Henning Husmo Kristiansen, Moa Metz, Lorena Silva-Garay, Fredrik Jutfelt, Robine H. J. Leeuwis 

Manually obtaining the length and other morphometric features of an animal can be time-
consuming, and consistent measurements are challenging with large datasets. By leveraging
high-throughput computing power and machine learning-based computer vision, such
phenotypic data can be rapidly collected with high accuracy. Here we present HusMorph, a
novel application with a simple and intuitive graphical user interface (GUI), based on the
same machine learning method used in other pipelines such as ML-morph. It consists of an
all-in-one package with the goal of making machine learning easy to use for non-experts. The
user starts by setting any number of landmarks on a set of photos captured with a
standardized setup. From this set, a machine learning model is generated by automatically
and randomly searching for the best performing parameters. Next, the user can apply the
model to predict landmarks on new standardized photos, and visually confirm and export the
results of the predictions. For measuring length between landmarks, an additional feature
allows for detecting a scale bar for each photo to convert the length from pixels to a metric
unit. Our application has been validated and applied to extract standard length from 1,935
photos of zebrafish and performs with about 99.5% accuracy compared to manual
measurements along with 100% scale bar detection.