2009 prostate segmentation challenge MICCAI

Goal

To discuss the state-of-art for prostate segmentation of MR images in the context of MRI-guided prostate therapy, through comparison of the segmentation methods using sample data.

Workshop

Thursday - 24th September at MICCAI 2009 (London, UK) http://www.miccai2009.org/

Sponsor

• National Center for Image Guided Therapy (NCIGT, PI: Jolesz, Tempany)
• National Alliance for Medical Image Computing (NA-MIC, PI: Kikinis)

Organizer

• Nobuhiko Hata (co-lead organizer, BWH, Boston, MA, Email hata at-mark bwh.harvard.edu)
• Gabor Fichtinger (co-lead organizer, Queens Univ, Kingston, ON)
• Sota Oguro (Data managing, BWH, Boston, MA)
• Haytham Elhawary (Slicer module for validation, BWH, Boston, MA)
• Theo van Walsum (Organizer of the umbrella MICCAI workshop "3D segmentation in Clinic").

Rules

1. Download 3D Slicer for viewing and manipulating the dataset we provide below. All files to be downloaded are available in the Download section below.
2. (Optional) Download the prostate image processing tutorial and its associated images.
3. Download the MR images of the prostate and the manually segmented gland (as training data). The segmentations have been performed by expert radiologists.
4. Perform the segmentation of the prostate gland with your algorithm. The segmentation result should be NRRD format and loadable to Slicer. The segmentation result should be aligned to the original MR image when loaded to Slicer. In addition the segmentation result should be a label map image maintaining the same pixel intensity value assigned to each segmentation image in the training dataset. A module has been created in Slicer3 which can calculate the 95% Hausdorff distance and Dice Similarity Coefficient between your segmentation and the manual segmentations of the prostate provided.
5. Create an account at this NA-MIC wiki: Account Request. Once your account has been accepted, upload the NRRD file with your segmentation results in the Contestants section of this page, along with the paper describing the method developed and applied.
6. The organizer will score the segmentation result by comparing it with the manual segmentation of the expert using the announced comparison scheme: 95% Hausdorff distance and Dice Similarity Coeeficient.
7. Repeat step above at the workshop with the competition dataset available on the day.

Those who submit paper and make their code available at Insight Journal (http://www.insight-journal.org/)  will get extra points. 
Those who submit paper and make their code available as a Slicer module will get extra recognition at the contest.

Dates

1. Training data available: June 2009
2. Submission of results of segmented training data: starts June 25th 2009 and continues until the paper submission deadline.
3. Submission of papers (with results) (no page limit, Springer/MICCAI format/PDF file): July 20, 2009
4. Workshop: September 2009

Downloads

1. 3D Slicer
2. Prostate MRI tutorial for 3D Slicer
3. Training MRI data with expert's segmentation
4. Subject data (for paper submission)
5. Contest Data on site

Contestants Results Submission

Use the following template to upload results:

1. Albert Gubern-Merida and Robert Marti, University of Girona, Link to Page
2. Jason Dowling (AEHRC CSIRO, Australia), Jurgen Fripp (AEHRC CSIRO, Australia), Peter Greer (Newcastle Mater Hospital, Australia), Sebastien Ourselin (UCL, UK), Olivier Salvado (AEHRC CSIRO, Australia), Link to Page

Contestants On-site Results Submission

1. Albert Gubern-Merida and Robert Marti, University of Girona, Link to Page
2. Jason Dowling (AEHRC CSIRO, Australia), Jurgen Fripp (AEHRC CSIRO, Australia), Peter Greer (Newcastle Mater Hospital, Australia), Sebastien Ourselin (UCL, UK), Olivier Salvado (AEHRC CSIRO, Australia), Link to Page

Results

File:MICCAI2009SegmentationChallenge.ppt

Evaluation Scheme

Two evaluation methods are used to determine the accuracy of the segmentation: the 95% Hausdorff distance and the Dice Similarity Coefficient.

• 95% Hausdorff distance (HD)

The HD is calculated between the contours of the segmented prostates which result from the manual segmentation and the proposed contestant algorithm. The HD is the maximum distance of a set to the nearest point in another set. A more formal description of the HD can be found in the following publication:

• • Archip N, Clatz O, Whalen S, et al. Non-rigid alignment of pre-operative MRI, fMRI, and DT-MRI with intra-operative MRI for enhanced visualization and navigation in image-guided neurosurgery. Neuroimage 2007; 35:609-624.

• Dice Similarity Coefficient (DSC)

The DSC, which has been extensively used in the evaluation of segmentation, gives a measure of the volumetric overlap between the two segmented prostates. The DSC indicates twice the number of voxels which are shared by or are common to both segmentations divided by the total number of voxels in both datasets. The DSC can range from zero to one, where zero is no alignment between segmented glands and one is perfect alignment. An example of the DSC used in prostate segmetation can be found at:

• • A. Bharatha, M. Hirose, N. Hata, S. K. Warfield, M. Ferrant, K. H. Zou, E. Suarez-Santana, J. Ruiz-Alzola, A. D'Amico, R. A. Cormack, R. Kikinis, F. A. Jolesz, and C. M. Tempany, "Evaluation of three-dimensional finite element-based deformable registration of pre- and intraoperative prostate imaging," Med Phys, vol. 28(12), pp. 2551-60, 2001.