Volume segmentation using convolutional neural networks with limited training data

Cheng, Hsueh-Chien; Varshney, Amitabh

doi:10.1109/icip.2017.8296349

Cited by 25 publications

(31 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We compare our method with both 2D methods and 3D methods, including both traditional methods based on hand-crafted features (Lucchi et al, 2013 ; Cetina et al, 2018 ; Peng and Yuan, 2020 ) and deep learning methods (Ronneberger et al, 2015 ; Çiçek et al, 2016 ; Cheng and Varshney, 2017 ; Xiao et al, 2018 ; Casser et al, 2020 ), on the EPFL dataset and Kasthuri++ dataset. Since our HED-Net takes 5-slice input, which is usually called 2.5D method, we also compare our method with 2D U-Net (Ronneberger et al, 2015 ) that takes five slices as input.…”

Section: Results and Analysismentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

“…Recently, various methods (Lucchi et al, 2013 ; Cheng and Varshney, 2017 ; Cetina et al, 2018 ; Xiao et al, 2018 ; Casser et al, 2020 ; Peng and Yuan, 2020 ; Yuan et al, 2021 ) have been introduced to address mitochondria segmentation. According to the features they used, mitochondria segmentation can be categorized into two classes: traditional methods with hand-crafted features (Lucchi et al, 2011 , 2013 ; Cetina et al, 2018 ; Peng and Yuan, 2020 ) and deep learning methods with automatically learned features (Cheng and Varshney, 2017 ; Xiao et al, 2018 ; Casser et al, 2020 ; Yuan et al, 2020 , 2021 ). Generally speaking, deep-learning-based methods, especially methods based on fully convolutional neural networks (Ronneberger et al, 2015 ; Litjens et al, 2017 ; Shelhamer et al, 2017 ), show better performance than traditional machine learning and computer vision methods (Lucchi et al, 2013 ; Cetina et al, 2018 ; Peng and Yuan, 2020 ).…”

Section: Introductionmentioning

confidence: 99%

“…Casser et al ( 2020 ) used a modified 2D U-Net with an on-the-fly data augmentation and Z-filtering postprocessing, and their model showed obviously improved performance over 2D U-Net. Cheng and Varshney ( 2017 ) improved the 2D/3D U-Net with factorized convolutions and online feature-level augmentations and showed improved results over the 2D/3D U-Net. Xiao et al ( 2018 ) proposed an effective approach using a modified 3D U-Net, a 3D residual convolutional network with deep supervision.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Hierarchical Encoder-Decoder With Soft Label-Decomposition for Mitochondria Segmentation in EM Images

2021

View full text Add to dashboard Cite

Semantic segmentation of mitochondria from electron microscopy (EM) images is an essential step to obtain reliable morphological statistics about mitochondria. However, automatically delineating plenty of mitochondria of varied shapes from complex backgrounds with sufficient accuracy is challenging. To address these challenges, we develop a hierarchical encoder-decoder network (HED-Net), which has a three-level nested U-shape architecture to capture rich contextual information. Given the irregular shape of mitochondria, we introduce a novel soft label-decomposition strategy to exploit shape knowledge in manual labels. Rather than simply using the ground truth label maps as the unique supervision in the model training, we introduce additional subcategory-aware supervision by softly decomposing each manual label map into two complementary label maps according to mitochondria's ovality. The three label maps are integrated with our HED-Net to supervise the model training. While the original label map guides the network to segment all the mitochondria of varied shapes, the auxiliary label maps guide the network to segment subcategories of mitochondria of circular shape and elliptic shape, respectively, which are much more manageable tasks. Extensive experiments on two public benchmarks show that our HED-Net performs favorably against state-of-the-art methods.

show abstract

Section: Results and Analysismentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Hierarchical Encoder-Decoder With Soft Label-Decomposition for Mitochondria Segmentation in EM Images

2021

View full text Add to dashboard Cite

show abstract

“…+ False Negative (25) commonly used in image segmentation. additionally, we report the VOC score, which is the mean between the IoU of the foreground class and the IoU of the background class, and was used in Lucchi et al (2015) in place of the IoU of the foreground class alone (Cheng and Varshney, 2017;Casser et al, 2018). These results are summarized in Table 4, averaging the performance across folds on the test set.…”

Section: Segmentation Photon Receptor Cells In 2d Retinal Imagesmentioning

confidence: 99%

Discriminative Training of Conditional Random Fields with Probably Submodular Constraints

Berman

Blaschko

2020

Int J Comput Vis

View full text Add to dashboard Cite

Problems of segmentation, denoising, registration and 3D reconstruction are often addressed with the graph cut algorithm. However, solving an unconstrained graph cut problem is NP-hard. For tractable optimization, pairwise potentials have to fulfill the submodularity inequality. In our learning paradigm, pairwise potentials are created as the dot product of a learned vector w with positive feature vectors. In order to constrain such a model to remain tractable, previous approaches have enforced the weight vector to be positive for pairwise potentials in which the labels differ, and set pairwise potentials to zero in the case that the label remains the same. Such constraints are sufficient to guarantee that the resulting pairwise potentials satisfy the submodularity inequality. However, we show that such an approach unnecessarily restricts the capacity of the learned models. Guaranteeing submodularity for all possible inputs, no matter how improbable, reduces inference error to effectively zero, but increases model error. In contrast, we relax the requirement of guaranteed submodularity to solutions that are probably approximately submodular. We show that the conceptually simple strategy of enforcing submodularity on the training examples guarantees with low sample complexity that test images will also yield submodular pairwise potentials. Results are presented in the binary and muticlass settings, showing substantial improvement from the resulting increased model capacity.

show abstract

Two Stream Active Query Suggestion for Active Learning in Connectomics

Lin

Wei

Jang

et al. 2020

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Volume segmentation using convolutional neural networks with limited training data

Cited by 25 publications

References 11 publications

Hierarchical Encoder-Decoder With Soft Label-Decomposition for Mitochondria Segmentation in EM Images

Hierarchical Encoder-Decoder With Soft Label-Decomposition for Mitochondria Segmentation in EM Images

Discriminative Training of Conditional Random Fields with Probably Submodular Constraints

Two Stream Active Query Suggestion for Active Learning in Connectomics

Contact Info

Product

Resources

About