VL-M2C: Leveraging deep learning approach for stage detection of malaria parasites

Abstract

Malaria is a parasitic infection that can be caused by the bite of infected anopheles’ mosquito and can progress from mild symptoms to severe forms which make it crucial to understand its potential consequences. This study majorly focusses on multiclass classification and provides an ensemble framework for the detection of stages of malaria parasite in thin blood smears. In this study we used publicly accessible dataset comprising 1320 images together with training and test json file. Initially pre-processing is applied to improve image quality, then key regions are extracted to retain important information during feature extraction phase. During this study we compared different classification techniques to find the best model for multiclass classification for malaria parasite stages. Several metrics, including accuracy, recall, precision, and loss, are used to analyze the performance of the model. In this study, the ensemble method VL-M2C ie VGG LSTM Multiclass Malaria Classification has been proposed that raises the overall accuracy and robustness of the model by considering the advantages of individual classifiers. It has been compared with VGG16, CNN and RCNN. Our proposed VL-M2C has the best accuracy (98.56%) and lowest loss (0.1240), thus proves promising diagnosis system.