An Optimized Convolutional Neural Network (CNN) For Malaria Parasite Detection in Blood Smear Images

Abstract

Malaria remains a leading cause of morbidity and mortality worldwide, and accurate
detection of the malaria parasite in blood smear images is essential for timely diagnosis and treatment,
especially in resource-limited settings. This research develops an optimized Convolutional Neural Network
(CNN) model for malaria parasite detection, employing Bayesian optimization to fine-tune
hyperparameters like learning rate, batch size, dropout rate, and filter sizes for improved performance and
computational efficiency. The model is evaluated against pre-trained models (VGG19, ResNet50, and
EfficientNet) using metrics such as accuracy, precision, recall, F1 score, specificity, inference time, and
memory usage. Results show that the optimized CNN achieved 99% accuracy, with precision and recall of
98.1% and 98.2%, respectively, and reduced inference time to 0.001 seconds per image, outperforming
VGG19 and ResNet50 and achieving comparable performance to EfficientNet with lower computational
demands. These findings highlight the potential of optimized CNNs for malaria detection in low-resource
environments, offering a promising solution for timely and cost-effective diagnosis. Future work may focus
on further optimizations and expanding the model’s applicability to other diseases.