LeDaphNet: An Automated Assessment of Daphnia magna Populations Using Digital Imaging and Machine Learning
Year: 2017
Type: report
Abstract: Linking population-level consequences of environmental stressors to initial events in the adverse outcome pathway has been an enduring challenge to environmental scientists.The quantification of populations has hindered progress in this arena.Daphnia magna is a water flea species often used to assess effects of natural changes or anthropromorphic influences in the environment.The small size and high fecundity of daphnids make them ideal for performing experiments at the organismal and population levels.Currently, the inefficiency in counting daphnids within a population hinders the rapid acquisition of longitudinal population-level data, thereby limiting the causal association between organismal responses and ecosystem adversity.We propose a novel method to semi-automate population-level data collection of daphnids using a document scanner and a 3D-printed imaging vessel to produce images of samples from daphnid populations.Using machine learning, specifically deep neural networks, we developed open-source classification software that counts the daphnids present in each image with over 92% accuracy.We validate our results with a 4-week population study examining the effect of the herbicide pendimethalin on daphnid populations.We found our classifier-based method to be nearly as accurate as the traditional method of hand counting, while also being substantially more efficient and easily scalable.
Language: English
Subfield: Biophysics
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W3194340998