DMLMD I dataset for machine learning based malaria detection using blood smears

AbstractBackground and Objective:Malaria remains a life-threatening disease requiring rapid and accurate diagnosis to reduce morbidity and mortality. Conventional microscopy is labor-intensive and prone to human error, particularly in resource-limited settings. This study aimed to develop and augment a thin blood smear image dataset for malaria detection using deep learning approaches to enhance diagnostic accuracy and support automated screening.Methods:An initial dataset of 631 high-resolution thin blood smear images representing five developmental stages of Plasmodium infection Multi Ring (M-Rg), Ring (Rg), Multi Trophozoite (M-Trp), Trophozoite (Trp), and Schizont (Sz) was expanded using image augmentation techniques. The augmented dataset, comprising 9,314 images, was employed to train and evaluate deep learning models for differentiating infected and uninfected red blood cells.Results:The augmented dataset significantly improved model training robustness and enabled effective automatic detection of malaria parasites. Deep learning models successfully identified subtle morphological patterns associated with different infection stages, reducing potential diagnostic errors and enhancing classification performance.Conclusion:The augmented thin blood smear dataset provides a valuable resource for developing AI-based malaria diagnostic systems. Such automated tools offer scalable, accurate, and accessible solutions for malaria detection, particularly in under-resourced regions, thereby supporting improved disease management and control.