Optimal Machine Learning Based Automated Malaria Parasite Detection and Classification Model Using Blood Smear Images

Abstract

Malaria is a deadly disease which can be spread by the Plasmodium parasites. The existence of malaria can be identified by professional microscopists who examine the microscopic blood smear images. But it remains a challenge owing to the unavailability of experts, poor resolution images, and insufficient diagnostic quality. Therefore, image processing and machine learning (ML) models can be employed to detection of malaria parasites using blood smear images. With this motivation, this study introduces an optimal machine learning based automated malaria parasite detection and classification (OML-AMPDC) model using blood smear images. The proposed OML-AMPDC technique primarily undergoes preprocessing in two stages namely adaptive filtering (AF) based noise removal and contrast enhancement using CLAHE technique. Besides, the feature extraction process was implemented using Local Derivative Radial Patterns (LDRP). In addition, random forest (RF) classifier is applied to allot proper class labels to the blood smear images. Finally, particle swarm optimization (PSO) algorithm was utilized for optimally choose two parameters of the RF model, named maximum number of levels in every decision tree (max_depth) and number of trees in the forest (n_estimators). The design of PSO algorithm helps for enhancing the classification performance of the RF method. A wide-ranging experimental analysis is performed using benchmark dataset and the results reported the betterment of the OML-AMPDC technique over the recent approaches.

Utilizing Image Analysis with Machine Learning and Deep Learning to Identify Malaria Parasites in Conventional Microscopic Blood Smear Images

“…It may be described as the first artificial intelligence-based procedure that, faster and more accurately than manual testing, can indicate whether a blood cell is highly likely to contain a parasite. Kundu et al [150] inaugurated the OML-AMPDC methodology, integrating a bifold pre-processing strategy utilizing adaptive filtering and CLAHE for the dual purposes of noise eradication and contrast augmentation. The feature extraction entails the utilization of Local Derivative Radial Patterns (LDRP), while a random forest classifier is enlisted for the categorization of blood smear images.…”

“…Employing a diverse array of methodologies, encompassing knowledge distillation, data augmentation, Autoencoder, CNN-based feature extraction, and subsequent classification using a Support Vector Machine (SVM) or K-Nearest Neighbors (KNN), the model undergoes meticulous training through three distinct procedures-termed general training, distillation training, and autoencoder training-to refine and elevate both accuracy and inference performance. Preprocessing applied filtering [64,116], Median filtering [9,63,92,93,111,127], SUSAN [33,78], Gaussian low pass filter [138], Morphological operation [96,100,147], Laplacian [93,96,145], Local histogram equalization [95,96,98,127], Forward discrete curve [98], Low pass filter [99], linear model [109], Gray world colour normalization [62] K-means [67], Quaternion Fourier Transform (QFT) [44] Thresholding [84,92,109,129] Bayesian classifier [40,100], Euclidean distance classifier [114], K-nearest Neighbors [99,102,147], K Mean [61], SVM [153], Decision tree [150], Template matching [130], Genetic algorithm…”