Automated classification of malaria parasite species in thin blood smears using advanced deep learning techniques

Abstract

Accurate identification of malaria parasite species from Giemsa-stained thin blood smears is essential for appropriate treatment, but remains difficult because of subtle morphological similarities between Plasmodium vivax and Plasmodium falciparum, staining variability, and imbalanced datasets. This study developed an artificial intelligence–based framework to improve automated parasite species classification. A total of 18,847 Giemsa-stained thin blood smear images from the Indian Council of Medical Research–National Institute of Malaria Research were analyzed, including uninfected erythrocytes and cells infected with P. vivax or P. falciparum . The proposed framework first learned informative image features from unlabeled microscopy images and then classified samples using a hierarchical strategy that sequentially identified infected cells and parasite species. Model performance was evaluated using three train–validation–test split strategies (60:20:20, 70:15:15, and 80:10:10) and compared with established deep learning models. The proposed framework achieved the highest performance across all evaluation strategies. Using the 70:15:15 split, it reached 98.20% accuracy, a 97.54% F1-score, a ROC–AUC of 99.66%, and a Matthews correlation coefficient of 95.42%. Performance remained robust under increasing class imbalance, achieving 80.61% accuracy and a 76.61% F1-score at 2% training prevalence. Model interpretation demonstrated that predictions were primarily based on parasite-associated regions within infected erythrocytes. The proposed framework enables accurate and robust automated classification of malaria parasite species from thin blood smear images. These findings support its potential application as a computer-assisted tool for malaria microscopy and clinical decision support.