A high-throughput pipeline for 3D reconstruction of soybean root system architecture using macro-CT scanning

Abstract. Root systems and their architectures are fundamental to plant development, resource acquisition, and stress resilience, yet root system phenotyping remains technically challenging because roots are hidden and the root systems architecture (RSA) is complex within opaque soil environments. In this study, we present a large-scale, non-destructive phenotyping framework for soybean root systems using a macro-CT scanner (CT: computed tomography) combined with a biologically informed 3D reconstruction pipeline. Soybean plants were cultivated in sand, a medium chosen for its favorable density contrast with root tissues and its compatibility with imaging-based root system analysis. An automated preprocessing pipeline first converted CT slice data into initial 3D root system point clouds, removing distant noise while leaving substantial noise near the roots. Subsequently, semi-automated particle tracking was conducted, initiated from manually segmented primary roots and selected lateral root tips. A distance-based pruning strategy was then employed to eliminate trajectories deviating into noise, while a neighborhood retrieval approach recovered root geometry around retained trajectories. Tested across diverse soybean genotypes in controlled greenhouse conditions, this approach robustly reconstructed coherent and biologically realistic RSAs from noisy CT datasets. This framework enables high-throughput, in-situ root system characterization at a scale large enough to be of value for modern breeding programs and provides a strong basis for further automation through deep learning-based primary root segmentation and root tip detection.