Optimized High Resolution 3D Dense-U-Net Network for Brain and Spine Segmentation

Kolarik, Martin; Bürget, Radim; Uher, Václav; Říha, Kamil; Dutta, Malay Kishore

doi:10.3390/app9030404

Cited by 88 publications

(62 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Using only a small training dataset (160 LDCT scans), our method achieved a mean Dice coefficient of 86.6% and labeling accuracy of 97.5% for VB segmentation and labeling. In terms of the absolute metric, our method performed worse in VB segmentation than that used in a previous study [24][25][26] but yet surpassed previously reported results [13,27] in VB labeling. Nevertheless, since different datasets were used in assessment of different methods, direct comparison of different methods should be interpreted cautiously.…”

Section: Discussioncontrasting

confidence: 55%

Automatic opportunistic osteoporosis screening using low-dose chest computed tomography scans obtained for lung cancer screening

Pan

Shi²,

Wang

et al. 2020

Eur Radiol

View full text Add to dashboard Cite

Objective Osteoporosis is a prevalent and treatable condition, but it remains underdiagnosed. In this study, a deep learning-based system was developed to automatically measure bone mineral density (BMD) for opportunistic osteoporosis screening using lowdose chest computed tomography (LDCT) scans obtained for lung cancer screening. Methods First, a deep learning model was trained and tested with 200 annotated LDCT scans to segment and label all vertebral bodies (VBs). Then, the mean CT numbers of the trabecular area of target VBs were obtained based on the segmentation mask through geometric operations. Finally, a linear function was built to map the trabecular CT numbers of target VBs to their BMDs collected from approved software used for osteoporosis diagnosis. The diagnostic performance of the developed system was evaluated using an independent dataset of 374 LDCT scans with standard BMDs and osteoporosis diagnosis. Results Our deep learning model achieved a mean Dice coefficient of 86.6% for VB segmentation and 97.5% accuracy for VB labeling. Line regression and Bland-Altman analyses showed good agreement between the predicted BMD and the ground truth, with correlation coefficients of 0.964-0.968 and mean errors of 2.2-4.0 mg/cm 3. The area under the curve (AUC) was 0.927 for detecting osteoporosis and 0.942 for distinguishing low BMD. Conclusion The proposed deep learning-based system demonstrated the potential to automatically perform opportunistic osteoporosis screening using LDCT scans obtained for lung cancer screening. Key Points • Osteoporosis is a prevalent but underdiagnosed condition that can increase the risk of fracture. • A deep learning-based system was developed to fully automate bone mineral density measurement in low-dose chest computed tomography scans. • The developed system achieved high accuracy for automatic opportunistic osteoporosis screening using low-dose chest computed tomography scans obtained for lung cancer screening. Keywords Bone mineral density. Deep learning. Osteoporosis. Screening Abbreviations AUC Area under the curve BMD Bone mineral density CNN Convolutional neural network CT Computed tomography DL Deep learning DXA Dual-energy X-ray absorptiometry LDCT Low-dose chest computed tomography QA Quality assurance QCT Quantitative computed tomography VB Vertebral body VOI Volume of interest Yaling Pan and Dejun Shi contributed equally to this work.

show abstract

Section: Discussioncontrasting

confidence: 55%

Automatic opportunistic osteoporosis screening using low-dose chest computed tomography scans obtained for lung cancer screening

Pan

Shi²,

Wang

et al. 2020

Eur Radiol

View full text Add to dashboard Cite

show abstract

“…Furthermore, integration of Attention Gates (AGs) [64] and Squeeze-and-Excitation (SE) blocks [65,66] are shown to increase the performance of the U-Net model in performing segmentation [64,66] and could be considered as another component for optimisation in the U-Net structure as performed in this paper. In addition, optimisation of the components for 3D neural network models would be a logical and next step to further the work in this paper as it processes 3D input and extracts 3D features, which can be beneficial for performing volumetric medical image segmentation [34].…”

Section: Discussionmentioning

confidence: 99%

“…In general, the most successful models are based on 3D neural networks, which include the fusion of 2D/3D models. There are many 3D networks such as, but not limited to, Hybrid Discriminative Network (HD-Net) [32], V-Net [33] and 3D Dense U-Net [34]. In this paper we are considering only 2D networks as the number of components to optimise over is significantly less than in the 3D case.…”

Section: Introductionmentioning

confidence: 99%

Optimisation of 2D U-Net Model Components for Automatic Prostate Segmentation on MRI

et al. 2020

View full text Add to dashboard Cite

In this paper, we develop an optimised state-of-the-art 2D U-Net model by studying the effects of the individual deep learning model components in performing prostate segmentation. We found that for upsampling, the combination of interpolation and convolution is better than the use of transposed convolution. For combining feature maps in each convolution block, it is only beneficial if a skip connection with concatenation is used. With respect to pooling, average pooling is better than strided-convolution, max, RMS or L2 pooling. Introducing a batch normalisation layer before the activation layer gives further performance improvement. The optimisation is based on a private dataset as it has a fixed 2D resolution and voxel size for every image which mitigates the need of a resizing operation in the data preparation process. Non-enhancing data preprocessing was applied and five-fold cross-validation was used to evaluate the fully automatic segmentation approach. We show it outperforms the traditional methods that were previously applied on the private dataset, as well as outperforming other comparable state-of-the-art 2D models on the public dataset PROMISE12.

show abstract

“…The selection of a deep learning or convolutional neural network model is the most important step in a medical image segmentation pipeline. There is a variety of model architectures and each has different strengths and weaknesses [7,8,10,11,[23][24][25][26][27][28][29].…”

Section: Model Architecturementioning

confidence: 99%

MIScnn: A Framework for Medical Image Segmentation with Convolutional Neural Networks and Deep Learning

Müller

2019

Submissions to the 2019 Kidney Tumor Segmentation Challenge: KiTS19

View full text Add to dashboard Cite

The increased availability and usage of modern medical imaging induced a strong need for automatic medical image segmentation. Still, current image segmentation platforms do not provide the required functionalities for plain setup of medical image segmentation pipelines. Already implemented pipelines are commonly standalone software, optimized on a specific public data set. Therefore, this paper introduces the open-source Python library MIScnn. The aim of MIScnn is to provide an intuitive API allowing fast building of medical image segmentation pipelines including data I/O, preprocessing, data augmentation, patch-wise analysis, metrics, a library with state-of-the-art deep learning models and model utilization like training, prediction, as well as fully automatic evaluation (e.g. cross-validation). Similarly, high configurability and multiple open interfaces allow full pipeline customization. Running a crossvalidation with MIScnn on the Kidney Tumor Segmentation Challenge 2019 data set (multi-class semantic segmentation with 300 CT scans) resulted into a powerful predictor based on the standard 3D U-Net model. With this experiment, we could show that the MIScnn framework enables researchers to rapidly set up a complete medical image segmentation pipeline by using just a few lines of code. The source code for MIScnn is available in the Git repository: https://github.com/frankkramer-lab/MIScnn.

show abstract

Optimized High Resolution 3D Dense-U-Net Network for Brain and Spine Segmentation

Cited by 88 publications

References 21 publications

Automatic opportunistic osteoporosis screening using low-dose chest computed tomography scans obtained for lung cancer screening

Automatic opportunistic osteoporosis screening using low-dose chest computed tomography scans obtained for lung cancer screening

Optimisation of 2D U-Net Model Components for Automatic Prostate Segmentation on MRI

MIScnn: A Framework for Medical Image Segmentation with Convolutional Neural Networks and Deep Learning

Contact Info

Product

Resources

About