Fig 5 - uploaded by Hendrik De Villiers
Content may be subject to copyright.
Model B. All convolutional kernel strides, including the ConvSoftmax output layer, are 1. This model produces patch predictions (see Figure 2b).

Model B. All convolutional kernel strides, including the ConvSoftmax output layer, are 1. This model produces patch predictions (see Figure 2b).

Source publication
Conference Paper
Full-text available
Convolutional neural networks have shown great promise in both general image segmentation problems as well as bioimage segmentation. In this paper, the application of different convolutional network architectures is explored on the C. elegans live/dead assay dataset from the Broad Bioimage Benchmark Collection. These architectures include a standar...

Contexts in source publication

Context 1
... simplest FCN model, model B (depicted in Figure 5) has four convolutional layers. The first three use PReLU, the last uses softmax normalization over its feature maps. ...
Context 2
... an alternative to the patch size shown in Figure 5, a smaller as well as a larger patch was explored to determine what influence the patch size has on the performance of model B. The smaller patch, 27 × 27, was chosen in such a way that the output size was a single pixel. Performance was similar to that of the model in Section III-B, the difference being that it had significantly lower segmentation precision. ...
Context 3
... simplest model, model B (Figure 5), was subjected to a 5-fold cross validation, using a patch target instead of single pixel target, as shown in Figure 2b. The results are shown in Table I, as well as Figures 8d and 9d. ...

Similar publications

Article
Full-text available
In recent years, with the continuous breakthrough in deep learning, convolutional neural networks (CNNs) have shown great potential for semantic segmentation. CNNs achieve better results by deepening or widening the network, but they increase the utilization rate of computing resources and even have the phenomenon of the vanishing gradient. A new c...
Preprint
Full-text available
We present an efficient convolution kernel for Convolutional Neural Networks (CNNs) on unstructured grids using parameterized differential operators while focusing on spherical signals such as panorama images or planetary signals. To this end, we replace conventional convolution kernels with linear combinations of differential operators that are we...
Article
Full-text available
We present a method for skin lesion segmentation for the ISIC 2017 Skin Lesion Segmentation Challenge. Our approach is based on a Fully Convolutional Network architecture which is trained end to end, from scratch, on a limited dataset. Our semantic segmentation architecture utilizes several recent innovations in particularly in the combined use of...
Article
Full-text available
We consider the problem of semantic image segmentation using deep convolutional neural networks. We propose a novel network architecture called the label refinement network that predicts segmentation labels in a coarse-to-fine fashion at several resolutions. The segmentation labels at a coarse resolution are used together with convolutional feature...

Citations

... To overcome the scarcity of dataset in segmentation of bioimages, a CNN network is introduced in [102]. It is based on FCN however, it has five convolutional layers using PReLU, with one max pooling layer attached to the first layer, attached to the last layer is a deconvolutional layer, which upsamples the feature maps before it is given to a convsoftmax layer. ...
Article
Full-text available
Microorganisms play a great role in ecosystem, wastewater treatment, monitoring of environmental changes and decomposition of waste materials. However, some of them are harmful to humans and animals such as tuberculosis bacteria and plasmodium. In such course, it is important to identify, track, analyse, consider the beneficial side and get rid of the negative effects of microorganisms using fast, accurate and reliable methods. In recent decades, image analysis techniques have been used to address the drawbacks of manual traditional approaches in the identification and analysis of microorganisms. As image segmentation being an important step (technique) in the detection, tracking, monitoring, feature extraction, modelling and analysis of microorganisms, different methods have been deployed, from classical approaches to current deep neural networks upon different challenges on microorganism images. This survey comprehensively analyses the various studies focused on developing microorganism image segmentation methods in the last 30 years (since 1989). In this survey, segmentation methods are categorised into classical and machine learning methods. Furthermore, these methods are sub-categorized into threshold based, region based and edge based which belong to classical methods, supervised and unsupervised machine leaning based methods which belong to machine learning category. A growth trend of different methods and most frequently used methods in each category are meticulously analysed. A clear explanation of the suitability of these methods for different segmentation challenges encountered on microscopic microorganism images is also enlightened.
... For training, the provided set included a 2×2×1.5 um 3 volume imaged from 30 sections and publicly available manual segmentations. For testing, the provided set included only image data, with segmentations kept [34] 0.977141154 0.987534429 CUMedVision [8] 0.976824580 0.988645822 LSTM [30] 0.975366444 0.987425430 Our approach (May 2016) 0.972797555 0.987597732 U-net [25] 0.972760748 0.986616590 (a) (b) Figure 6. Prediction on test data (Slice 28/30). ...
Article
Full-text available
Electron microscopic connectomics is an ambitious research direction with the goal of studying comprehensive brain connectivity maps by using high-throughput, nano-scale microscopy. One of the main challenges in connectomics research is developing scalable image analysis algorithms that require minimal user intervention. Recently, deep learning has drawn much attention in computer vision because of its exceptional performance in image classification tasks. For this reason, its application to connectomic analyses holds great promise, as well. In this paper, we introduce a novel deep neural network architecture, FusionNet, for the automatic segmentation of neuronal structures in connectomics data. FusionNet leverages the latest advances in machine learning, such as semantic segmentation and residual neural networks, with the novel introduction of summation-based skip connections to allow a much deeper network architecture for a more accurate segmentation. We demonstrate the performance of the proposed method by comparing it with state-of-the-art electron microscopy (EM) segmentation methods from the ISBI EM segmentation challenge. We also show the segmentation results on two different tasks including cell membrane and cell body segmentation and a statistical analysis of cell morphology.
... The aforementioned U-net architecture also won the ISBI 2015 cell tracking challenge, and forms the basis of this study. In our previous work [6], a similar architecture to U-net was shown to achieve state-of-the-art performance on the BBBC C. elegans live/dead assay data set. In Poudel et al. [7], a recurrent FCN was proposed that achieved state-of-the-art performance on two segmentation data sets, the MICCAI 2009 LV segmentation challenge and the PRETERM data set. ...
Conference Paper
Full-text available
Unsupervised pre-training of neural networks has been shown to act as a regularization technique, improving performance and reducing model variance. Recently, fully convolutional networks (FCNs) have shown state-of-the-art results on various semantic segmentation tasks. Unfortunately, there is no efficient approach available for FCNs to benefit from unsupervised pre-training. Given the unique property of FCNs to output segmentation maps, we explore a novel variation of unsupervised pre-training specifically designed for FCNs. We extend an existing FCN, called U-net, to facilitate end-to-end unsupervised pre-training and apply it on the ISBI 2012 EM segmentation challenge data set. We performed a battery of significance tests for both equality of means and equality of variance, and show that our results are consistent with previous work on unsupervised pre-training obtained from much smaller networks. We conclude that end-to-end unsupervised pre-training for FCNs adds robustness to random initialization, thus reducing model variance.
Thesis
Modern microscopy techniques acquire images at very high rates, high spatial resolution and with a large field of view. To analyze the large image data-sets acquired with such microscopes, accurate and scalable automated analysis is desperately needed. A key component is the instance segmentation of structures of interest, such as cells, neurons or organelles. In this thesis, we develop scalable methods for boundary based instance segmentation. We make use of Lifted Multicut graph partitioning and develop a method achieving state-of-the-art results on challenging benchmark data-sets. In order to scale this approach up, we introduce a new approximate solver for Multicut and Lifted Multicut, which can solve problems that were previously infeasible. We further establish a method to incorporate domain knowledge into the segmentation problem, which can significantly improve quality. To overcome the brittleness of seeded watersheds, used extensively in segmentation for microscopy, we introduce the Mutex Watershed. This efficient algorithm can segment images directly from pixels without the need for seeds or thresholds. Finally, we apply our methods in collaborative work, demonstrating their utility to answer biological research questions. In summary, our contributions enable scalable instance segmentation, thus eliminating one of the major obstacles to the automated analysis of large microscopy data-sets.
Article
Full-text available
Electron microscopy (EM) image segmentation plays an important role in computer-aided diagnosis of specific pathogens or disease. However, EM image segmentation is a laborious task and needs to impose experts knowledge, which can take up valuable time from research. Convolutional neural network (CNN)-based methods have been proposed for EM image segmentation and achieved considerable progress. Among those CNN-based methods, UNet is regarded as the state-ofthe- art method. However, the UNet usually has millions of parameters to increase training difficulty and is limited by the issue of vanishing gradients. To address those problems, the authors present a novel highly parameter efficient method called DenseUNet, which is inspired by the approach that takes particular advantage of recent advances in both UNet and DenseNet. In addition, they successfully apply the weighted loss, which enables us to boost the performance of segmentation. They conduct several comparative experiments on the ISBI 2012 EM dataset. The experimental results show that their method can achieve state-of-the-art results on EM image segmentation without any further post-processing module or pre-training. Moreover, due to smart design of the model, their approach has much less parameters than currently published encoder- decoder architecture variants for this dataset.