A novel artificial intelligence–powered cell counting tool coupled with digital microscopy for rapid field‐assessment of harmful cyanobacterial blooms

ABSTRACT Historical quantification of cyanobacterial harmful algal blooms (cHABs) typically involved labor‐intensive manual cell counting. We developed a novel, cost‐effective, field‐validated system to perform cell counts of six common toxin‐producing cyanobacterial genera within 30 s of upload with 10‐min sample preparation. Using a portable field microscope, users can quickly evaluate the type and quantity of freshwater cyanobacteria for use in ecological monitoring, human health, and water quality. Participating groups ( n = 21) received digital microscopes and sampling equipment and submitted images from 170 cHAB events occurring in 36 US lakes for machine learning (ML) assisted cHAB analysis via a smartphone app. The accuracy of ML identification was compared to human taxonomic identification, while cell concentrations were compared against FluoroProbe (bbe Moldaenke GmbH) blue‐green algal (BGA) chlorophyll fluorescence. Machine learning classification of 497 photos containing 4002 colonies performed as well as, or better than, human taxonomic analysis in 94% of cases. There was a weak correlation between BGA chlorophyll and ML‐derived cell counts ( R 2 = 0.33), biovolume ( R 2 = 0.13) or total pixel counts ( R 2 = 0.32) across all samples, but there was a strong correlation ( R 2 = 0.76) between ML cell concentrations and BGA chlorophyll in samples not subject to overnight shipping and handling, where stress induced by transport and dark conditioning of cyanobacteria was hypothesized to drive more error in quantification by fluorescence. Cell counting minimizes errors introduced by fluorescence measurements and could improve risk assessment. The described quantification tool is easy‐to‐use and readily accessible to users with different levels of expertise in cHAB science.