Book Chapter10.1007/978-3-031-17027-0_2
DeepEdit: Deep Editable Learning for Interactive Segmentation of 3D Medical Images
Andres Diaz-Pinto,Pritesh Mehta,Sachidanand Alle,Muhammad Hamza Asad,Richard Brown,Vishwesh Nath,Alvin Ihsani,Michela Antonelli,Daniel Palkovics,Csaba Pinter,Ron N. Alkalay,Steve Pieper,Holger R. Roth,Daguang Xu,Prerna Dogra,Tom Vercauteren,Andrew Feng,Abood Quraini,Sebastien Ourselin,M. Jorge Cardoso +19 more
- 18 May 2023
Vol. abs/2305.10655, pp 11-21
TL;DR: DeepEdit as mentioned in this paper combines the power of two methods: a non-interactive (i.e. automatic segmentation using nnU-Net, UNET or UNETR) and an interactive segmentation method (e.g. DeepGrow), into a single deep learning model.
read more
Abstract: Automatic segmentation of medical images is a key step for diagnostic and interventional tasks. However, achieving this requires large amounts of annotated volumes, which can be tedious and time-consuming task for expert annotators. In this paper, we introduce DeepEdit, a deep learning-based method for volumetric medical image annotation, that allows automatic and semi-automatic segmentation, and click-based refinement. DeepEdit combines the power of two methods: a non-interactive (i.e. automatic segmentation using nnU-Net, UNET or UNETR) and an interactive segmentation method (i.e. DeepGrow), into a single deep learning model. It allows easy integration of uncertainty-based ranking strategies (i.e. aleatoric and epistemic uncertainty computation) and active learning. We propose and implement a method for training DeepEdit by using standard training combined with user interaction simulation. Once trained, DeepEdit allows clinicians to quickly segment their datasets by using the algorithm in auto segmentation mode or by providing clicks via a user interface (i.e. 3D Slicer, OHIF). We show the value of DeepEdit through evaluation on the PROSTATEx dataset for prostate/prostatic lesions and the Multi-Atlas Labeling Beyond the Cranial Vault (BTCV) dataset for abdominal CT segmentation, using state-of-the-art network architectures as baseline for comparison. DeepEdit could reduce the time and effort annotating 3D medical images compared to DeepGrow alone. Source code is available at https://github.com/Project-MONAI/MONAILabel .
read more
Chat with Paper
AI Agents for this Paper
Find similar papers on Google Scholar, PubMed and Arxiv
Write a critical review of this paper
Analyze citations of this paper to find unaddressed research gaps
Figures
Table 1. Whole prostate segmentation mean Dice scores ± one standard deviation for the 193 PROSTATEx dataset patients used in the ten-fold cross-validation, for nnUNet, DeepEdit-0 (equivalent to DeepGrow), DeepEdit-0.25, and DeepEdit-0.5. The highest mean Dice in each column is shown in bold. 
Table 3. Dice scores for single and multilabel segmentation on the validation set using the BTCV dataset. For single label we used the spleen organ and multilabel we used spleen, liver, and left and right kidneys. We show the results obtained for 0, 1, 5, and 10 clicks simulated during validation. Highest Dice scores in each column are shown in bold. 
Table 2. Prostatic lesion segmentation mean Dice scores ± one standard deviation for the 200 PROSTATEx dataset patients used in the ten-fold cross-validation, for nnUNet, DeepEdit-0 (equivalent to DeepGrow), DeepEdit-0.25, and DeepEdit-0.5. The highest mean Dice in each column is shown in bold. 
Fig. 3. Prostatic lesion segmentation: Dice score box plots for the 200 PROSTATEx dataset patients used in the ten-fold cross-validation, for DeepEdit-0 (equivalent to DeepGrow), DeepEdit-0.25, and DeepEdit-0.5. 
Fig. 1. General schema of the DeepEdit: (a) Training and (b) Inference Mode. DeepEdit training process consists of two modes: the automatic segmentation mode and the interactive mode. Simulated clicks for all labels plus background are added to a backbone network as input channels . Input tensor could be either the image with zero-tensors (automatic segmentation mode) or the image with tensors representing label clicks and background clicks provided by the user (interactive mode). 
Fig. 2. Whole prostate segmentation: Dice score box plots for the 193 PROSTATEx dataset patients used in the ten-fold cross-validation, for DeepEdit-0 (equivalent to DeepGrow), DeepEdit-0.25, and DeepEdit-0.5.
Citations
MONAI Label: A framework for AI-assisted Interactive Labeling of 3D Medical Images
Andres Diaz-Pinto,Sachidanand Alle,Alvin Ihsani,Muhammad Hamza Asad,Vishwesh Nath,Fernando Pérez-García,Pritesh Mehta,Wenqi Li,Holger R. Roth,Tom Vercauteren,Daguang Xu,Prerna Dogra,Sebastien Ourselin,Andrew Feng,M. Jorge Cardoso +14 more
TL;DR: MONAI Label as mentioned in this paper is a free and open-source framework that facilitates the development of applications based on artificial intelligence (AI) models that aim at reducing the time required to annotate radiology datasets.
Increasing the impact of vertebrate scientific collections through 3D imaging: The openVertebrate (oVert) Thematic Collections Network
DC Blackburn,Doug M Boyer,Jaimi A Gray,Julie Winchester,John M. Bates,Stephanie L Baumgart,Emily Braker,Daryl Coldren,Kevin W Conway,Alison R. Davis Rabosky,Noé de la Sancha,Casey B Dillman,Jonathan L. Dunnum,Catherine M Early,Benjamin W. Frable,Matt W Gage,James Hanken,Jessica A. Maisano,Ben D Marks,Katherine P. Maslenikov,John E. McCormack,Ramon S Nagesan,Gregory G. Pandelis,H. L. Prestridge,Daniel L. Rabosky,Zachary S Randall,Mark B Robbins,Lauren A Scheinberg,Carol L Spencer,Adam P Summers,Leif Tapanila,Cody W Thompson,Luke Tornabene,Greg J Watkins-Colwell,Luke J Welton,Edward L. Stanley +35 more
TL;DR: High-fidelity 3D imaging of museum specimens increases their impact by making them accessible to a broad audience.
25
MONAI Label: A framework for AI-assisted Interactive Labeling of 3D Medical Images
23 Mar 2022
TL;DR: MONAI Label as discussed by the authors is a free and open-source framework that facilitates the development of applications based on artificial intelligence (AI) models that aim at reducing the time required to annotate radiology datasets.
Artificial intelligence CAD tools in trauma imaging: a scoping review from the American Society of Emergency Radiology (ASER) AI/ML Expert Panel
David Dreizin,Pedro V. Staziaki,Garvit D. Khatri,Nicholas M. Beckmann,Zhaoyong Feng,Yuanyuan Liang,Zachary S. DelProposto,Maximiliano Klug,J. S. Spann,Nathan Sarkar,Yunting Fu +10 more
TL;DR: Trauma imaging CAD tools are likely to improve patient care but are currently in an early stage of maturity, with few FDA-approved products for a limited number of uses and the scarcity of high-quality annotated data remains a major barrier.
15
SAMMed: A medical image annotation framework based on large vision model
Chenglong Wang,Dexuan Li,Sucheng Wang,Chengxiu Zhang,Yida Wang,Yun Liu,Guang Yang +6 more
TL;DR: An enhanced framework for medical image annotation that leverages the capabilities of large vision model, Segment Anything Model, and demonstrates promising results in medical image annotation.
References
•Proceedings Article
Attention is All you Need
Ashish Vaswani,Noam Shazeer,Niki Parmar,Jakob Uszkoreit,Llion Jones,Aidan N. Gomez,Lukasz Kaiser,Illia Polosukhin +7 more
- 12 Jun 2017
TL;DR: This paper proposed a simple network architecture based solely on an attention mechanism, dispensing with recurrence and convolutions entirely and achieved state-of-the-art performance on English-to-French translation.
U-Net: Convolutional Networks for Biomedical Image Segmentation
Olaf Ronneberger,Philipp Fischer,Thomas Brox +2 more
- 05 Oct 2015
TL;DR: Neber et al. as discussed by the authors proposed a network and training strategy that relies on the strong use of data augmentation to use the available annotated samples more efficiently, which can be trained end-to-end from very few images and outperforms the prior best method (a sliding-window convolutional network) on the ISBI challenge for segmentation of neuronal structures in electron microscopic stacks.
Snakes : Active Contour Models
TL;DR: This work uses snakes for interactive interpretation, in which user-imposed constraint forces guide the snake near features of interest, and uses scale-space continuation to enlarge the capture region surrounding a feature.
Normalized cuts and image segmentation
Jianbo Shi,Jitendra Malik +1 more
TL;DR: This work treats image segmentation as a graph partitioning problem and proposes a novel global criterion, the normalized cut, for segmenting the graph, which measures both the total dissimilarity between the different groups as well as the total similarity within the groups.
Normalized cuts and image segmentation
Jianbo Shi,Jitendra Malik +1 more
- 17 Jun 1997
TL;DR: This work treats image segmentation as a graph partitioning problem and proposes a novel global criterion, the normalized cut, for segmenting the graph, which measures both the total dissimilarity between the different groups as well as the total similarity within the groups.