Malaria Detection Using Deep Convolution Neural Network

According to the WHO World Malaria Report 2024, malaria cases rose to 263 million, with 597,000 deaths—most of them children under five in sub-Saharan Africa. Despite advances, progress in malaria control has slowed due to limited resources and dependence on manual microscopy for diagnosis. Artificial intelligence offers a promising path for rapid, low-cost screening. In this work, we present a lightweight convolutional neural network (CNN) optimized via Bayesian hyperparameter tuning for automated malaria detection from blood smear images. The proposed model achieves high accuracy in classifying parasitized versus uninfected cells. To support deployment in real-world settings, we developed a middleware framework for mobile inference. In addition, we integrate an Out-of-Distribution (OOD) detector based on RRRCF, which flags anomalous inputs responsible for rare CNN misclassifications. Together, the CNN and OOD components yield effective prediction accuracy, demonstrating that compact, mobile-compatible deep learning systems can significantly improve malaria diagnostics in resource-limited environments.