Collaborative Research: ABI Innovation: Algorithms for recovering root architecture from 3D imaging

点击翻译按钮获取中文摘要

Roots, the "hidden" half of the plant, play many critical roles in the plant's development such as the uptake of water and nutrients, providing anchorage, and stabilizing the soil. These functions, in turn, are closely associated with the root architecture. Quantifying root architecture is not only a fundamental aspect of plant science but is a critical component in crop breeding for sustainable agriculture. Recent advances in 3D imaging (e.g., CT and MRI) have made it possible to capture 3D root structures in natural growing environments and monitor their growth over time. Unfortunately, the potential of the imaging techniques has been largely held back by the lack of effective computational tools for interpreting the images and distilling biological insights. This project, to be conducted by a group of computer scientists, mathematicians, and biologists across three institutes in the St. Louis region, aims at developing efficient and robust computational methods for automated analysis of root architecture from 3D images. The research looks a step ahead of the current cutting edge phenotype data collection, to how we will derive accurate representations of growing root systems, and therefore gain insight into the plant phenome. The team is committed to providing training to more than ten students over the course of the project, leveraging the existing NSF REU programs at two of the institutes. The team will pursue outreach activities not only within the research communities but also locally in the St. Louis area with a focus on grade schools.Deriving root architecture from 3D images involves a number of technically challenging tasks, including inferring individual roots from a segmented image, reconstructing their branching structure, and tracking the architecture in a time series of segmented images. This research draws from, and extends upon, methods from computer graphics and computational geometry to address these tasks. Specifically, the research will develop three novel classes of methods. Given a noise-ridden root segmentation, the first class of methods produces a curve skeleton that captures the topology and branching structure of the root system. The second class then uses the curve skeleton to automatically infer architectural components such as the root hierarchy and types. The third class improves the accuracy of the algorithms in the 1rst two classes by utilizing a sequence of segmentations and further annotates the root architecture with a time function. These algorithms enable the extraction of detailed root traits for root phenotyping, and both the algorithms and traits will be evaluated by a suite of representative real-world imaging data.Besides the design of automatic algorithms, a graphical software will be prototyped that offers fast and interactive means to inspect and edit the results produced by the algorithms. The software will be tested by biologists in the team and freely distributed to the research community.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.