High-throughput time-stretch imaging flow cytometry for multi-class classification of phytoplankton

Abstract

Time-stretch imaging has been regarded as an attractive technique for high-throughput imaging flow cytometry primarily owing to its real-time, continuous ultrafast operation. Nevertheless, two key challenges remain: (1) sufficiently high time-stretch image resolution and contrast is needed for visualizing sub-cellular complexity of single cells, and (2) the ability to unravel the heterogeneity and complexity of the highly diverse population of cells - a central problem of single-cell analysis in life sciences - is required. We here demonstrate an optofluidic time-stretch imaging flow cytometer that enables these two features, in the context of high-throughput multi-class (up to 14 classes) phytoplantkton screening and classification. Based on the comprehensive feature extraction and selection procedures, we show that the intracellular texture/morphology, which is revealed by high-resolution time-stretch imaging, plays a critical role of improving the accuracy of phytoplankton classification, as high as 94.7%, based on multi-class support vector machine (SVM). We also demonstrate that high-resolution time-stretch images, which allows exploitation of various feature domains, e.g. Fourier space, enables further sub-population identification - paving the way toward deeper learning and classification based on large-scale single-cell images. Not only applicable to biomedical diagnostic, this work is anticipated to find immediate applications in marine and biofuel research.

Hidden Secrets Behind Dots: Improved Phytoplankton Taxonomic Resolution Using High‐Throughput Imaging Flow Cytometry

“…Nevertheless, certain technical limitations still occur. Species identification could be very accurate (16), as was also demonstrated by several authors using images from different imaging cytometers (44,48,49). For taxa in the size range of 2-30 μm, this magnification is not sufficient for distinguishing species due to poor resolution (45).…”

“…IFC could probably overcome limitations of both methods: microscopy and FCM. In recent studies making use of machine learning and imaging cytometry some misclassifications between diatoms and cyanobacteria (15,49), diatoms and green algae (44,49), cryptophytes and green algae (16), and dinophytes and green algae (59) ( Table 4) could have been excluded by using flow cytometric differentiation of larger taxonomic groups as presented in this study. In contrast to previous studies, this study aims to make use of multispecies flow cytometric differentiation as meaningful supplemental for machine-learning approaches.…”

“…As Rodenacker et al (62) point out a critical aspect of this microscopic approach is the overlap of cells and detritus particles. Several new IFC approaches for phytoplankton are emerging (47,63,65) and are based on time-stretch imaging in microfluidic systems (49) or fluorescence microscopy (63,65). With respect to analysis time and particle overlap, the application of standard FCM overcomes all mentioned disadvantages of microscopy but it is limited due to low taxonomic resolution.…”

All-passive pixel super-resolution of time-stretch imaging

“…In practice, multiple morphological metrics can be measured from the pixel-SR image to improve classification accuracy734. Owing to the discrete structure of senedesmus colonies, it is possible to separate the two sub-types: two-daughter and four-daughter colonies more effectively in a higher-dimensional morphology metric space.…”

Automated Diatom Classification (Part A): Handcrafted Feature Approaches

“…That is, the methods rely on expert knowledge to extract the most relevant features versus CNN approaches that learn features from data. However, still, the handcrafted methods present limited results as in [25], where 14 classes were classified with Support Vector Machine (SVM) 10 fcv, using 44 GLCM features that describe only geometric and morphological properties. They obtained an accuracy of 94.7%.…”