Automated detection of plasmodium species using Machine-Learning technique

Abstract

Background: Malaria is one of the widespread mosquito-borne diseases. The WHO report of year 2018 confirms a recorded disease rate of 219 million and a death rate of 435,000. Accurate diagnosis of malaria infection is essential for appropriate management of cases. WHO's Global Technical Strategy for Malaria 2016–2030 recommends that surveillance strategies should be integrated as core interventions to accelerate progress towards elimination. The proposed work aims to develop a Machine-Learning (ML) system for the assessment and classification of the parasite existing in the stained thin blood smear images recorded with a digital microscope. This work implements a multi-class classifier to categorize the plasmodium according to its shape and texture features. Methods and materials: In this work, an automated system is proposed to extract, evaluate and classify the plasmodium species available in the considered test images. The proposed system initially implements a multi-level thresholding to enhance the image and plasmodium in enhanced image is extracted using morphological segmentation, later the texture/shape features are extracted and finally a multi class classifier is implemented to categorize the test images based on the parasite and its growth phase. In order to attain better classification accuracy, this work implements prevailing feature selection and multi-class classification based on Support-Vector-Machine (SVM). Results: The developed system was tested on 2000 numbers of RGB scale thin blood smear pictures of size 512 x 512 pixels and the results obtained in the proposed technique was evaluated to validate the performance of proposed ML system. The results of this study provides an overall classification accuracy of 97.05% with the SVM and relative assessment with the comparable procedures in the literature also verifies that, ML technique offers superior values of accuracy, precision, sensitivity, specificity and F1score. Conclusion: This work implements a Machine-Learning tool to asses and classifies the plasmodium species using the stained thin blood smear images. The outcome of the proposed tool confirms that, it works well on considered microscopic image database and helps to attain better classification accuracy. In future, the proposed approach can be used to examine the clinical grade thin blood smear images of malaria infected patients.