Abstract
Malaria remains a public health burden in sub-Saharan Africa and other resource-limited settings, where early and accurate diagnosis can be impeded by the scarcity of adequately trained personnel and diagnostic tools availability. Traditional diagnostic methods, including examining stained blood smears with a microscope, are time-consuming and prone to human error. This study addresses these challenges by exploring an automated deep-learning approach using the ConvNeXt architecture based on transfer learning and data augmentation to detect malaria parasites in thin blood smear images. This study first addresses the primary research question: how well does the ConvNeXt model increase malaria diagnostic accuracy while maintaining adaptability in varying image capture conditions? To ensure that the ConvNeXt models are capable of classifying malaria cases, they were fine-tuned on a balanced dataset of parasitised and uninfected samples.
Results show that the ConvNeXt V2 Tiny re-modified model achieved the highest classification accuracy of 98% over both other models like Swin Tiny, ResNet18 and ResNet50. The findings suggest that the model could serve as a cost-effective and efficient diagnostic tool in resource-limited settings. This work adds to existing published research that contributes to advancing artificial intelligence (AI)-driven malaria diagnostics by presenting a reliable model that is applicable to clinical use in broadly distributed and resource-limited environments. By integrating explainable AI techniques in the study, the model's decision-making process becomes transparent so that healthcare providers can have greater clinical trust and make more informed diagnostic decisions. The importance of adaptable high-performance models to support malaria control in resource-limited settings is emphasized in this research.