Enumeration and Classification of Freshwater Algal Samples Using Semi-automated Imaging Flow Cytometry and Supervised Machine Learning Techniques

Abstract

The manual enumeration and identification of algal samples is a time-intensive process requiring considerable expertise in algal taxonomy. Semi-automated enumeration using imaging flow cytometry, coupled with an accurate classification model, can process samples and deliver data on a timescale representing a fraction of the processing time of traditional, manual counts. In this talk, we will discuss sample processing using an Imaging Flow Cytobot (IFCB) and subsequent classification utilizing two supervised machine learning methods. We evaluated the ability of random forest and convolutional neural networks to classify over 250 freshwater phytoplankton samples. These samples represent 137 classes separated by taxonomy and functional group and include 200,000+ images based on live material. Various methods were evaluated to determine algal targets, augment images, and extract relevant features. These methods were implemented in both the MatLab and Python programming environments. We will discuss the accuracy of both supervised learning approaches, considerations that should be made, and how semi-automated algal classification can be utilized to better understand and evaluate algal assemblages.

Authors

Cory Sauve, Denise Clark, Hannah Schroeder, and Ann St. Amand