International Workshop on Image Analysis Methods for Plant Science University of Nottingham Jubilee Campus September 6 th 2012

s Talks 1. Automated phenotyping of root behavior in mutant and naturally varying populations Prof Edgar Spalding, University of Wisconsin Plant biologists are justified in pursuing a comprehensive mapping of genotype to phenotype because the goal is important and the genomic half of the equation in some reference systems is highly developed. However, methodologies for quantifying phenotypes with the same precision and throughput are needed if the desired map is to be attained. Unlike genotypes, phenotypes continuously change during development and differently in different environmental conditions. Automated image acquisition and feature extraction (machine vision) is increasingly seen as a method for quantitatively describing the 'phenotype space' as required for producing the ultimate genotype-to-phenotype description. In this presentation, I will describe our efforts to quantify time-dependent phenotypes displayed by Arabidopsis and maize seedling roots in genetically structured populations. I will emphasize technical challenges resulting from the age-old tension between throughput and resolution, and from the newer problem of coupling automated image acquisition to the computational resources needed for analysis and subsequent statistical genetic modeling. One example of a completed study will show how a subtle phenotype resulting from mutation of a glutamate receptor-like gene could be computationally discovered and described. Another will show how the time axis can be added to a quantitative trait locus (QTL) map of root gravitropism as a result of the machine vision approach. 2. Image processing in field crop phenotyping Dr Norbert Kirchgessner, Prof Achim Walter, ETH Zurich In the crop science group at the Institute of Agricultural Sciences, ETH Zürich, we seek to provide innovative pathways to identify and generate more versatile and efficiency-oriented crop production systems. Imaging procedures acting at different spatial and temporal scales are core elements of the toolbox of our interdisciplinary team. In climate chambers, greenhouses and on field sites, we will apply visual, nearinfrared and thermal imaging to quantify shoot and root architecture, dynamic and short-term growth processes as well as photosynthesis, gas exchange and compound composition of major and alternative crops alike. These 'phenotyping' analyses, in concert with a range of approaches from plant ecophysiology, breeding and molecular analysis, will help elucidating differences between plant genotypes. Moreover they will facilitate the optimization of crop production systems to regionally differing ecological niches or to an altering climate. Most importantly, these phenotyping approaches will facilitate an improved understanding of basic rules governing plant-environment-management interactions. This in turn is a necessary prerequisite to ameliorate the knowledge transfer between lab and field as well as between plant biology and agricultural sciences, thereby allowing for improved agricultural plants and practices in the future. Crucial to a refined understanding of plant performance in the field is a more detailed understanding of dynamic growth processes. In the past, a focal point of some of our team members was hence the refinement and application of optical-flow-based growth analysis techniques, which also required the establishment of physiologial coordinate systems of plant organs. In the future, a focus of our lab phenotyping activities will be put on non-destructive growth analyses of roots via CT. On our experimental field site, we will establish a rigged camera system (field phenotyping platform FIP) equipped with multiple sensors. 3. Automatic plant structure analysis in the field Dr Alexander Bucksch, Georgia Institute of Technology Quantitative description of branching structures in plants is important for creating and validating mathematical growth models, estimating the plants interactions with the environment and to describe phenotypic traits of the plants. An additional challenge arises when the extraction of such branching structures is carried out under field conditions. 2D digital photography and emerging 3D technologies like terrestrial laser scanning enable us to capture the branching hierarchy in a short time. However, there are many challenges to decipher the resulting 2D or 3D data and to extract the captured branching structure. We give 2 examples of automatic parameter extraction under field conditions: 1.) Tree canopies measured with terrestrial laser scanners and represented as 3D point clouds. We analyze the point clouds by means of optimal network theory. 2.) Bean roots represented as a 2D digital photograph to extract phenotyping parameters in the field Both examples rely on (distinct) skeletonization algorithms to represent and analyze the branching structure. We validate both methods against manually collected field data. For the tree canopies the branching hierarchy and branch lengths were measured in the field. Based on these two quantities we discriminate two growing conditions for a test set of six trees, by analyzing the side-branch-statistics and the internode length. In case of the roots we followed the Shovelomics protocol for manual root phenotyping in the field. We extracted basal root number and angles from the 2D images as typical parameters for this protocol. 4. Mars-Alt v2: toward a robust imaging platform for studying multicellular systems Prof Christophe Godin, INRIA MARS-ALT is a pipeline of algorithms that has been recently developed to segment cells and to track cell lineages in multicellular organisms observed with laser microscopy. In this talk I will present the recent developments made to improve the robustness of the system in addressing a greater number of biological systems, using different laser imaging protocols, making the system available to a wider number of users and developers. 5. Extraction and analysis of structured networks of plant cells from confocal images Dr Michael Pound, University of Nottingham It is increasingly important in life sciences that many cell-scale and tissue-scale measurements are quantified from confocal microscope images. However, extracting and analyzing large-scale confocal image data sets represents a major bottleneck for researchers. To aid this process, CellSeT software has been developed, which utilizes tissue-scale structure to help segment individual cells. We provide examples of how the CellSeT software can be used to quantify fluorescence of hormone-responsive nuclear reporters, determine membrane protein polarity, extract cell and tissue geometry for use in later modeling, and take many additional biologically relevant measures using an extensible plug-in toolset. Application of CellSeT promises to remove subjectivity from the resulting data sets and facilitate higher-throughput, quantitative approaches to plant cell research 6. Local interest point detectors for cell detection in digital microscopy images Dr Pedro Quelhas, Instituto de Engenharia Biomedica Porto Automatic image analysis approaches used in microscopy cell image analysis have enabled the objective analysis of large scale biology research image data, removing a large workload from the biology researcher. However, most of the approaches used in this automated analysis systems are based on classic automatic image segmentation, which is neither trivial to apply nor robust to image quality variations. A way to analyze digital microscopy images with increased robustness and performance is through the use of local interest point detectors. This has gained recent popularity due to the performance in high noise low contrast situations where automatic image segmentation fails. Local interest point detectors are designed to have high response in specific locations in the image where certain shapes are visible. By relying in prior knowledge of the shape of cells in microscopy images under analysis their application for this task becomes trivial. Additionally, the parameters needed in these methods are much more intuitive as they relate directly to the shape and size of the cells. 7. Rapid seedling reconstruction in 3D and other vision research projects in the Agrifood industry Dr Rick van de Zedde, Wageningen UR Rick van de Zedde is a senior scientist at Wageningen UR in The Netherlands, with an M. Sc degree in Artificial Intelligence. His specialism is computer vision/ robotics. He is the projectmanager of several computer vision related research projects, and is the coordinator of the centre of expertise for computer vision in Wageningen UR – GreenVision (greenvision.wur.nl). Together with various machine builders, Wageningen UR GreenVision is developing novel sensor modules for diverse machine-automation solutions in the Agrifood industry. Several research projects will be presented. One interesting project is about automating the manual grading of seedlings at nurseries. Very labour-intensive, time consuming and expensive. We have developed an automated system that uses 3D-vision techniques to grade seedlings according to their quality. In our system the 3D plant model is created based on the information of 10 cameras. The current version processes 18.500 seedlings on a single line conveyer belt, the software reconstructs 3D model within 30ms. 8. Imaging Arabidopsis shoots: from rosette to cell Dr Stijn Dhondt, Dirk Inzé, VIB Plant Systems Biology Phenotyping is widely recognized as the most laborious and technically challenging part in the process of understanding the genomic code and implementing this knowledge towards applications, making it costly and time consuming. Nevertheless, this ‘phenotyping bottleneck’ can now be addressed by combining novel image capturing technologies, robotics, image analysis, and data integration. We developed a series of imaging setups and data analysis pipelines to increase the throughput of a number of phenotypic studies, both on the organismal, organ, and cellular level. To cover these different levels, our image analysis algorithms handle a variety of input images. Visible and NIR imaging is applied to extract rosette growth rates, dark field imaging is employed to segment venation patterns in leaves, and differential interference contrast imaging is utilized in a proof of concept to extract the cellular content of the leaf's epidermis. Furthermore, we investigate the possibility to use X-ray computed tomography to study overall plant morphology and cellular organization in three dimensions. Such imaging tools can ensure a fast and precise phenotypic description of biological structures, limiting laborious, costly, and often repetitive manual intervention. Furthermore, in comparison to manual screening experiments, often more phenotypic traits are recorded, increasing the throughput and output of phenotypic analyses. 9. 3D reconstruction of plant canopies – parameterization, registration and further image analysis of stereo images of sugar beet and barley Dr Mark Mueller-Linow, Research Center Juelich Three-dimensional canopies form complex architectures with spatially and temporally changing distributions of leaf orientations which serve as important indicators for canopy function and plant state. The 3-d reconstruction from stereo images taken by ordinary SLR-cameras is a first step to analyze these structural properties. Our methodology comprises several automated steps including camera calibration, image rectification, area of interest selection, correlation-based block methods to solve the correspondence problem and the implementation of several filters for image post-processing, e.g. the detection of occlusions, outliers and non-plant background. Further image analysis of the 3-d stereos uses image segmentation techniques as a prerequisite in order to derive the leaf angle distribution at the leaf and at the pixel level. These imaging processing techniques have been applied on stereo images of sugar beet and barley which have been recorded at our outdoor research site in Klein-Altendorf (University of Bonn). At the same time hyperspectral images were co-registered for later matching and parametrization of hyperspectral data by the 3-d canopy structure . The methodology will be overviewed and first results on sugar beet and barley will be given. 10. SmartRoot: A novel image analysis toolbox enabling quantitative analysis of root system architecture Mr Guillaume Lobet, Université catholique de Louvain SmartRoot is a novel, semi-automated image analysis software to streamline the quantitative analysis of root growth and architecture of complex root systems. The software combines a vectorial representation of root objects with a powerful tracing algorithm which accommodates to a wide range of image sources and quality. The root system is treated as a collection of roots (possibly connected) that are individually represented as parsimonious sets of connected segments. Pixel coordinates and grey level are therefore turned into intuitive biological attributes such as segment diameter and orientation, distance to any other segment or topological position. As a consequence, user interaction and data analysis directly operate on biological entities (roots) and are not hampered by the spatially discrete, pixel-based nature of the original image. The software supports a sampling-based analysis of root system images, in which detailed information is collected on a limited number of roots selected by the user according to speciifc research requirements. SmartRoot, is an operating system independent freeware based on ImageJ and relies on cross-platform standards for communication with data analysis softwares.