Unimodal to Multimodal Machine Learning in Malaria Diagnosis: Challenges and Opportunities

Malaria continues to pose an enormous global health burden, particularly in Sub-Saharan Africa, where early and correct diagnosis is essential for effective treatment. Traditional unimodal machine learning (ML) models have been widely used for malaria detection, mostly based on microscopic blood smeared pictures. Nevertheless, these models are limited in their generalizability, interpretability, and reliance on just one data source. The shift to multimodal ML, which incorporates imaging, clinical, and genomic data, opened up new avenues for improved diagnostic accuracy and early-stage malaria diagnosis. This research investigated the obstacles and potential of transitioning from unimodal to multimodal machine learning for malaria diagnosis. We examined existing unimodal approaches, highlighted fusion strategies for multimodal learning, and addressed issues such as data integration difficulty, computing needs, and interpretability. Finally, the potential of multimodal ML for increased diagnostic reliability, early illness identification, and clinical decision-making was discussed. This study emphasized the necessity for more research in multimodal ML for malaria diagnosis.