Reducing costs of Giardia spp. cyst enumeration using machine learning-based systems

Abstract Giardia spp. cyst enumeration is a laboratory procedure that is frequently required in water treatment-related research. Currently, detection conducted by experts using fluorescence microscopy on samples stained with specific markers for Giardia spp. cysts is still the standard method, despite its high costs limiting its usage worldwide and, ultimately, hindering waterborne analyses in low-income countries. We present an approach based on darkfield imaging and machine learning to reduce costs associated with Giardia spp. cyst enumeration and the lack of experts. Automated counts were compared to manual counts, achieving an average sensitivity (SE) rate of 88%, specificity (SP) of 100% and accuracy of 88% across a wide range of cyst concentrations. By using machine learning in conjunction with darkfield microscopy, a low-cost illumination technique that can be easily integrated into standard laboratory microscopes, we have significantly reduced the costs associated with Giardia spp. cyst detection, all while still maintaining the SE and SP of fluorescence microscopy. Based on the findings, the proposed system has the potential to be a useful tool to enumerate Giardia spp. cyst suspensions. It can be accessed by virtually any microbiology laboratory as it is consumable-free and expert-independent.