HVSMR-2.0: Released 2024
HVSMR-2.0 is an expanded dataset for whole-heart segmentation from 3D cardiovascular MR images in patients with congenital heart disease. It contains 60 cardiovascular MR images with even more heart defects, and separate segmentations of eight structures: the four cardiac structures and four great vessels.
The dataset is described in Nature Scientific Data (see pdf) and available on figshare.
Overview
HVSMR 2016 will be held in the afternoon on October 17th, 2016 in conjunction with the Medical Image Computing and Computer Assisted Intervention (MICCAI) conference in Athens, Greece.
Segmenting the blood pool and myocardium from a 3D cardiovascular magnetic resonance (CMR) image is a prerequisite before creating patient-specific heart models for pre-procedural planning of children with complex congenital heart disease (CHD). Manual segmentation is the most reliable method but performing this on many slices of the 3D CMR image is very labor-intensive and subject to inter- and intra-observer variability.
Implementation of semi-automatic or automatic segmentation for 3D whole-heart CMR datasets is also challenging because of the high anatomical variability in heart defects, signal intensity variations, and low signal- and contrast-to-noise ratios.
This combination of a workshop and challenge will provide a snapshot of the current progress in the field through extended discussions, and offer researchers an opportunity to characterize their methods on a newly created and released standardized 3D whole-heart CMR dataset. The dataset will be freely available both during and after the challenge.
Three invited speakers, one each from medicine, academia and industry, will give presentations about the technical challenges of whole-heart segmentation and the benefits of virtual and physical heart models in clinical practice. The workshop will also include oral and/or poster presentations for each accepted paper. This will be followed by an open discussion on the segmentation methods, theory, metrics of segmentation quality, and applications of whole-heart segmentation for surgical planning in congenital heart disease cases.
LNCS Publication
The HVSMR proceedings were published in collaboration with the MICCAI 2016 RAMBO workshop. The papers are available at Springer.
Challenge Winners
Congratulations to the following winners of the HVSMR 2016 challenge! Winners were selected based on their performance on the testing dataset, as well as implementation factors such as reported segmentation time.
- 1st place (sponsored by IBM):
Lequan Yu, Xin Yang, Jing Qin and Pheng-Ann Heng
3D FractalNet: Dense volumetric segmentation for cardiovascular MRI volumes - 2nd place (sponsored by Arterys):
Jelmer M. Wolterink, Tim Leiner, Max A. Viergever and Ivana Isgum
Dilated convolutional neural networks for cardiovascular MR segmentation in congenital heart disease - 3rd place (sponsored by Philips):
Rahil Shahzad, Shan Gao, Qian Tao, Oleh Dzyubachyk and Rob van der Geest
Automated cardiovascular segmentation in patients with congenital heart disease from 3D CMR scans: Combining multi-atlases and level-sets
Recent Updates
[October 17, 2016] HVSMR held at MICCAI 2016
[September 20, 2016] Final paper notifications released
[August 22, 2016] Testing segmentations uploaded
[August 15, 2016] Testing images released
[August 8, 2016] Paper notifications released
[June 27, 2016] Paper submission deadline extended to July 8, 2016.
[June 27, 2016] Paper submission system opened
[June 7, 2016] Challenge submission system opened
[April 7, 2016] Cropped and short-axis training data options released
[March 29, 2016] Training datasets released
[March 28, 2016] HVSMR 2016 website launched
Sponsors
Awards for the best segmentation algorithms will be provided by: