Automatic Multi-Organ Segmentation on Abdominal CT With Dense V-Networks

Gibson, Eli; Giganti, Francesco; Hu, Yipeng; Bonmati, Ester; Bandula, Steve; Gurusamy, Kurinchi Selvan; Davidson, Brian R; Pereira, Stephen P.; Clarkson, Matthew J.; Barratt, Dean C.

doi:10.1109/tmi.2018.2806309

Cited by 524 publications

(434 citation statements)

References 42 publications

Supporting

Mentioning

420

Contrasting

Unclassified

Order By: Relevance

“…A residual learning mechanism [36] was used to classify each voxel in the VoxResNet [27]. More recently, training through a shapeprior method was proposed in a densely connected V-net-like structure [35].…”

Section: Cnns In Medical Image Segmentationmentioning

confidence: 99%

See 1 more Smart Citation

Pose-aware instance segmentation framework from cone beam CT images for tooth segmentation

Chung

Lee

Hong

et al. 2020

Computers in Biology and Medicine

View full text Add to dashboard Cite

Individual tooth segmentation from cone beam computed tomography (CBCT) images is an essential prerequisite for an anatomical understanding of orthodontic structures in several applications, such as tooth reformation planning and implant guide simulations. However, the presence of severe metal artifacts in CBCT images hinders the accurate segmentation of each individual tooth. In this study, we propose a neural network for pixel-wise labeling to exploit an instance segmentation framework that is robust to metal artifacts. Our method comprises of three steps: 1) image cropping and realignment by pose regressions, 2) metal-robust individual tooth detection, and 3) segmentation. We first extract the alignment information of the patient by pose regression neural networks to attain a volume-of-interest (VOI) region and realign the input image, which reduces the interoverlapping area between tooth bounding boxes. Then, individual tooth regions are localized within a VOI realigned image using a convolutional detector. We improved the accuracy of the detector by employing non-maximum suppression and multiclass classification metrics in the region proposal network. Finally, we apply a convolutional neural network (CNN) to perform individual tooth segmentation by converting the pixel-wise labeling task to a distance regression task. Metal-intensive image augmentation is also employed for a robust segmentation of metal artifacts. The result shows that our proposed method outperforms other state-of-the-art methods, especially for teeth with metal artifacts. Our method demonstrated 5.68% and 30.30% better accuracy in the F1 score and aggregated Jaccard index, respectively, when compared to the best performing state-of-the-art algorithms. The primary significance of the proposed method is two-fold: 1) an introduction of pose-aware VOI realignment followed by a robust tooth detection and 2) a metal-robust CNN framework for accurate tooth segmentation.Index Terms-Cone beam computed tomography image segmentation, pose-aware tooth detection, pose regression neural network, tooth instance segmentation.

show abstract

Section: Cnns In Medical Image Segmentationmentioning

confidence: 99%

“…9 and 10). The DenseVNet [35] showed relatively higher precision than the 3D U-net [26] and VoxResNet [27] as DenseVNet employed a shape-prior based on trainable parameters [35]. The trained shape-prior based on well-bounded images suppressed the false positive responses.…”

Section: B Segmentation Performancementioning

confidence: 99%

Pose-aware instance segmentation framework from cone beam CT images for tooth segmentation

Chung

Lee

Hong

et al. 2020

Computers in Biology and Medicine

View full text Add to dashboard Cite

show abstract

“…30 used a modified 2D FCN to segment four organs at risks from CT images and apply Conditional Random Fields (CRF) to further improve the segmentation performance. Gibson et al 31…”

Section: Introductionmentioning

confidence: 99%

A method of rapid quantification of patient‐specific organ doses for CT using deep‐learning‐based multi‐organ segmentation and GPU‐accelerated Monte Carlo dose computing

Zhao¹,

Fang²,

Yan³

et al. 2020

Medical Physics

View full text Add to dashboard Cite

Purpose: The ability to obtain patient-specific organ doses for CT will open the door to new applications such as personalized selection of scan factors and individualized risk assessment, leading to the ultimate goal of achieving lowdose and optimized CT imaging. One technical barrier to advancing CT dosimetry has been the lack of computational tools for automatic patient-specific multi-organ segmentation of CT images, coupled with rapid organ dose quantification.This study aims to demonstrate the feasibility of combining deep-learning algorithms for automatic segmentation of radiosensitive organs from CT images and GPU-based Monte Carlo rapid organ dose calculation. Methods: A deep convolutional neural network (CNN) based on the U-Net for organ segmentation is developed and trained to automatically delineate radiosensitive organs from CT images. Two databases are used: the Lung CT Segmentation Challenge 2017 (LCTSC) dataset that contains 60 thoracic CT scan patients each with 5 segmented organs, and the Pancreas-CT (PCT) dataset that contains 43 abdominal CT scan patients each with 8 segmented organs. A five-fold cross-validation of the new method is performed on both sets of data. Dice Similarity Coefficient (DSC) is used to evaluate the segmentation performance against the ground truth. A GPU-based Monte Carlo dose code, ARCHER, is used to calculate patient-specific CT organ doses. The proposed method is tested in terms of Relative Dose Error (RDE). To demonstrate the potential improvement of the new methods, organ dose results are compared against those obtained for population-average phantoms used in an off-line dose reporting software, VirtualDose, at Massachusetts General Hospital. Results: For the group of 60 patients from LCTSC dataset, the median DSCs are found to be 0.97 (right lung), 0.96 (left lung), 0.93 (heart), 0.88 (spinal cord) and 0.78 (esophagus). For the group of 43 patients from PCT dataset, the median DSCs are found to be 0.96 (spleen), 0.96 (liver), 0.95 (left kidney), 0.89 (stomach), 0.87 (gall bladder), 0.79(pancreas), 0.74 (esophagus), and 0.64 (duodenum). Comparing with the organ dose results from population-averaged phantoms, the new patient-specific method achieved the smaller RDE range on all organs: -4.3%~1.5% (vs -31.5%~33.9%) for the lung, -7.0%~2.3% (vs -15.2%~125.1%) for the heart, -18.8%~40.2% (vs -10.3%~124.1%) for the esophagus, -5.6%~1.6% (vs -20.3%~57.4%) for the spleen, -4.5%~4.6% (vs -19.5%~61.0%) for the pancreas, -2.3%~4.4% (vs -37.8%~75.8%) for the left kidney, -14.9%~5.4% (vs -39.9% ~14.6%) for the gallbladder, -0.9%~1.6%(vs -30.1%~72.5%) for the liver, and -23.0%~11.1% (vs -52.5%~-1.3%) for the stomach. The trained automatic segmentation tool takes less than 5 seconds in a patient for all 103 patients in the dataset. The Monte Carlo radiation dose calculations performed in parallel with the segmentation using the GPU-accelerated ARCHER code takes less than 4 seconds in a patient to achieve <0.5% statistical uncertainty in all organ doses for all 103 patients in the...

show abstract

“…However, as mentioned in the introduction, FCN‐based segmentation methods have poor localization around complex object boundary. To improve their performance, a lot of work have been focusing on changing the architectures of the networks . However, it seems that most networks can still make prediction only on the pixel level; there is no explicit mechanism to ensure smoothness of the object boundary.…”

Section: Discussionmentioning

confidence: 99%

Automatic liver segmentation by integrating fully convolutional networks into active contour models

2019

View full text Add to dashboard Cite

Purpose Automatic and accurate three‐dimensional (3D) segmentation of liver with severe diseases from computed tomography (CT) images is a challenging task. Fully convolutional networks (FCNs) have emerged as powerful tools for automatic semantic segmentation, with multiple potential applications in medical imaging. However, the use of a large receptive field and multiple pooling layers in the network leads to poor localization around object boundaries. The network usually makes pixel‐wise prediction independently, making it difficult to respect local label consistency and enforce the smoothness of the object boundary. Methods We have developed an automatic liver segmentation method based on a novel framework that integrates fully convolutional network predictions into active contour models (ACM). We use only a single network architecture to generate a pixel label map containing spatial regional information (foreground and background) as well as layered boundary information. We exploit the structured network outcome to define an external constraint force of active contour models. A unique property of the designed force is that both its strength and direction are adaptive to its position and relative distance to the object boundary. The resulting integrated active contour models have the advantages of incorporating both high‐level and low‐level image information simultaneously, while enforcing the smoothness of the contour. Because the external constraint force can push the evolving contour to the liver boundary and exists everywhere in the image domain, it allows us to place the initial contour far away from the liver boundary. It potentially allows us to control the evolution of the contour in order to preserve the topology of the liver. Results We have trained and evaluated our model on 73 liver CT scans from a clinic study. The integrated ACM model yields mean dice coefficients (DICE) 95.8 ± 1.4 (%). Without further fine‐tuning the network weights for two independent datasets, it yields mean DICE 96.2 ± 0.9 (%) for the SLIVER07 training dataset, and mean DICE 94.3 ± 2.7 (%) for the LiTS training dataset. In comparison with FCN alone model, the integrated ACM model yields improvements in terms of surface distance and DICE values for almost all the cases. Furthermore, the initialization of the active contour can be very far away from the liver boundary. Conclusions Experimental results for segmenting livers (with severe diseases on CT images resulting in shape and density abnormalities) have revealed that our proposed model improves segmentation results in comparison with FCN alone. Without further fine‐tuning the network weights for two independent datasets, the model is capable of handling image variations from different datasets due to its inherent deformable nature. It is relatively easy to integrate more advanced (either existing or future) FCN architecture into our framework to further improve the segmentation performance.

show abstract

Automatic Multi-Organ Segmentation on Abdominal CT With Dense V-Networks

Cited by 524 publications

References 42 publications

Pose-aware instance segmentation framework from cone beam CT images for tooth segmentation

Pose-aware instance segmentation framework from cone beam CT images for tooth segmentation

A method of rapid quantification of patient‐specific organ doses for CT using deep‐learning‐based multi‐organ segmentation and GPU‐accelerated Monte Carlo dose computing

Automatic liver segmentation by integrating fully convolutional networks into active contour models

Contact Info

Product

Resources

About