Entropy Guided Unsupervised Domain Adaptation for Cross-Center Hip Cartilage Segmentation from MRI

Zeng, Guodong; Schmaranzer, Florian; Lerch, Till; Boschung, Adam; Zheng, Guoyan; Burger, Jürgen; Gerber, Kate; Tannast, Moritz; Siebenrock, Klaus-Arno; Kim, Young-Jo; Novais, Eduardo N.; Gerber, Nicolas U.

doi:10.1007/978-3-030-59710-8_44

Cited by 11 publications

(6 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Differences in system calibration and protocols across centres can also reduce the segmentation accuracy. Domain adaption techniques are being developed to allow the optimization of the segmentation network for a specific centre without the need for time consuming manual MRI annotations [35].…”

Section: Discussionmentioning

confidence: 99%

Automated quantification of cartilage quality for hip treatment decision support

Ruckli

Schmaranzer²,

Meier³

et al. 2022

Int J CARS

Self Cite

View full text Add to dashboard Cite

Purpose Preservation surgery can halt the progress of joint degradation, preserving the life of the hip; however, outcome depends on the existing cartilage quality. Biochemical analysis of the hip cartilage utilizing MRI sequences such as delayed gadolinium-enhanced MRI of cartilage (dGEMRIC), in addition to morphological analysis, can be used to detect early signs of cartilage degradation. However, a complete, accurate 3D analysis of the cartilage regions and layers is currently not possible due to a lack of diagnostic tools. Methods A system for the efficient automatic parametrization of the 3D hip cartilage was developed. 2D U-nets were trained on manually annotated dual-flip angle (DFA) dGEMRIC for femoral head localization and cartilage segmentation. A fully automated cartilage sectioning pipeline for analysis of central and peripheral regions, femoral-acetabular layers, and a variable number of section slices, was developed along with functionality for the automatic calculation of dGEMRIC index, thickness, surface area, and volume. Results The trained networks locate the femoral head and segment the cartilage with a Dice similarity coefficient of 88 ± 3 and 83 ± 4% on DFA and magnetization-prepared 2 rapid gradient-echo (MP2RAGE) dGEMRIC, respectively. A completely automatic cartilage analysis was performed in 18s, and no significant difference for average dGEMRIC index, volume, surface area, and thickness calculated on manual and automatic segmentation was observed. Conclusion An application for the 3D analysis of hip cartilage was developed for the automated detection of subtle morphological and biochemical signs of cartilage degradation in prognostic studies and clinical diagnosis. The segmentation network achieved a 4-time increase in processing speed without loss of segmentation accuracy on both normal and deformed anatomy, enabling accurate parametrization. Retraining of the networks with the promising MP2RAGE protocol would enable analysis without the need for B1 inhomogeneity correction in the future.

show abstract

Section: Discussionmentioning

confidence: 99%

Automated quantification of cartilage quality for hip treatment decision support

Ruckli

Schmaranzer²,

Meier³

et al. 2022

Int J CARS

Self Cite

View full text Add to dashboard Cite

show abstract

“…Let SAE(y T 1 i , y T2•RN i ; θ) be the Shape AutoEncoder, thus refined outputs Ŷ are defined via Eq. (7).…”

Section: Selfmentioning

confidence: 99%

“…Many efforts have been devoted to addressing the domain shift problem. Among them, the most widely used method is domain adaptation to align the latent feature distributions of the two domains [6,7,8]. Unfortunately, one limitation is that it requires concurrent access to the input images of both the source and target domains.…”

Section: Introductionmentioning

confidence: 99%

Plug-and-play Shape Refinement Framework for Multi-site and Lifespan Brain Skull Stripping

Li¹,

Dan²,

Wang³

et al. 2022

Preprint

View full text Add to dashboard Cite

Skull stripping is a crucial prerequisite step in the analysis of brain magnetic resonance (MR) images. Although many excellent works or tools have been proposed, they suffer from low generalization capability. For instance, the model trained on a dataset with specific imaging parameters (source domain) cannot be well applied to other datasets with different imaging parameters (target domain). Especially, for the lifespan datasets, the model trained on an adult dataset is not applicable to an infant dataset due to the large domain difference. To address this issue, numerous domain adaptation (DA) methods have been proposed to align the extracted features between the source and target domains, requiring concurrent access to the input images of both domains. Unfortunately, it is problematic to share the images due to privacy. In this paper, we design a source-free domain adaptation framework (SDAF) for multi-site and lifespan skull stripping that can accomplish domain adaptation without access to source domain images. Our method only needs to share the source labels as shape dictionaries and the weights trained on the source data, without disclosing private information from source domain subjects. To deal with the domain shift between multi-site lifespan datasets, we take advantage of the brain shape prior which is invariant to imaging parameters and ages. Experiments demonstrate that our framework can significantly outperform the state-of-the-art methods on multi-site lifespan datasets.

show abstract

“…Inspired by the fact that the segmentation outputs of images from two domains should have considerable similarities, e.g., spatial layout and local context, many recent methods [9], [11], [29] tended to perform structure adaptation between the two domains in the output level. Working along this line, [11] proposed an entropy-based adversarial learning to penalize low-confident predictions on target domain.…”

Section: A Unsupervised Domain Adaptationmentioning

confidence: 99%

Margin Preserving Self-paced Contrastive Learning Towards Domain Adaptation for Medical Image Segmentation

Liu,

Zhu,

Zheng

et al. 2021

Preprint

View full text Add to dashboard Cite

To bridge the gap between the source and target domains in unsupervised domain adaptation (UDA), the most common strategy puts focus on matching the marginal distributions in the feature space through adversarial learning. However, such category-agnostic global alignment lacks of exploiting the class-level joint distributions, causing the aligned distribution less discriminative. To address this issue, we propose in this paper a novel margin preserving self-paced contrastive Learning (MP-SCL) model for cross-modal medical image segmentation. Unlike the conventional construction of contrastive pairs in contrastive learning, the domain-adaptive category prototypes are utilized to constitute the positive and negative sample pairs. With the guidance of progressively refined semantic prototypes, a novel margin preserving contrastive loss is proposed to boost the discriminability of embedded representation space. To enhance the supervision for contrastive learning, more informative pseudo-labels are generated in target domain in a self-paced way, thus benefiting the category-aware distribution alignment for UDA. Furthermore, the domain-invariant representations are learned through joint contrastive learning between the two domains. Extensive experiments on cross-modal cardiac segmentation tasks demonstrate that MPSCL significantly improves semantic segmentation performance, and outperforms a wide variety of state-of-the-art methods by a large margin.

show abstract

Entropy Guided Unsupervised Domain Adaptation for Cross-Center Hip Cartilage Segmentation from MRI

Cited by 11 publications

References 12 publications

Automated quantification of cartilage quality for hip treatment decision support

Automated quantification of cartilage quality for hip treatment decision support

Plug-and-play Shape Refinement Framework for Multi-site and Lifespan Brain Skull Stripping

Margin Preserving Self-paced Contrastive Learning Towards Domain Adaptation for Medical Image Segmentation

Contact Info

Product

Resources

About