Benchmarking table recognition performance on biomedical literature on neurological disorders.

MOTIVATION: Table recognition systems are widely used to extract and structure quantitative information from the vast amount of documents that are increasingly available from different open sources. While many systems already perform well on tables with a simple layout, tables in the biomedical domain are often much more complex. Benchmark and training data for such tables are however very limited. RESULTS: To address this issue, we present a novel, highly curated benchmark dataset based on a hand-curated literature corpus on neurological disorders, which can be used to tune and evaluate table extraction applications for this challenging domain. We evaluate several state-of-the-art table extraction systems based on our proposed benchmark and discuss challenges that emerged during the benchmark creation as well as factors that can impact the performance of recognition methods. For the evaluation procedure, we propose a new metric as well as several improvements that result in a better performance evaluation. AVAILABILITY AND IMPLEMENTATION: The resulting benchmark dataset (https://zenodo.org/record/5549977) as well as the source code to our novel evaluation approach can be openly accessed. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Temperature Stable Ion Exchange Resins as Catalysts for the Manufacturing of Vitamin Precursors by Aldol Reaction.

This study explores the potential of robust, strongly basic type I ion exchange resins-specifically, Amberlyst® A26 OH and Lewatit® K 6465-as catalysts for the aldol condensation of citral and acetone, yielding pseudoionone. Emphasis is placed on their long-term stability and commendable performance in continuous operational settings. The aldol reaction, which traditionally is carried out using aqueous sodium hydroxide as the catalyst, holds the potential for enhanced sustainability and reduced waste production through the use of basic ion exchange resins in heterogeneous catalysis. Density Functional Theory (DFT) calculations are employed to investigate catalyst deactivation mechanisms. The result of these calculations indicates that the active sites of Amberlyst® A26 OH are cleaved more easily than the active sites of Lewatit® K 6465. However, the experimental data show a gradual decline in catalytic activity for both resins. Batch experiments reveal Amberlyst® A26 OH's active sites diminishing, while Lewatit® K 6465 maintains relative consistency. This points to distinct deactivation processes for each catalyst. The constant count of basic sites in Lewatit® K 6465 during the reaction suggests additional factors due to its unique polymer structure. This intriguing observation also highlights an exceptional temperature stability for Lewatit® K 6465 compared to Amberlyst® A26 OH, effectively surmounting one of the prominent challenges associated with the utilization of ion exchange resins in catalytic applications.

A 3D printed plant model for accurate and reliable 3D plant phenotyping.

BACKGROUND: This study addresses the importance of precise referencing in 3-dimensional (3D) plant phenotyping, which is crucial for advancing plant breeding and improving crop production. Traditionally, reference data in plant phenotyping rely on invasive methods. Recent advancements in 3D sensing technologies offer the possibility to collect parameters that cannot be referenced by manual measurements. This work focuses on evaluating a 3D printed sugar beet plant model as a referencing tool. RESULTS: Fused deposition modeling has turned out to be a suitable 3D printing technique for creating reference objects in 3D plant phenotyping. Production deviations of the created reference model were in a low and acceptable range. We were able to achieve deviations ranging from -10 mm to +5 mm. In parallel, we demonstrated a high-dimensional stability of the reference model, reaching only ±4 mm deformation over the course of 1 year. Detailed print files, assembly descriptions, and benchmark parameters are provided, facilitating replication and benefiting the research community. CONCLUSION: Consumer-grade 3D printing was utilized to create a stable and reproducible 3D reference model of a sugar beet plant, addressing challenges in referencing morphological parameters in 3D plant phenotyping. The reference model is applicable in 3 demonstrated use cases: evaluating and comparing 3D sensor systems, investigating the potential accuracy of parameter extraction algorithms, and continuously monitoring these algorithms in practical experiments in greenhouse and field experiments. Using this approach, it is possible to monitor the extraction of a nonverifiable parameter and create reference data. The process serves as a model for developing reference models for other agricultural crops.

3D U-Net Segmentation Improves Root System Reconstruction from 3D MRI Images in Automated and Manual Virtual Reality Work Flows.

Magnetic resonance imaging (MRI) is used to image root systems grown in opaque soil. However, reconstruction of root system architecture (RSA) from 3-dimensional (3D) MRI images is challenging. Low resolution and poor contrast-to-noise ratios (CNRs) hinder automated reconstruction. Hence, manual reconstruction is still widely used. Here, we evaluate a novel 2-step work flow for automated RSA reconstruction. In the first step, a 3D U-Net segments MRI images into root and soil in super-resolution. In the second step, an automated tracing algorithm reconstructs the root systems from the segmented images. We evaluated the merits of both steps for an MRI dataset of 8 lupine root systems, by comparing the automated reconstructions to manual reconstructions of unaltered and segmented MRI images derived with a novel virtual reality system. We found that the U-Net segmentation offers profound benefits in manual reconstruction: reconstruction speed was doubled (+97%) for images with low CNR and increased by 27% for images with high CNR. Reconstructed root lengths were increased by 20% and 3%, respectively. Therefore, we propose to use U-Net segmentation as a principal image preprocessing step in manual work flows. The root length derived by the tracing algorithm was lower than in both manual reconstruction methods, but segmentation allowed automated processing of otherwise not readily usable MRI images. Nonetheless, model-based functional root traits revealed similar hydraulic behavior of automated and manual reconstructions. Future studies will aim to establish a hybrid work flow that utilizes automated reconstructions as scaffolds that can be manually corrected.

Pheno4D: A spatio-temporal dataset of maize and tomato plant point clouds for phenotyping and advanced plant analysis.

Understanding the growth and development of individual plants is of central importance in modern agriculture, crop breeding, and crop science. To this end, using 3D data for plant analysis has gained attention over the last years. High-resolution point clouds offer the potential to derive a variety of plant traits, such as plant height, biomass, as well as the number and size of relevant plant organs. Periodically scanning the plants even allows for performing spatio-temporal growth analysis. However, highly accurate 3D point clouds from plants recorded at different growth stages are rare, and acquiring this kind of data is costly. Besides, advanced plant analysis methods from machine learning require annotated training data and thus generate intense manual labor before being able to perform an analysis. To address these issues, we present with this dataset paper a multi-temporal dataset featuring high-resolution registered point clouds of maize and tomato plants, which we manually labeled for computer vision tasks, such as for instance segmentation and 3D reconstruction, providing approximately 260 million labeled 3D points. To highlight the usability of the data and to provide baselines for other researchers, we show a variety of applications ranging from point cloud segmentation to non-rigid registration and surface reconstruction. We believe that our dataset will help to develop new algorithms to advance the research for plant phenotyping, 3D reconstruction, non-rigid registration, and deep learning on raw point clouds. The dataset is freely accessible at https://www.ipb.uni-bonn.de/data/pheno4d/.

Object class segmentation of RGB-D video using recurrent convolutional neural networks.

Object class segmentation is a computer vision task which requires labeling each pixel of an image with the class of the object it belongs to. Deep convolutional neural networks (DNN) are able to learn and take advantage of local spatial correlations required for this task. They are, however, restricted by their small, fixed-sized filters, which limits their ability to learn long-range dependencies. Recurrent Neural Networks (RNN), on the other hand, do not suffer from this restriction. Their iterative interpretation allows them to model long-range dependencies by propagating activity. This property is especially useful when labeling video sequences, where both spatial and temporal long-range dependencies occur. In this work, a novel RNN architecture for object class segmentation is presented. We investigate several ways to train such a network. We evaluate our models on the challenging NYU Depth v2 dataset for object class segmentation and obtain competitive results.

Two-layer contractive encodings for learning stable nonlinear features.

Unsupervised learning of feature hierarchies is often a good strategy to initialize deep architectures for supervised learning. Most existing deep learning methods build these feature hierarchies layer by layer in a greedy fashion using either auto-encoders or restricted Boltzmann machines. Both yield encoders which compute linear projections of input followed by a smooth thresholding function. In this work, we demonstrate that these encoders fail to find stable features when the required computation is in the exclusive-or class. To overcome this limitation, we propose a two-layer encoder which is less restricted in the type of features it can learn. The proposed encoder is regularized by an extension of previous work on contractive regularization. This proposed two-layer contractive encoder potentially poses a more difficult optimization problem, and we further propose to linearly transform hidden neurons of the encoder to make learning easier. We demonstrate the advantages of the two-layer encoders qualitatively on artificially constructed datasets as well as commonly used benchmark datasets. We also conduct experiments on a semi-supervised learning task and show the benefits of the proposed two-layer encoders trained with the linear transformation of perceptrons.