Paludism Detection using Convolutional Neural Networks

This study introduces a convolutional neural network (CNN) approach for diagnosing malaria (paludism) using a hybrid dataset, integrating 27,558 images from public sources with 442 locally-collected samples from southern Algeria to ensure the model's relevance for regional healthcare applications. We compared our CNN architecture against popular transformer-based models, VGG19 and VGG16, as well as an earlier CNN architecture. While VGG19 and VGG16 are renowned for their performance in medical imaging, our older CNN architecture outperformed both, achieving an accuracy of 99.85%, specificity of 99.92%, and sensitivity of 99%. This model's robust performance across diverse blood smear images, including regional variations, positions it as an effective diagnos-tic tool tailored to local needs, supporting healthcare providers in accurately identifying malaria infections.