Towards Whole-body CT Bone Segmentation

Klein, André; Warszawski, Jan; Hillengaß, Jens; Maier‐Hein, Klaus H.

doi:10.1007/978-3-662-56537-7_59

Cited by 9 publications

(3 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the musculoskeletal field, promising results have been shown using CNNs to segment vertebrae, whole body, and the proximal femurs in both clinical MR and CT scans [15][16][17][18][19] . However, previous studies have created CNNs that perform well on one dataset [15][16][17][18][19] .…”

Section: Introductionmentioning

confidence: 99%

“…CNNs have been successfully used for classification, object detection, and segmentation of medical image data and have been shown to consistently outperform traditional approaches 14 . In the musculoskeletal field, promising results have been shown using CNNs to segment vertebrae, whole body, and the proximal femurs in both clinical MR and CT scans [15][16][17][18][19] . However, previous studies have created CNNs that perform well on one dataset [15][16][17][18][19] .…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Deep learning-based segmentation of high-resolution computed tomography image data outperforms commonly used automatic bone segmentation methods

Patton

Henning

Goulet

et al. 2021

Preprint

View full text Add to dashboard Cite

Segmenting bone from background is required to quantify bone architecture in computed tomography (CT) image data. A deep learning approach using convolutional neural networks (CNN) is a promising alternative method for automatic segmentation. The study objectives were to evaluate the performance of CNNs in automatic segmentation of human vertebral body (micro-CT) and femoral neck (nano-CT) data and to investigate the performance of CNNs to segment data across scanners. Scans of human L1 vertebral bodies (microCT [North Star Imaging], n=28, 53μm3) and femoral necks (nano-CT [GE], n=28, 27μm3) were used for evaluation. Six slices were selected for each scan and then manually segmented to create ground truth masks (Dragonfly 4.0, ORS). Two-dimensional U-Net CNNs were trained in Dragonfly 4.0 with images of the [FN] femoral necks only, [VB] vertebral bodies only, and [F+V] combined CT data. Global (i.e., Otsu and Yen) and local (i.e., Otsu r = 100) thresholding methods were applied to each dataset. Segmentation performance was evaluated using the Dice index, a similarity metric of overlap. Kruskal-Wallis and Tukey-Kramer post-hoc tests were used to test for significant differences in the accuracy of segmentation methods. The FN U-Net had significantly higher Dice indices (i.e., better performance) than the global (Otsu: p=0.001; Yen: p=0.001) and local (Otsu [r=100]: p=0.001) thresholding methods and the VB U-Net (p=0.001) but there was no significant difference in model performance compared to the FN + VB U-net (p=0.783) on femoral neck image data. The VB U-net had significantly higher Dice coefficients than the global and local Otsu (p=0.001 for both) and FN U-Net (p=0.001) but not compared to the Yen (p=0.462) threshold or FN + VB U-net (p=0.783) on vertebral body image data. The results demonstrate that the U-net architecture outperforms common thresholding methods. Further, a network trained with bone data from a different system (i.e., different image acquisition parameters and voxel size) and a different anatomical site can perform well on unseen data. Finally, a network trained with combined datasets performed well on both datasets, indicating that a network can feasibly be trained with multiple datasets and perform well on varied image data.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Deep learning-based segmentation of high-resolution computed tomography image data outperforms commonly used automatic bone segmentation methods

Patton

Henning

Goulet

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…The major bottleneck in using deep learning for medical image segmentation is the need for large dataset of manually contoured slices for training and validation. The bone segmentation study used a simpler thresholding based algorithm to generate the bone segments which were then corrected manually by an expert (Klein et al , 2018). …”

Section: Introductionmentioning

confidence: 99%

Fully automated tissue classifier for contrast-enhanced CT scans of adult and pediatric patients

et al. 2018

View full text Add to dashboard Cite

Purpose: To develop a consistent, fully-automated classifier for all tissues within the trunk and to more accurately discriminate between tissues (such as bone) and contrast medium with overlapping high CT numbers Methods: Twenty-eight contrast enhanced NCAP (neck-chest-abdomen-pelvis) CT scans (4 adult and 3 pediatric patients) were used to train and test a tissue classification pipeline. The classifier output consisted of 6 tissue classes: lung, fat, muscle, solid organ, blood/bowel contrast and bone. The input features for training were selected from 28 2D image filters and 12 3D filters, and one hand crafted spatial feature. To improve differentiation between tissue and blood/bowel contrast classification, 70 additional CT images were manually classified. Two different training data sets consisting of manually classified tissues from different locations in body were used to train the models. Training used the random forest algorithm in WEKA (Waikato Environment for Knowledge Analysis); the number of trees was optimized for best out-of-bag error. Automated classification accuracy was compared with manual classification by calculating Dice Similarity Coefficient (DSC). Results: Model performance was tested on 21 manually classified slices (2 adult and 1 pediatric patient). The overall DSC at image locations represented in the training dataset were - lung: 0.98, fat: 0.90, muscle: 0.85, solid organ: 0.75, blood/contrast: 0.82, and bone: 0.90. The overall DSC for slice locations that were not represented in the training dataset were - lung: 0.97, fat: 0.89, muscle: 0.76, solid organ: 0.79, blood: 0.56, and bone: 0.74. Analyzing the classification maps for the entire scan volume revealed that except for misclassifications in the trabecular bone region of the spinal column, and solid organ and blood/contrast interfaces within the abdomen, the results were acceptable. Conclusions: A fully-automated whole-body tissue classifier for adult and pediatric contrast-enhanced CT using random forest algorithm and intensity-based image filters was developed.

show abstract

Development of Artificial Intelligence Applied to the Production of Biopolymers with a Focus on Sustainability

de Oliveira Marcelo Júnior,

Gomes,

Guerra

et al. 2023

Advances in Sustainability Science and Technology

View full text Add to dashboard Cite

Towards Whole-body CT Bone Segmentation

Cited by 9 publications

References 11 publications

Deep learning-based segmentation of high-resolution computed tomography image data outperforms commonly used automatic bone segmentation methods

Deep learning-based segmentation of high-resolution computed tomography image data outperforms commonly used automatic bone segmentation methods

Fully automated tissue classifier for contrast-enhanced CT scans of adult and pediatric patients

Development of Artificial Intelligence Applied to the Production of Biopolymers with a Focus on Sustainability

Contact Info

Product

Resources

About