721 / 2024-09-19 10:17:00
Progress in Developing FluoSieve Imaging Flow Cytometry for Marine Phytoplankton Observation
phytoplankton,AI,in situ,HAB
Session 12 - Alleviating the impact of emerging Harmful Algal Blooms (HABs) to coastal ecosystems and seafood safety for a sustainable and healthy Ocean
Abstract Accepted
The development of automated technologies for in-situ analyzing photosynthetically active phytoplankton cells and colonies in natural seawater is of great significance for biological oceanography and HAB monitoring. However, the composition of natural seawater is highly complex. The size range of phytoplankton spans at least 3 orders of magnitude, from single cells <1μm to large diatoms or colonies >500μm. In addition, seawater also contains countless non-phytoplankton particles. These facts present enormous challenges in specificity, sensitivity, and spatial resolution for existent imaging flow cytometers (IFC) such as CytoSense and IFCB to observe phytoplankton in situ.
Ocean observation and HAB monitoring prefer high-throughput methods in analyzing more seawater within less time to extract more realistic phytoplankton information. Since most phytoplankton are tiny, IFCs usually adopt slow flow with high magnification to obtain sufficient resolution for imaging phytoplankton. However, to enhance imaging throughput, IFCs should use higher flow rates with lower magnifications, though may be very likely at the cost of imaging resolution and quality sacrifice, to gain increased seawater sampling capability. The compromise between imaging resolution and observation accuracy of current IFCs essentially limits their ultimate throughput.
We are trying to unite "low-magnification imaging" plus "computational image restoration" in a fluorescence imaging flow cytometer system named FluoSieve, which was previously reported in XMAS 2019, to balance this trade-off. By building up a large-scale phytoplankton fluorescence image dataset, we are training an image restoration CNN network called IfPhytoRS. The preliminary results indicate that the IfPhytoRS model can restore the poorer resolution and quality images acquired by lower magnification lens into much better counterparts as if were acquired by a much higher magnification lens. This would be very beneficial for downstream tasks such as phytoplankton taxonomic recognition or size measurement while simultaneously achieving much higher observation throughput. In this presentation, we will report on the progress of this research.
Ocean observation and HAB monitoring prefer high-throughput methods in analyzing more seawater within less time to extract more realistic phytoplankton information. Since most phytoplankton are tiny, IFCs usually adopt slow flow with high magnification to obtain sufficient resolution for imaging phytoplankton. However, to enhance imaging throughput, IFCs should use higher flow rates with lower magnifications, though may be very likely at the cost of imaging resolution and quality sacrifice, to gain increased seawater sampling capability. The compromise between imaging resolution and observation accuracy of current IFCs essentially limits their ultimate throughput.
We are trying to unite "low-magnification imaging" plus "computational image restoration" in a fluorescence imaging flow cytometer system named FluoSieve, which was previously reported in XMAS 2019, to balance this trade-off. By building up a large-scale phytoplankton fluorescence image dataset, we are training an image restoration CNN network called IfPhytoRS. The preliminary results indicate that the IfPhytoRS model can restore the poorer resolution and quality images acquired by lower magnification lens into much better counterparts as if were acquired by a much higher magnification lens. This would be very beneficial for downstream tasks such as phytoplankton taxonomic recognition or size measurement while simultaneously achieving much higher observation throughput. In this presentation, we will report on the progress of this research.