Enhanced Malaria Detection Model using Deep Convolutional Neural Network with Comprehensive Data Augmentation and Grad-CAM Explainability for Clinical Trustworthiness

Abstract

Malaria remains a major global health challenge, particularly in sub-Saharan Africa and parts of Asia, where accurate and timely diagnosis is essential for effective treatment and control. Traditional microscopic examination, while widely used, is labor-intensive, subjective, and prone to misdiagnosis. To address these limitations, this study proposes deep learning-based approaches for automated malaria parasite detection from thin blood smear images. An enhanced malaria detection model using deep convolutional neural network with comprehensive data augmentation and Grad-CAM was developed. Using the NIH Malaria Dataset comprising 27,514 validated images, the models were trained and tested with rigorous preprocessing, augmentation, and stratified sampling. Results show that the CNN model achieved 96.37% accuracy, 98.40% recall for parasitized cells, and an AUC of 0.9935, outperforming conventional methods and providing robust generalization for unseen data. This study highlights the potential of deep learning in advancing malaria diagnostics while also addressing critical deployment considerations, including error calibration and clinical applicability. This enhances clinical Trustworthiness.