International Journal of Science, Engineering and TechnologyAuthors: Paramjeet Singh, Dr Raj Kumar
Abstract: Malaria remains one of the most devastating infectious diseases worldwide, claiming over 600,000 lives annually, with the majority of fatalities occurring in sub-Saharan Africa and South Asia. Traditional microscopic examination of blood smears, while being the gold standard for malaria diagnosis is time-consuming, expert-dependent, and prone to human error—particularly in resource-limited settings. This paper proposes MDCE-Net (Multi-scale Deep Convolutional Ensemble Network), a novel deep learning architecture that combines multi-scale feature extraction, Squeeze-and-Excitation attention mechanisms, and ensemble transfer learning using EfficientNetB4 and ResNet50 backbones for automated malaria parasite detection from thin blood smear images. The proposed model was trained and evaluated on a combined dataset of 56,480 cell images sourced from the NIH Malaria Dataset and the Kaggle Malaria Cell Images Dataset, encompassing both parasitized and uninfected cells. Extensive data augmentation strategies including random rotation, horizontal and vertical flipping, zoom, brightness adjustment, and Gaussian noise injection were employed to enhance model robustness. The MDCE-Net architecture achieved a classification accuracy of 99.21%, precision of 99.08%, recall of 99.34%, and F1-score of 99.21% on the test set, outperforming existing state-of-the-art methods including standalone VGG16, ResNet50, and EfficientNetB4 architectures. The model also demonstrates strong generalization performance on independent validation cohorts. This work presents a significant step toward automated, scalable, and deployable malaria diagnostic tools suitable for integration into point-of-care systems in rural and resource-constrained healthcare environments. Source code and model weights are made publicly available to facilitate reproducibility.
