FishInspector

The FishInspector software allows the user-friendly and easy annotation of features in zebrafish embryo 2-dimensional images. Features are detected automatically but may require correction by the user (implemented in the software). At present only lateral images can be analysed, but the software could be extended to analyse dorsoventral images as well. The software has been initially developed for images obtained with the an automated capillary position system. However, images from other sources can be used as well after appropriate conversion (see below). During the installation process an appropriate version of MATLAB runtime wll be installed as well. However, no MATLAB installation is required. FishInspector has been tested on Windows 7 but may run on other windows versions as well.

Recently (version 1.7, see also paragraph Future Developments) the software has been improved by using deep learning for the detection of structural features. The use of trained models provide high flexibility for the image source and allows to consider also other orientations.

More details on the software and the subsequent data analysis can be found in the following publications:

Teixido, E., Kießling, T. R., Krupp, E., Quevedo, C., Muriana, A. & Scholz, S. 2019. Automated morphological feature assessment for zebrafish embryo developmental toxicity screens. Toxicological Sciences, 167(438–449.

Teixidó, E., Kieβling, T. R., Klüver, N. & Scholz, S. 2022. Grouping of chemicals into mode of action classes by automated effect pattern analysis using the zebrafish embryo toxicity test. Archives of Toxicology, 1-17.

Conversion tool for non-capillary images

Images from other origin can be used as well but require automatic conversion to an image with a virtual capillary. This can be done with a KNIME workflow and an embedded imageJ macro. The workflow can handle multiple images simultaneously. An installation of KNIME and imageJ on your computer is required. Depending on the image quality certain parameters of the workflow may have to be adjusted. Images of embryos from a lateral orientation are required.

You will have to install various KNIME extentions in order to run the workflows (indicated for each workflow). Particularly the KNIME community contributions for image analysis will be required.

Subsequent data processing

Please note that the main purpose of FishInspector is providing the coordinates of features as JSON files. For subsequent analyses appropriate workflows are required. We had been mainly interested in concentration-response analysis of fish embryos exposed to chemicals. Therefore, various KNIME workflows with embedded R-scripts are provided for subsequent data processing and analysis. Prior to concentration-response analysis it is required to analyse the distribution/variability of features in control embryos. A range of a two-fold standard deviation around the mean is used as a threshold to indicate deviation from control embryos and calculation of quantal concentration-response curves.

In order to run the workflows a KNIME and R installation is required. You may have to install certain additional packages in R (please open workflows in KNIME for further instructions). Our workflows have been established and tested with R version 3.4.1 but other versions may work as well.

Future Developments

A new version (1.7) of FishInspector is close to be released. The new version has sigificantly improved. Most importantly, recognistion of structural features as now based on trained semantic segmentation models and metrics based on the coordinates of the structural features (e.g. calculation of body lenght, otolith-eye distance, yolk elongation, fin length have been included).

The new version will be officially released (as open source) upon publication. However, evaluation copies with principally unrestricted use can be provided on request. For this purpose contact Dr. Stefan Scholz (contact details below).

Acknowledgements

We acknowledge Tobias Kießling (TKS3) for implementation of our ideas into a user friendly program, establishing the MATLAB code and compiling the program as a standalone windows program. We also thank Prof. Chih Lai, St. Thomas University, MN, for support in implementing the deep learning approach in the latest (unpublished) version of FishInspector (1.7).

The FishInspector software was developed with support of various research grant. Initial support was from the ZFminus1 project, which is funded by the German Ministery of Education and Research (research funding scheme Alternatives to Animal Testing). The project ZFminus1 aimed at the reduction of the number of animal tests with mammals conducted for developmental toxicity testing, particularly at avoiding the need to test chemicals in a second mammalian species. Recently, further amendments of the software are supported by the BMBF grant "ZF-AOP" and the European Uniion project "PrecisionTox".

Contact

If you are interested in further developing the software and in collaborations, please do not hesitate to contact us.