Malaria remains a major global health challenge. Current diagnosis relies on microscopic analysis of blood smears, a process that is time-consuming, expertise-dependent, and limited in resource-constrained settings. Deep learning approaches show promise for automated diagnosis but achieving high species-level accuracy from complex microscopic images remains challenging. In this study, we apply Mask R-CNN, an advanced instance segmentation architecture, to automate the detection and species-level classification of four Plasmodium species (P. falciparum, P. vivax, P. malariae, and P. ovale). The model was trained and evaluated on 971 annotated microscopic images collected from healthcare facilities in Rwanda. Our approach achieved a high combined mean average precision (mAP = 0.892), with a strong performance for P. vivax (mAP = 0.958) and P. malariae (mAP = 0.946). These results demonstrate the potential of Mask R-CNN to provide accurate, efficient, and scalable support for malaria diagnosis.