DeepCenterline: A Multi-task Fully Convolutional Network for Centerline Extraction
Zhihui Guo,Junjie Bai,Lu Yi,Xin Wang,Kunlin Cao,Qi Song,Milan Sonka,Youbing Yin +7 more
- 02 Jun 2019
- pp 441-453
TL;DR: This is the first deep-learning based centerline extraction method that guarantees single-pixel-wide centerline for a complex tree-structured object.
read more
Abstract: A novel centerline extraction framework is reported which combines an end-to-end trainable multi-task fully convolutional network (FCN) with a minimal path extractor. The FCN simultaneously computes centerline distance maps and detects branch endpoints. The method generates single-pixel-wide centerlines with no spurious branches. It handles arbitrary tree-structured object with no prior assumption regarding depth of the tree or its bifurcation pattern. It is also robust to substantial scale changes across different parts of the target object and minor imperfections of the object’s segmentation mask. To the best of our knowledge, this is the first deep-learning based centerline extraction method that guarantees single-pixel-wide centerline for a complex tree-structured object.
read more
Chat with Paper
AI Agents for this Paper
Find similar papers on Google Scholar, PubMed and Arxiv
Write a critical review of this paper
Analyze citations of this paper to find unaddressed research gaps
Figures
Fig. 5: Visual comparison of two methods and Hausdorff distance distributions. a) Red is centerlines from DeepCL. Cyan is centerlines from baseline method. Green shows the overlap of both centerlines. First row: DeepCL finds branches missed by baseline method. Second row: DeepCL avoids wrong bifurcations generated by baseline. Third row: DeepCL generates centerlines well-positioned at central location, avoiding taking straight shortcuts at complex bifurcation regions or tortuous segments. The last figure with red border shows a failure case for both DeepCL and baseline. b) Hausdorff distance distribution from voxels to the nearest centerline points for both methods. 
Table 2: Difference in performance of DeepCL and baseline. 
Fig. 1: Schematic workflow of DeepCenterline 
Fig. 2: The proposed multi-task FCN architecture. The input is 3D segmentation mask volume. The two tasks, centerline distance map and endpoint confidence map computation, share the same encoder path and have separate decoder paths. Skip-connections are added among features of same scale to facilitate good use of information. An attention module is added for the centerline distance map task to further boost accuracy. 
Fig. 3: Spatial-wise and channel-wise attention 
Fig. 6: Comparison of centerline distance map prediction with and without attention. a) Coronary artery segmetnation mask. b) A cross-sectional view of segmentation mask. c) Centerline distance map without attention module. d) Centerline distance map with attention module. e) Centerline distance map values at the profile line shown as double-arrowed line in b). With attention, the centerline distance map shows a high peak around centerline instead of a plateau by the model without attention.
Citations
Deep learning for cardiac image segmentation: A review
Chen Chen,Chen Qin,Huaqi Qiu,Giacomo Tarroni,Giacomo Tarroni,Jinming Duan,Wenjia Bai,Daniel Rueckert +7 more
TL;DR: In this article, a review of deep learning-based segmentation methods for cardiac image segmentation is provided, which covers common imaging modalities including magnetic resonance imaging (MRI), computed tomography (CT), and ultrasound.
A Review of Deep Learning in Medical Imaging: Imaging Traits, Technology Trends, Case Studies With Progress Highlights, and Future Promises
S. Kevin Zhou,Hayit Greenspan,Christos Davatzikos,James S. Duncan,Bram van Ginneken,Anant Madabhushi,Jerry L. Prince,Daniel Rueckert,Ronald M. Summers +8 more
- 26 Feb 2021
TL;DR: In this paper, the authors present traits of medical imaging, highlight clinical needs and technical challenges in medical imaging and describe how emerging trends in deep learning are addressing these issues, and conclude with a discussion and presentation of promising future directions.
612
A review of deep learning in medical imaging: Imaging traits, technology trends, case studies with progress highlights, and future promises
S. Kevin Zhou,Hayit Greenspan,Christos Davatzikos,James S. Duncan,Bram van Ginneken,Anant Madabhushi,Jerry L. Prince,Daniel Rueckert,Ronald M. Summers +8 more
TL;DR: This survey article presents traits of medical imaging, highlights both clinical needs and technical challenges in medical Imaging, and describes how emerging trends in DL are addressing these issues, including the topics of network architecture, sparse and noisy labels, federating learning, interpretability, uncertainty quantification, and so on.
465
Learning tree-structured representation for 3D coronary artery segmentation.
TL;DR: A novel tree-structured convolutional gated recurrent unit (ConvGRU) model is proposed to learn the anatomical structure of the coronary artery, which considers the local spatial correlations in the input data as the convolutions are used for input-to-state as well as state- to-state transitions, thus more suitable for image analysis.
85
Multi-task deep learning for medical image computing and analysis: A review
TL;DR: Multi-task deep learning (MTDL) as discussed by the authors is a joint learning paradigm that harnesses the inherent correlation of multiple related tasks to achieve reciprocal benefits in improving performance, enhancing generalizability, and reducing the overall computational cost.
84
References
U-Net: Convolutional Networks for Biomedical Image Segmentation
Olaf Ronneberger,Philipp Fischer,Thomas Brox +2 more
- 05 Oct 2015
TL;DR: Neber et al. as discussed by the authors proposed a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently, which can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks.
Progressive Attention Guided Recurrent Network for Salient Object Detection
Xiaoning Zhang,Tiantian Wang,Jinqing Qi,Huchuan Lu,Gang Wang +4 more
- 18 Jun 2018
TL;DR: A novel attention guided network which selectively integrates multi-level contextual information in a progressive manner and introduces multi-path recurrent feedback to enhance this proposed progressive attention driven framework.
Coronary centerline extraction from CT coronary angiography images using a minimum cost path approach
TL;DR: The presented results show that minimum cost path approaches can effectively be applied as a preprocessing step for subsequent analysis in clinical practice and biomedical research.
109
Robust and accurate coronary artery centerline extraction in CTA by combining model-driven and data-driven approaches.
Yefeng Zheng,Huseyin Tek,Gareth Funka-Lea +2 more
- 22 Sep 2013
TL;DR: The automatically segmented chambers are exploited to predict the initial position of the major coronary centerlines and define a vessel-specific region-of-interest (ROI) to constrain the following centerline refinement.
A robust and efficient curve skeletonization algorithm for tree-like objects using minimum cost paths
TL;DR: A new robust and efficient curve skeletonization algorithm for three-dimensional (3-D) elongated fuzzy objects using a minimum cost path approach, which avoids spurious branches without requiring post-pruning.
74