Anahita Fathi Kazerooni1, Laura C. Bell2, Floris Van den Abeele3, Ruben Verhack3, Xinze Zhou4, Salman Rezaei5, Zaki Ahmed6, Rianne Van Der Heijden7, Seyed Ali Nabavizadeh4, Leland S. Hu8, Hamidreza Saligheh Rad5, and Steven Sourbron9
1Department of Radiology, UPenn, Philadelphia, PA, United States, 2Division of Neuroimaging Research, Barrow Neurological Institute, Phoenix, PA, United States, 3Hyperfusion.ai, Gent, Belgium, 4University of Pennsylvania, Philadelphia, PA, United States, 5Quantitative MR Imaging and Spectroscopy Group, Research Center for Molecular and Cellular Imaging, Tehran University of Medical Sciences, Tehran, Iran (Islamic Republic of), 6Mayo Clinic, Rochester, MN, United States, 7Department of Radiology & Nuclear Medicine, Erasmus MC University Medical Center, Rotterdam, Netherlands, 8Neuroradiology Division, Department of Radiology, Mayo Clinic, Phoenix, AZ, United States, 9University of Sheffield, Sheffield, United Kingdom
Synopsis
Dynamic contrast enhanced (DCE-) MRI is widely acquired as a
part of neuroimaging protocol for evaluation of glioblastoma tumors before the
start of therapy or monitoring and assessment of treatment response in
longitudinal scans. Nonetheless, lack of a standardized quantification method
has limited its application in clinical settings, multi-institutional studies and
clinical trials. These challenges have motivated efforts to validate DCE-MRI
using benchmark biomedical image analysis methods. The Open-Source Initiative
for Perfusion Imaging (OSIPI) has designed the OSIPI-DCE challenge to evaluate
and compare DCE tools in terms of accuracy, repeatability, and reproducibility
of Ktrans estimation in the brain.
Purpose
Research findings cannot be reproduced when researchers evaluate
new methods on personal datasets and local code that is not shared with the
public [1]. With a growing attention to
reproducible research at ISMRM [2], community challenges with
public datasets can play an important role in validating and benchmarking image
quantification methods [3]. While dynamic contrast
enhanced (DCE-) MRI has been widely used to evaluate a diversity of brain
pathologies, there is a lack of standardized analysis methods, limiting its
application in clinical practice, multi-institutional studies, or clinical
trials.
The Open-Source Initiative for Perfusion Imaging (OSIPI), an
initiative of ISMRM perfusion study group and sponsored by ISMRM, was founded
to promote reproducible research and open science in perfusion imaging and
facilitate translation of software tools into clinical practice. In accordance
with the aims of OSIPI, we have designed the DCE (OSIPI-DCE) challenge for
evaluating DCE-MRI methods that estimate the volume transfer constant (Ktrans)
in brain tumors. Challenge Overview
We will evaluate and compare the analysis methods submitted
by the competing teams for quantification of Ktrans from
DCE-MRI in terms of accuracy, repeatability, and reproducibility. The accuracy
of Ktrans estimations by the participating teams will be scored
based on synthetic data specifically designed for this challenge; repeatability
will be rated based on test-retest DCE-MRI scans of 10 patients with glioblastoma
brain tumor; reproducibility will be assessed based on an independent analysis by
a neutral team. The submissions will be ranked according to a global score
reflecting that an ideal method should be accurate AND repeatable AND reproducible
(Table 1). The challenge will be advertised in http://challenge.ismrm.org, results will
be announced in the Perfusion Study Group Workshop in February 2022, and the
top three teams will present their methods at the ISMRM 2022 annual meeting.
Teams interested in this challenge will be asked to register
for free on http://challenge.ismrm.org.
Upon registration, the teams will receive submission guidelines and a link to
the data, which is uploaded in a repository. The challenge will run over a
course of 9 months, but each competing group will be granted 3 months from the
time of their registration to submit their reports and results (Figure 1).
At the end of the challenge, all the submissions will be processed and
evaluated by the organizing team.Data
The following data will be provided for the competing teams:
- Synthetic: This synthetic
data is designed based on a two-compartment exchange model (2CXM) and the signal
model of a spoiled gradient-echo sequence in steady-state. It contains multiple
structures with specifications of normal gray matter, normal white matter,
tumor, necrosis, and arterial input function (AIF) taking a range of values for
PS, Ve, Vp, and F. Temporal
sampling time is 6 s, the total scan time is 6 min and Rician noise has been
added. T1 values for AIF and tissue are 1400ms and 1000ms, respectively.
- Clinical: A set of
test-retest DCE-MR images from 10 patients with newly diagnosed glioblastoma
scanned on 3T scanners at two scan dates typically 2-5 days apart [4,5] will be provided (details of
imaging parameters can be found in Table 2).
Submissions and evaluation
Competing teams will be required to submit (1) matrices of voxel-wise Ktrans
maps for the synthetic and clinical data (2 visits per patient) in NIfTI format,
and (2) Standard Operating Procedures (SOPs) that will allow a neutral team to reproduce
the results. The SOPs should explain how to access and install the software
used. There is no requirement to release the source code, but for commercial
software or in-house software that is not freely available, a temporary license
should be granted to the neutral team so the results can be reproduced. The
submissions will be ranked according to the OSIPI-DCE score defined in Table
1.Discussion
The OSIPI-DCE challenge will provide a ranking of available
software tools that will help future users select an appropriate method for
their analysis. After the challenge is concluded, the data and procedures will
be made freely available to serve as a benchmark for any future developments in
Ktrans mapping in the brain. The expectation is that providing
robust benchmarks will ultimately promote a more standardized approach to the
analysis of DCE and provide an evidence base for a future consensus [6]. As such, this challenge can
support the uptake of Ktrans as a biomarker in
multi-institutional clinical trials, drug development and ultimately clinical
practice. Acknowledgements
The
authors would like to acknowledge Dr. Keyvan Farahani, from National Cancer
Institute, USA, and Dr.Natenael Semmineh, from Barrow Neurological Institute, USA,
for their help in providing information for the design of this challengeReferences
- Maier-Hein, L. et al. Why
rankings of biomedical image analysis competitions should be interpreted with
care. Nat. Commun. 9, 5217 (2018).
- Stikov,
N., Trzasko, J. D. & Bernstein, M. A. Reproducibility and the future of MRI
research. Magn. Reson. Med. 82, 1981–1983 (2019).
- Maier-Hein,
L. et al. BIAS: Transparent reporting of biomedical image analysis
challenges. Med. Image Anal. 66, 101796 (2020).
- Prah,
M. A. et al. Repeatability of Standardized and Normalized Relative CBV
in Patients with Newly Diagnosed Glioblastoma. Am. J. Neuroradiol. 36,
1654 LP – 1661 (2015).
- Clark,
K. et al. The Cancer Imaging Archive (TCIA): Maintaining and Operating a
Public Information Repository. J. Digit. Imaging 26, 1045–1057
(2013).
- Boxerman,
J. L. et al. Consensus Recommendations for a Dynamic Susceptibility
Contrast MRI Protocol for Use in High-Grade Gliomas. Neuro. Oncol.
(2020) doi:10.1093/neuonc/noaa141.