting the targets using the features extracted by the first part.We evaluated our Faster-RCNN model using the publicly available free datasets from the Artificial Intelligence and Data Science Research Group of Makerere University in Uganda.One of the dataset has a total of 2703 images taken by the camera from thick blood films of 133 individuals,and the picture contained 50,255 malaria parasites.Experimental results show that the original image can be cut into small blocks and faster-RCNN can achieve an accuracy of 89.98%that is better than other methods using this datasets.This model can be easily extended to the detection of other parasites or blood pathogens,which is of great significance for the detection of global diseases,especially malaria parasites.