V2SeqNet: A robust deep learning framework for malaria parasite stage classification in thin smear microscopic images

Abstract

Malaria is a global health problem, and it requires good diagnostic tools for early treatment. This research aims to develop an advanced deep learning-based model for the classification of malaria parasite stages using the MP-IDB dataset. Advanced preprocessing techniques such as resizing, normalization and data augmentation are used in research to enhance the quality of images and the variability of the dataset. A new architecture V2SeqNet ie VGG with 2 LSTM Layers for Sequential Classification Network combining VGG16 with LSTM layers and geometric feature extraction shows a good performance with accuracy of 99.69%, recall of 100% and precision at 98.78%. Results are compared to other state-of-the-art detection and classification models like MobileNet, EfficientNet, ResNet-50, DenseNet-121etc. Robustness, precision and scalability of the proposed model will give it an opportunity to become the best choice for automated malaria diagnosis. Future extension of the model is to identify more than one malaria species and stages, thus increasing its utility for comprehensive malaria diagnosis.