Machine Learning‐Assisted Infectious Disease Detection in Low‐Income Areas: Toward Rapid Triage of Dengue and Zika Virus Using Open‐Source Hardware

Abstract

Neglected tropical diseases represent a significant global health problem disproportionately affecting impoverished regions with resource‐limited sources. These diseases include mainly dengue, zika, and similar arbovirus infections. Rapid, accurate, and low‐cost diagnoses are crucial for effective disease management and control, improving live quality. However, conventional diagnostic methods often face logistical challenges, high costs, and complex infrastructure requirements. As an alternative, point‐of‐care technologies have emerged for diagnostics in remote and vulnerable areas. Even then, similar biomarkers are difficult to distinguish, compromising the diagnosis. This study shows an affordable platform for dengue and zika detection based on fluorine‐doped tin oxide thin films modified with gold nanoparticles and functionalized with DNA aptamers. Nonstructural protein 1 (NS1) proteins from dengue and zika were used as biomarkers. Electrochemical impedance spectroscopy was used to verify aptasensor performance and a custom machine learning model was implemented to distinguish both NS1 protein samples when tested individually and mixed in different proportions. Further the use of low‐cost and open‐source hardware for the implementation of this model for simultaneous detection, and the limitation of the associated hardware, is discussed. The results show an overall accuracy of 95.3% and a maximum RAM usage of 9.4 kB, enabling its use in low‐cost hardware‐limited devices (<25 USD).