Quantitative morphological transformation of vascular bundles in the culm of moso bamboo (Phyllostachys pubescens)

Vascular bundles of bamboo are determinants for mechanical properties of bamboo material and for physiological properties of living bamboo. The morphology of vascular bundles reflecting mechanical and physiological functions differs not only within internode tissue but also among different internodes in the culm. Although the distribution of vascular bundle fibers has received much attention, quantitative evaluation of the morphological transformation of vascular bundles associated with spatial distribution patterns has been limited. In this study deep learning models were used to determine quantitative changes in the distribution and morphology of vascular bundles in the culms of moso bamboo (Phyllostachys pubescens). A precise model for extracting vascular bundles from cross-sectional images was constructed using the U-Net model. Analyses of extracted vascular bundles from different internodes showed significant changes in vascular bundle distribution and morphology among internodes. Vascular bundles in lower internodes showed outer relative position and larger area than those in upper internodes. Aspect ratio and eccentricity indicate that vascular bundles in internodes near the base have more elliptical morphology, with a long axis in the radial direction. The variational autoencoder model using extracted vascular bundles enabled simulation of the morphological transformation of vascular bundles along with radial direction. These deep learning models enabled highly accurate quantification of vascular bundle morphologies, and will contribute to a further understanding of bamboo development as well as evaluation of the mechanical and physiological properties of bamboo.