Deep Learning–Based Automated Diagnosis of Malaria Using Blood Smear Microscopy Images
Résumé
Malaria remains a major global health burden, particularly in sub-Saharan Africa and Southeast Asia. Conventional diagnosis using light microscopy is considered the gold standard but is highly dependent on skilled personnel and prone to inter-observer variability. Timely and rapid diagnosis is crucial for faster and proper malaria treatment planning. Microscopic examination is the gold standard for malaria diagnosis, where hundreds of millions of blood films are examined annually. This study presents a deep learning–based automated system for malaria diagnosis using Giemsa-stained thin blood smear microscopy images. A dataset of 12,000 labeled images was used to train and validate a convolutional neural network (CNN) model based on transfer learning with ResNet-50. Data preprocessing included normalization, augmentation, and artifact removal. The model was evaluated using accuracy, sensitivity, specificity, F1-score, and ROC-AUC. The proposed system achieved an accuracy of 97.8%, sensitivity of 98.4%, specificity of 97.1%, and AUC of 0.98. Statistical comparison using McNemar’s test demonstrated significant improvement over traditional machine learning classifiers (p < 0.05). The findings indicate that deep learning can provide reliable, scalable, and cost-effective malaria diagnosis support in low-resource settings. Automated malaria diagnosis has significant potential for low-resource and high-burden settings where skilled microscopists are limited. With further multicenter validation and optimization for field deployment, deep learning systems can serve as scalable decision-support tools, contributing to improved surveillance, case management, and malaria elimination efforts.