The Cancer Imaging Program of the National Cancer Institute (NCI) in collaboration with the International Society for Biomedical Imaging (ISBI) will launch a grand challenge in segmentation of internal structures of the prostate gland based on magnetic resonance imaging data. The challenge will take place at the ISBI Symposium which takes place April 7-11, 2013 in San Francisco, CA. A full list of challenges taking place at the ISBI Symposium can be found at http://www.biomedicalimaging.org/2013/program/isbi-challenges/.
Recently, the high spatial resolution and soft-tissue contrast offered by MRI makes it the most accurate method available for obtaining this kind of information. This, combined with the potential of MRI to localize and grade prostate cancer, has led to a rapid increase in its adoption and increasing research interest in its use for this application. Furthermore, the medical image analysis and clinical community has been very interested in developing accurate prostate MRI segmentation methods which apart from accurate, automated prostate volume segmentation also serve an important purpose for computer-aided detection and diagnostic algorithms, as well as a number of multi-modality image registration algorithms, which aim to enable MRI-derived information on anatomy and tumour location and extent to aid therapy planning and guidance.
The prostate gland consists of internal structures including the peripheral zone (PZ), central zone (CZ), and transition zone (TZ), where the latter 2 structures are jointly referred to as the central gland (CG). While most tumors are found in the PZ, tumors can also be found in the CG, and CG tumors can have drastically different appearance than PZ tumors. In the PZ (where most cancers are found), tumors are typically chacterized by hypointense regions on MRI images, in stark contrast to the usually hyperintense PZ regions. However, in the CG, tumors are typically noticeable due to their homogeneous texture, as compared to the traditionally hetergeneous texture in the CG. In recent years, several computer aided detection (CAD) systems have been developed for detecting tumors from prostate MRI imagery. Since the tumors in the PZ can appear drastically different from tumors in the CT, CAD systems would invariably benefit from knowing where each internal prostate structure was located.
In addition, treatment options can even be tailored to an individual patient, as CG tumors have been found to be significantly less aggressive compared to PZ tumors. However, most extant prostate segmentation systems only consider the prostate capsule boundary. A methodology to automatically and simultaneously segment the prostate, PZ, and CG from T2-weighted MRI alone could also allow for the development of a spatially aware CAD system for prostate cancer detection, by leveraging the explicit, automated segmentations of the different prostate zones. A second application is to create patient-specific treatment models based on the zonal location of the tumor.
This year's ISBI challenges participants to outline those two non-overlapping adjacent regions of the gland. Advancing solutions to this vexing clinical problem will surely gain recognition of the power of computer image processing to address a widespread but critical healthcare puzzle. Algorithms applicable to both 1.5T and 3T will be most welcome since the scanner installed base is presently mostly 1.5T.
Data for the prostate challenge has been provided by Boston University and Radboud University, Nijmegen Medical Centre (the Netherlands). The challenge data has been divided into three groups:
A 60 case prostate MRI 3D series "training set" consisting of DICOM images with matched .NRRD markups that defines the tissue "central zone (CZ)" and "peripheral zone (PZ)" are available from the National Cancer Institute (NCI) Cancer Imaging Archive – "TCIA".
There will be 60 cases, each consisting of 15+ (4 mm thick slices at 3T) or 28+ (3 mm thick at 1.5T) adjacent axial cross-section cuts. Half of them were obtained at 1.5 T (Philips Achieva) with an endo-rectal receiver coil (from (BMC) Boston Medical Center) and the other half at 3T (Siemens TIM) with a surface coil (from Radboud University Nijmegen Medical Centre (RUNMC), Netherlands). They were acquired as T2 weighted MR axial pulse sequence. Each file contains nearly all DICOM acquisition parameters except tags that specifically identify Private Health Information (PHI).
To complete the training data set, each case has had individual slice CZ and PZ outlines marked by Drs Nicolas Bloch (at Boston University School of Medicine) / Mirabela Rusu (Case Western U.) or Drs Henkjan Huisman / Geert Litjens / Futterer at RUNMC Netherlands. Use the following links to download the .NRRD markups:
Once you've trained your algorithm on those 60 cases the ISBI Challenge Committee will provide you access to a 10 new case "Leaderboard" set to which you will not have .NRRD markup data. "Leaderboard" is a popular term (borrowed from golf) for a procedure to see how well you are doing on a pre-test set to get a sense your relative competitive rank before submitting to a final test. The results of your participation in the Leaderboard will be returned to you as feedback on a subset of sequestered markups retained on the judging analytic software using Kitware's MIDAS software. You may then use those results for further tuning your algorithm to prepare yourself for the final "Test" set of 10 totally new cases on which you will run your algorithm while attending the ISBI meeting in April 2013.
For a comprehensive background on the prostate, its anatomy, imaging and clinical observations see the following publications:
Committee and publications that promote a PI-RADS reporting standard see:
For healthcare context one MRI case as reported by a radiologist see linked report on training case ProstateDx-01-0001:
Questions about the structure and conduct of the Challenge please contact:
Questions about the TCIA image archive and downloading procedures please review the landing page documentation or contact:
Questions about 3D Slicer and .nrrd file structure contact:
Questions about how Kitware designed its scoring software called MIDAS see below attached document link:
Keyvan Farahani, PhD (farahank@mail.nih.gov)
Anant Madabhushi, PhD (anant.madabhushi@case.edu)
Henkjan Huisman, PhD (H.Huisman@rad.umcn.nl)
Nicolas Bloch, MD, (nicolas.bloch@bmc.org)
Mirabela Rusu, PhD (mirabela.rusu@case.edu)
Geert Litjens MS (g.litjens@rad.umcn.nl)
Stephen Aylward. PhD (stephen.aylward@kitware.com)
Andinet Enquobahrie (andinet.enqu@kitware.com)
Patrick Reynolds (patrick.reynolds@kitware.com)
Michael Grauer (michael.grauer@kitware.com)
John Freymann (freymanj@mail.nih.gov)
Justin Kirby, (kirbyju@mail.nih.gov)
Carl Jaffe MD (carl.jaffe@bmc.org)