Digitally deconstructing leaves in 3D using X-ray microcomputed tomography and machine learning

Authors: THEROUX-RANCOURT, GUILLAUME - University Of Natural Resources & Applied Life Sciences - Austria; JENKINS, MATTHEW - University Of California, Davis; BRODERSEN, CRAIG - Yale University; McElrone, Andrew; FORRESTEL, ELISABETH - University Of California, Davis; EARLES, J - University Of California, Davis

Submitted to: Applications in Plant Sciences
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/20/2020
Publication Date: 7/31/2020
Citation: Theroux-Rancourt, G., Jenkins, M.R., Brodersen, C.R., McElrone, A.J., Forrestel, E.J., Earles, J.M. 2020. Digitally deconstructing leaves in 3D using X-ray microcomputed tomography and machine learning. Applications in Plant Sciences. 8(7). Article e11380. https://doi.org/doi:10.1002/aps3.11380.
DOI: https://doi.org/10.1002/aps3.11380

Interpretive Summary:

Technical Abstract: X-ray microcomputed tomography (microCT) can be used to measure 3D leaf internal anatomy, providing a more holistic view of tissue organization in relation to leaf function. Previously, the substantial time needed by the user to segment tissues within microCT images limited this technique to small datasets, thus restricting its utility for plant phenotyping experiments and, more generally, limiting our confidence in the conclusion of these studies due to low replication numbers. We present a Python-based random-forest machine learning segmentation and 3D leaf anatomical traits quantification program, which dramatically reduces the time required to process leaf microCT scans. By training the model on six hand segmented image slices the program can achieve more than 90% accuracy in background and tissue automated segmentation, except for veins and bundle sheaths when taken separately and not as a single tissue. Overall, this 3D segmentation and quantification pipeline will reduce one of the major barriers to using X-ray microCT imaging in high-throughput plant phenotyping.