Rhizomatic: Next Generation Image Processing for In Situ Fine Root Measurement


Zack Stoecker-Sylvia ([email protected]), Ed Pier*, Dylan Kobayashi


Oceanit Laboratories, Honolulu, HI



Oceanit Laboratories have developed “Rhizomatic,” an automatic root analysis and characterization tool to help plant scientists better understand the “hidden half” of plant systems and produce research more quickly, more reliably, and more reproducibly. Rhizomatic has been developed as a web application to which users can upload images of in situ root systems taken by minirhizotrons or other imaging methods and either use existing prediction models to automatically find and characterize the roots present or, without a background in machine learning, easily train their own models customized to the specifics of their root study. Predicted roots are tracked over time across multiple sessions and growth and topology can be automatically measured and tracked.

Unlike other root prediction software, Rhizomatic only requires the standard, line-based root marking commonly performed in minirhizotron root tracing projects to train new models, enabling the development of models without new root tracing. Rhizomatic is currently compatible with Rootfly, RootSnap, rhizoTrak, and WinRHIZO Tron programs; marking software and other root tracing formats can be added easily. Additionally, as the model library grows, users will be able to reliably find pre-trained prediction models shared by other scientists, both reducing the need for human-traced roots and providing a tracing solution that will produce tracings and metrics that can be directly compared to each other without the concern that different root tracers have different standards. Collection of a body of ever-improving models is expected over time.

In addition to the automatic root prediction core, Rhizomatic includes an image processing pipeline that can address and fix common errors in root imaging. It can screen out bad images if they are significantly dissimilar to others from the same window, repair severely misaligned and mislabeled images by finding their correct window, and adjust root marks copied from a previous image and not updated to compensate for imager misalignment. Finally, as many prior studies may have incomplete tracings that may not produce the best prediction models, Rhizomatic’s root review tool can be used to review predicted roots against prior tracings, approve correct predictions, and then retrain with updated tracings to bootstrap into a higher quality root prediction model.