Augmenting vascular disease diagnosis by vasculature-aware unsupervised learning

Wang, Yong; Ji, Mengqi; Jiang, Shengwei; Wang, Xukang; Wu, Jiamin; Duan, Feng; Fan, Jian; Huang, Laiqiang; Ma, Shaohua; Fang, Lü; Dai, Qionghai

doi:10.1038/s42256-020-0188-z

Cited by 15 publications

(8 citation statements)

References 54 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Frameworks that can optimize DNNs without strict reliance on large and high-quality annotated training datasets can significantly narrow the gap between research and clinical Recent deep-learning works have presented encouraging performance in handling certain types of imperfect datasets in isolation [41][42][43][44][45] but have not yet shown general applicability by addressing all three types. Methods such as data augmentation 41,46 , transfer learning 42,43 , semi-supervised learning 44,47 , and selfsupervised learning 48,49 have been extensively investigated to handle cases with limited training annotations or no target domain annotations. By contrast, much less attention has been given to noisy label learning in medical imaging 16,50 .…”

Section: Discussionmentioning

confidence: 99%

Annotation-efficient deep learning for automatic medical image segmentation

et al. 2021

View full text Add to dashboard Cite

Automatic medical image segmentation plays a critical role in scientific research and medical care. Existing high-performance deep learning methods typically rely on large training datasets with high-quality manual annotations, which are difficult to obtain in many clinical applications. Here, we introduce Annotation-effIcient Deep lEarning (AIDE), an open-source framework to handle imperfect training datasets. Methodological analyses and empirical evaluations are conducted, and we demonstrate that AIDE surpasses conventional fully-supervised models by presenting better performance on open datasets possessing scarce or noisy annotations. We further test AIDE in a real-life case study for breast tumor segmentation. Three datasets containing 11,852 breast images from three medical centers are employed, and AIDE, utilizing 10% training annotations, consistently produces segmentation maps comparable to those generated by fully-supervised counterparts or provided by independent radiologists. The 10-fold enhanced efficiency in utilizing expert labels has the potential to promote a wide range of biomedical applications.

show abstract

Section: Discussionmentioning

confidence: 99%

Annotation-efficient deep learning for automatic medical image segmentation

et al. 2021

View full text Add to dashboard Cite

show abstract

“…We have shown that two possible ways to obtain such auxiliary masks: using a parametric shape model to generate a set of auxiliary masks for simple structures such as optic disc and fetal head, and taking advantage of masks of the object from another domain (e.g., public datasets) for complex structures such as the lung and the liver. For more complex structures such as the brain and vessels [71], it might be more challenging to leverage existing unpaired labels from a different dataset for shape constraint. The effectiveness of our method in such cases will be investigated in the future.…”

Section: Discussionmentioning

confidence: 99%

Annotation-Efficient Learning for Medical Image Segmentation Based on Noisy Pseudo Labels and Adversarial Learning

Wang

Guo

Wang

et al. 2021

IEEE Trans. Med. Imaging

View full text Add to dashboard Cite

Despite that deep learning has achieved state-of-theart performance for medical image segmentation, its success relies on a large set of manually annotated images for training that are expensive to acquire. In this paper, we propose an annotationefficient learning framework for segmentation tasks that avoids annotations of training images, where we use an improved Cycle-Consistent Generative Adversarial Network (GAN) to learn from a set of unpaired medical images and auxiliary masks obtained either from a shape model or public datasets. We first use the GAN to generate pseudo labels for our training images under the implicit high-level shape constraint represented by a Variational Auto-encoder (VAE)-based discriminator with the help of the auxiliary masks, and build a Discriminator-guided Generator Channel Calibration (DGCC) module which employs our discriminator's feedback to calibrate the generator for better pseudo labels. To learn from the pseudo labels that are noisy, we further introduce a noise-robust iterative learning method using noise-weighted Dice loss. We validated our framework with two situations: objects with a simple shape model like optic disc in fundus images and fetal head in ultrasound images, and complex structures like lung in X-Ray images and liver in CT images. Experimental results demonstrated that 1) Our VAEbased discriminator and DGCC module help to obtain highquality pseudo labels. 2) Our proposed noise-robust learning method can effectively overcome the effect of noisy pseudo labels.3) The segmentation performance of our method without using annotations of training images is close or even comparable to that of learning from human annotations.

show abstract

“…Deep learning has revolutionized fields such as computer vision and natural language processing. 19 In the chemical and biological sciences, artificial intelligence is reshaping research methodologies for applications such as chemical exploration, 20,21 prediction of chemical reactivity 22 and reaction performance, 23 prediction and design of synthetic 24,25 and retrosynthetic routes, 26 automated reaction optimization, 27 accelerated materials discovery, [28][29][30] materials characterization, 31 label-free cell classification, 32,33 detection of DNA modifications, 34 study of oncogenic differentiation, 35 prediction of protein structure, 36 and diagnosis of medical images, 37,38 providing scientists with powerful tools for feature extraction, classification, and prediction.…”

Section: Progress and Potentialmentioning

confidence: 99%