Toxicology studies in early fish life stages serve an important function in measuring the impact of potentially harmful substances, such as crude oil, on marine life. Morphometric analysis of larvae can reveal the effects of such substances in retarding growth and development. These studies are labor intensive and time consuming, typically resulting in only a small number of samples being considered. An automated system for imaging and measurement of experimental animals, using flow-through imaging and an artificial neural network to allow faster sampling of more individuals, has been described previously and used in toxicity experiments. This study compares the performance of the automated imaging and analysis system with traditional microscopy techniques in measuring biologically relevant endpoints using two oil treatments as positive controls. We demonstrate that while the automated system typically underestimates morphometric measurements relative to analysis of manual microscopy images, it shows similar statistical results to the manual method when comparing treatments across most endpoints. It allows for many more individual specimens to be sampled in a shorter time period, reducing labor requirements and improving statistical power in such studies, and is noninvasive allowing for repeated sampling of the same population.
Keywords: Atlantic cod; automated imaging; computer vision; machine learning; morphometrics.