In situ digital holographic microscopy for rapid detection and monitoring of the harmful dinoflagellate, Karenia brevis

• Published in: Harmful algae Vol. 123; p. 102401
• Main Authors: Barua, Ranjoy, Sanborn, Delaney, Nyman, Lisa, McFarland, Malcolm, Moore, Timothy, Hong, Jiarong, Garrett, Matt, Nayak, Aditya R.
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 01.03.2023
• Subjects: Animals; Convolutional neural network; Dinoflagellida; Forecasting; Gulf of Mexico; Harmful Algal Bloom; Holography; Humans; In situ imaging; Karenia brevis; Microscopy; Plankton monitoring; Red tide; Gulf of Mexico; Red tide; Convolutional neural network; Holography; In situ imaging; Gulf of Mexico; Harmful algal bloom; Karenia brevis; Plankton monitoring
• ISSN: 1568-9883; 1878-1470
• DOI: 10.1016/j.hal.2023.102401

•An in situ holographic microscope developed to monitor Karenia brevis.•K. brevis automatically classified from images using a convolutional neural network.•Results showed good agreement with manual counts and flow cytometry.•Towing system paired with sensor allows for spatial mapping of plankton/blooms.•Promising method for adaptation in long term plankton/bloom monitoring networks. Karenia brevis blooms, also known as red tide, are a recurring problem in the coastal Gulf of Mexico. These blooms have the capacity to inflict substantial damage to human and animal health as well as local economies. Thus, monitoring and detection of K. brevis blooms at all life stages and cell concentrations is essential for ensuring public safety. Current K. brevis monitoring methods have several limitations, including size resolution limits and concentration ranges, limited capacity for spatial and temporal profiling, and/or small sample volume processing. Here, a novel monitoring method wherein an autonomous digital holographic imaging microscope (AUTOHOLO), that overcomes these limitations and can characterize K. brevis concentrations in situ, is presented. Using the AUTOHOLO, in situ field measurements were conducted in the coastal Gulf of Mexico during an active K. brevis bloom over the 2020–21 winter season. Surface and sub-surface water samples collected during these field studies were also analyzed in the lab using benchtop holographic imaging and flow cytometry for validation. A convolutional neural network was trained for automated classification of K. brevis at all concentration ranges. The network was validated with manual counts and flow cytometry, yielding a 90% accuracy across diverse datasets with varying K. brevis concentrations. The usefulness of pairing the AUTOHOLO with a towing system was also demonstrated for characterizing particle abundance over large spatial distances, which could potentially facilitate characterization of spatial distributions of K. brevis during bloom events. Future applications of the AUTOHOLO can include integration into existing HAB monitoring networks to enhance detection capabilities for K. brevis in aquatic environments around the world.