Felix Bollenbeck

Differentiation of endosperm transfer cells of barley: a comprehensive analysis at the micro-scale.

Barley endosperm cells differentiate into transfer cells (ETCs) opposite the nucellar projection. To comprehensively analyse ETC differentiation, laser microdissection-based transcript and metabolite profiles were obtained from laser microdissected tissues and cell morphology was analysed. Flange-like secondary-wall ingrowths appeared between 5 and 7 days after pollination within the three outermost cell layers. Gene expression analysis indicated that ethylene-signalling pathways initiate ETC morphology. This is accompanied by gene activity related to cell shape control and vesicle transport, with abundant mitochondria and endomembrane structures. Gene expression analyses indicate predominant formation of hemicelluloses, glucuronoxylans and arabinoxylans, and transient formation of callose, together with proline and 4-hydroxyproline biosynthesis. Activation of the methylation cycle is probably required for biosynthesis of phospholipids, pectins and ethylene. Membrane microdomains involving sterols/sphingolipids and remorins are potentially involved in ETC development. The transcriptional activity of assimilate and micronutrient transporters suggests ETCs as the main uptake organs of solutes into the endosperm. Accordingly, the endosperm grows maximally after ETCs are fully developed. Up-regulated gene expression related to amino acid catabolism, C:N balances, carbohydrate oxidation, mitochondrial activity and starch degradation meets high demands for respiratory energy and carbohydrates, required for cell proliferation and wall synthesis. At 10 days after pollination, ETCs undergo further differentiation, potentially initiated by abscisic acid, and metabolism is reprogrammed as shown by activated storage and stress-related processes. Overall, the data provide a comprehensive view of barley ETC differentiation and development, and identify candidate genes and associated pathways.

Robust classification of the nutrition state in crop plants by hyperspectral imaging and artificial neural networks

Hyperspectral imaging of crop plants offers the means for a non-invasive, precise and high-throughput plant-phenotyping in plant research and precision agriculture. We already reported the successful separation of spectral signatures by means of unsupervised learning (e.g. clustering) of tobacco leaves grown from different genetic background and under different nutritional conditions [1,2]. In this contribution we evaluate supervised methods to predict the plants nutrition state by classification and whether they are robust towards dominant sources of data variation like leaf age or intra-leaf pixel position which are irrelevant for the task at hand. Support Vector Machine (SVM)[3], Supervised Relevance Neural Gas (SRNG) [4], Generalized Relevance Learning Vector Quantization (GRLVQ) [5] and a Radial Basis Function (RBF) Network [6] adopted to perform relevance learning as well were tested. Leaf age snowed the largest impact on classification performance, where SVM and RBF produced robust results while SRNG and GRLVQ methods were reduced to near guessing level. Three cameras covering the VIS/SWIR range were tested and relevance of spectral bands towards nutrition prediction were calculated.

Clustering of crop phenotypes by means of hyperspectral signatures using artificial neural networks

Hyperspectral imaging linked to subsequent neural networks based analysis has proven its suitability to unravel complex information in a number of different application areas, such as geology, defence, etc. The extension of this approach to crop plant research, plant breeding, and agriculture has started quite recently. Here, the image acquisition ranges from airborne sensing mainly for agricultural applications down to single leaf analysis in the context of precision and high-throughput plant phenotyping. All these applications have in common, that particular relevant compounds of the plant need to be determined by means of hyperspectral signatures as substitute to extensive biochemical analysis. This paper describes the quantitative assessment of a number of genetically different tobacco varieties (Nicotiana tabacum) that were grown under different environmental and nutritional conditions. The analysis of the measured hyperspectral signatures was done by artificial neural networks.

Fast registration-based automatic segmentation of serial section images for high-resolution 3-D plant seed modeling

We propose a deformation-based approach for fast and robust segmentation of histological section images into multiple tissues. Derived from deformable registration techniques, it does not solely rely on information present in the image, but uses a-priori information in terms of reference segmentations. The experimental evaluation against state-of-the-art feature based classifiers demonstrates the high performance in segmentation accuracy and the effectiveness of this approach. This serves as basis for processing high-resolution serial section datasets comprising several thousand images towards three-dimensional atlases of plant organs.

Three-Dimensional Multimodality Modelling by Integration of High-Resolution Interindividual Atlases and Functional MALDI-IMS Data

We present an approach for the analysis of phenotypic diversity in morphology and internal composition of biological specimen by means of high resolution 3-D models of developing barley grains. Three-dimensional histological structures are resolved by reconstructing specimen from large stacks of serially sectioned material, which is a preliminary for the spatial assignment of key tissues in differentiation. By sampling and constructing models at different developmental time steps from multiple individuals, we address two aims in a computational phenomics context: i) Generation of averaging atlases as structural references for integration of functional data, and ii) building the basis for a mathematical model of grain morphogenesis. We have established an algorithmic pipeline for automated processing of large image stacks towards phenotypic 3-D models and data-integration, comprising registration, multi-label segmentation, and alignment of functional measurements. The described algorithms allow high-throughput reconstruction and tissue recognition of datasets comprising thousands of images. The usefulness of the approach is demonstrated by automated model generation, allowing volumetric measurements of tissue composition, three-dimensional analysis of diversity, and the integration of MALDI-IMS data by mutual information based registration, which is a significant contribution to a systematic analysis of differentiation and development.

Fuzzy image segmentation by potential fields

Many natural phenomena, and human knowledge about these respectively, can only be described by gradually varying or diffuse entities. This generally motivates imprecise or fuzzy information processing. However, since the transition from exact to fuzzy descriptions offers not just more degrees of freedom but often completely different structures to specify particular knowledge, it requires extended methods or tools enabling the potential user to appropriately transfer his knowledge into a machine-readable form. In terms of image processing, fuzzy techniques have become widely spread. Nevertheless, just this knowledge transfer from a human expert to a potentially available computer system is still an open issue in many cases. The present paper addresses this by means of fuzzy image segmentation against the background of biomedical image processing, where, for example the borderline between adjacent tissues often can not be specified sharply and unequivocally. Despite its particular application in the described context of plant biology, the presented approach is much more versatile and can be applied to a large variety of similar problems.

A multivariate wavelet-PCA denoising-filter for hyperspectral images

In this paper we investigate the use of multivariate multiresolution principal component analysis for filtering and denoising of signals. From the proposed model we deduce several properties that particularly address the properties of hyper-spectral image data. We thereby aim at overcoming shortcomings of other methods close to the approach specifically for hyperspectral applications. The performance is evaluated by generating synthetic pure and noised signals from a physical model for spectral reflectance images. From benchmark experiments we deduce that the performance of the proposed method is equal or higher compared to univariate multiresolution denoising algorithms, while being less computationally complex. The described algorithm is used for processing of large close-range outdoor data sets of sensed crop plants.

Joint Registration and Segmentation of Histological Volume Data by Diffusion-Based Label Adaption

Three-dimensional serial section imaging delivers high spatial resolution and histological detail, which facilitates analysis of differentiation and development by exact labelling of tissues and cells, unknown to other 3-D imaging modalities. We propose an algorithm for interleaved reconstruction and segmentation of tissues in serial section volumes by diffusion-based registration and adaption of two-dimensional reference labellings. Iterative refinement of the global image congruence and local deformation of labellings delivers an efficient algorithm for processing of large volume data-sets. The benefits of the approach are shown by means of reconstruction and segmentation of giga-voxel serial section volumes of plant specimen.

Computational Intelligence in Biomedical Image Processing

The description of phenotypical properties of living organisms has emerged as an urging topic in biology. In the context of automated high-throughput processing of large stacks of serial section light-microscopic images the recognition and segmentation of all prevailing biological structures and tissues is a crucial task for high-throughput processing of such data.

Agent Transportation Layer Adaptation System

Heuristic computing has consolidated into two streams of research. One that personifies software to exhibit human behaviour and an oher that provides innovative software or smart products [1]. The Turing test [2] was pivotal in providing researchers with a generally accepted method of classifying the work that now defines the major problems pursued within Artificial Intelligence (AI). Cognitive Science is one of these fields and Research in Multi-Agent System (MAS) has revealed that Agents must enter into a voluntarily trust relationship in order to collaborate, otherwise the imposed goal(s) may be aborted or fail completely [3, 4]. Current agent architectures present a finite limit to functionality when supporting one or more of these paradigms.