Cascade Decoder: A Universal Decoding Method For Biomedical Image Segmentation

Liang, Peixian; Chen, Jianxu; Zheng, Hao; Yang, Lin; Zhang, Yizhe; Chen, Danny Z.

doi:10.1109/isbi.2019.8759430

Cited by 11 publications

(8 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…1) Datasets: We evaluate our H-EMD on six 2D+time cell video datasets, including two in-house datasets (P. aeruginosa [54] and M. xanthus [44]) and four public datasets from the Cell Tracking Challenge [53] (Fluo-N2DL-HeLa, PhC-C2DL-PSC, PhC-C2DH-U373, and Fluo-N2DH-SIM+), and two 3D datasets, including one in-house Fungus [52], [55], [56] For the in-house datasets, instance segmentation annotations were manually labeled by experts. For the public datasets from the Cell Tracking Challenge, three types of instance segmentation annotations are provided for the training sequences: ground truth, gold truth, and silver truth.…”

Section: Methodsmentioning

confidence: 99%

“…The public Fluo-N2DL-HeLa dataset [53] [46] 96 Our H-EMD TUG-AT [10] CALT-US [13] BGU-IL [19] KIT-Sch-GE [18] DKFZ-GE [48] MU-Ba-US [49] UNSW-AU [50] UVA-NL [15] FR-Ro-GE [6] RWTH-GE [11] BRF-GE [51] KTH-SE [46] Threshold values Scores (%) Fungus. The in-house Fungus dataset [52], [55], [56] contains 4 3D electron microscopy images for segmenting fungus cells captured from body tissues of ants, whose 2D slices are of 853 × 877 pixels each. 16 slices in one stack are used as training data, and the other 3 stacks are used as test data.…”

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

H-EMD: A Hierarchical Earth Mover’s Distance Method for Instance Segmentation

Liang¹,

Zhang²,

Ding³

et al. 2022

IEEE Trans. Med. Imaging

Self Cite

View full text Add to dashboard Cite

Deep learning (DL) based semantic segmentation methods have achieved excellent performance in biomedical image segmentation, producing high quality probability maps to allow extraction of rich instance information to facilitate good instance segmentation. While numerous efforts were put into developing new DL semantic segmentation models, less attention was paid to a key issue of how to effectively explore their probability maps to attain the best possible instance segmentation. We observe that probability maps by DL semantic segmentation models can be used to generate many possible instance candidates, and accurate instance segmentation can be achieved by selecting from them a set of "optimized" candidates as output instances. Further, the generated instance candidates form a well-behaved hierarchical structure (a forest), which allows selecting instances in an optimized manner. Hence, we propose a novel framework, called hierarchical earth mover's distance (H-EMD), for instance segmentation in biomedical 2D+time videos and 3D images, which judiciously incorporates consistent instance selection with semantic-segmentation-generated probability maps. H-EMD contains two main stages. (1) Instance candidate generation: capturing instance-structured information in probability maps by generating many instance candidates in a forest structure. (2) Instance candidate selection: selecting instances from the candidate set for final instance segmentation. We formulate a key instance selection problem on the instance candidate forest as an optimization problem based on the earth mover's distance (EMD), and solve it by integer linear programming. Extensive experiments on eight biomedical video or 3D datasets demonstrate that H-EMD consistently boosts DL semantic segmentation models and is highly competitive with stateof-the-art methods.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

H-EMD: A Hierarchical Earth Mover’s Distance Method for Instance Segmentation

Liang¹,

Zhang²,

Ding³

et al. 2022

IEEE Trans. Med. Imaging

Self Cite

View full text Add to dashboard Cite

show abstract

“…A Brief Review of Related DL Techniques 3D Medical Image Segmentation. An array of 2D (Ronneberger, Fischer, and Brox 2015;Wolterink et al 2017;Shen et al 2017) and 3D (Ç içek et al 2016;Yu et al 2017;Liang et al 2019;Zheng et al 2019b) FCNs has been developed that significantly improved segmentation performance on various 3D medical image datasets (Pace et al 2015;Shen et al 2017). Scale-level (Ronneberger, Fischer, and Brox 2015) and block-level (He et al 2016;Huang et al 2017) skip-connections allow substantially deeper architecture design and ease the training by alleviating the vanishing gradient problem.…”

Section: (B))mentioning

confidence: 99%

“…3D image segmentation is one of the most important tasks in medical image applications, such as morphological and pathological analysis (Lee et al 2015b;Hou et al 2019), disease diagnosis (Pace et al 2015), and surgical planning (Kordon et al 2019). Recently, 3D deep learning (DL) models have been widely used in medical image segmentation and achieved state-of-the-art performance (Ronneberger, Fischer, and Brox 2015;Yu et al 2017;Liang et al 2019), most of which were trained with fully annotated 3D image stacks. The performance of DL models (when applied to testing images) is highly dependant on the amount and variety of labeled data used in model training.…”

Section: Introductionmentioning

confidence: 99%

An Annotation Sparsification Strategy for 3D Medical Image Segmentation via Representative Selection and Self-Training

Zheng¹,

Zhang²,

Yang³

et al. 2020

AAAI

Self Cite

View full text Add to dashboard Cite

Image segmentation is critical to lots of medical applications. While deep learning (DL) methods continue to improve performance for many medical image segmentation tasks, data annotation is a big bottleneck to DL-based segmentation because (1) DL models tend to need a large amount of labeled data to train, and (2) it is highly time-consuming and label-intensive to voxel-wise label 3D medical images. Significantly reducing annotation effort while attaining good performance of DL segmentation models remains a major challenge. In our preliminary experiments, we observe that, using partially labeled datasets, there is indeed a large performance gap with respect to using fully annotated training datasets. In this paper, we propose a new DL framework for reducing annotation effort and bridging the gap between full annotation and sparse annotation in 3D medical image segmentation. We achieve this by (i) selecting representative slices in 3D images that minimize data redundancy and save annotation effort, and (ii) self-training with pseudo-labels automatically generated from the base-models trained using the selected annotated slices. Extensive experiments using two public datasets (the HVSMR 2016 Challenge dataset and mouse piriform cortex dataset) show that our framework yields competitive segmentation results comparing with state-of-the-art DL methods using less than ∼20% of annotated data.

show abstract

“…Although supervised deep learning has achieved great success on medical image segmentation [15,17,25,34], it heavily relies on sufficient good-quality manual annotations which are usually hard to obtain due to expensive acquisition, data privacy, etc. Public medical image datasets are normally smaller than the generic image datasets (see Fig.…”

Section: Introductionmentioning

confidence: 99%

Hierarchical Self-Supervised Learning for Medical Image Segmentation Based on Multi-Domain Data Aggregation

Zheng

Han

Wang

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

A large labeled dataset is a key to the success of supervised deep learning, but for medical image segmentation, it is highly challenging to obtain sufficient annotated images for model training. In many scenarios, unannotated images are abundant and easy to acquire. Self-supervised learning (SSL) has shown great potentials in exploiting raw data information and representation learning. In this paper, we propose Hierarchical Self-Supervised Learning (HSSL), a new self-supervised framework that boosts medical image segmentation by making good use of unannotated data. Unlike the current literature on task-specific self-supervised pretraining followed by supervised fine-tuning, we utilize SSL to learn task-agnostic knowledge from heterogeneous data for various medical image segmentation tasks. Specifically, we first aggregate a dataset from several medical challenges, then pre-train the network in a self-supervised manner, and finally fine-tune on labeled data. We develop a new loss function by combining contrastive loss and classification loss, and pre-train an encoder-decoder architecture for segmentation tasks. Our extensive experiments show that multi-domain joint pretraining benefits downstream segmentation tasks and outperforms singledomain pre-training significantly. Compared to learning from scratch, our method yields better performance on various tasks (e.g., +0.69% to +18.60% in Dice with 5% of annotated data). With limited amounts of training data, our method can substantially bridge the performance gap with respect to denser annotations (e.g., 10% vs. 100% annotations).

show abstract

Cascade Decoder: A Universal Decoding Method For Biomedical Image Segmentation

Cited by 11 publications

References 17 publications

H-EMD: A Hierarchical Earth Mover’s Distance Method for Instance Segmentation

H-EMD: A Hierarchical Earth Mover’s Distance Method for Instance Segmentation

An Annotation Sparsification Strategy for 3D Medical Image Segmentation via Representative Selection and Self-Training

Hierarchical Self-Supervised Learning for Medical Image Segmentation Based on Multi-Domain Data Aggregation

Contact Info

Product

Resources

About