Multi-echo gradient echo (MEGRE) MRI data with 8 different echo times (2.38-23.6 ms) from 326 + 40 prostate cancer patients with gold fiducial markers inserted into the prostate (train/validation + test dataset). Underlying description of the technique and its first use has been described in a previous publication. A scientific paper that utilizes this dataset for deep learning is being reviewed for publication.
The data contains an image volume for each patient, for each echo time. The center of mass from the three inserted prostate gold fiducial markers was manually defined. The ground truth label for this dataset consist of spherical objects with a radius of 1-12 mm, inserted in the center of mass defined locations. Method in the paper uses 9 mm radius.
Keywords: Gold fiducial marker, MRI only, Radiotherapy, MRI radiation therapy, Prostate, Cancer, Radiology.
Sample images with reduced image quality. Please click to preview.
|Cite as||Christian Jamtheim Gustafsson (2020) Multi-echo gradient echo (MEGRE) doi:10.23698/aida/megre|
|Title||Multi-echo gradient echo (MEGRE)|
Christian Jamtheim Gustafsson
|Resolution||Scan resolution 2.8x1.46x1.46 mm, reconstructed to 0.47x0.47x2.8 mm. Each patient has 28-34 slices with 512x512 image matrix.|
GE Discovery 750w 3T
|Copyright||Copyright 2020 Skånes Universitetssjukhus, Christian Jamtheim Gustafsson|
|Download||Please contact Christian Jamtheim Gustafsson, Joel Hedlund, or Claes Lundstrom to request access.|
For 100 out of the 326 patients in the train/validation dataset the MEGRE images were annotated in MATLAB where the center of mass (CoM) coordinates of the gold fiducial marker signal voids were defined by one observer. For the remaining 226 of the patients in the train/validation dataset and the 40 patients in the test dataset MEGRE images, echo number one or two was exported to the treatment planning system Eclipse v.15.1 where the CoM points of the gold fiducial marker signal voids were defined by multiple observers and exported as DICOM RT-structures. Further details on this are provided in an article currently in review for publication. All fiducial identifications were confirmed to be correct using the corresponding CT images.
See github repository for pre-processing and method source code
Compressed NIfTI (.nii.gz) file format for both images and ground truth segmentations. The data contains an image volume for each patient, for each echo time.
Data augmentation was performed for each of the 326 patients with random image rotations in an interval between -15 to 15 degrees around the superior-inferior patient axis using linear interpolation. Image data for each echo for each patient time was handled separately but subjected to the same amount of rotation. The same rotation was applied to the ground truth labels using nearest neighbor interpolation (to avoid producing non-binary mask values). The augmented data and labels for each patient data was saved as a new subject, producing a total of 652 subjects in the train/validation dataset, which equaled a total of 1946 fiducial objects. No data augmentation was performed for the test data set containing 40 patients.
All image data for all echos in all datasets was independently N4 bias field corrected in an image pre-processing pipeline.
Train/validation data set and ground truth label:
Pat 1-150 = Real patients
Pat 151-300 = Augmented versions of Pat 1-150
Pat 301-476 = Real patients
Pat 477-652 = Augmented versions of Pat 301-476
Test data set and ground truth label:
Pat 1-40 = real patients
Patient 1 had four fiducial markers inserted and was excluded in the deep learning study (in review). Each patient volume for each echo in test data set Pat 1-40 was subjected to Z-score normalization after N4 bias field correction.
Patients40processedN4v2Normalized for Pat 1-40
Patients326processedPooledAugN4v2 for Pat 1-652
Folder structure for images:
Folder structure for ground truth:
Copyright 2020 Skånes Universitetssjukhus, Christian Jamtheim Gustafsson
Permission to use, copy, modify, and/or distribute this data within Analytic Imaging Diagnostics Arena (AIDA) for any purpose with or without fee is hereby granted, provided that the above copyright notice and this permission notice appear in all copies, and that publications resulting from the use of this data cite the following works:
Deep learning paper (in review).
Christian Jamtheim Gustafsson (2020) Multi-echo gradient echo (MEGRE) doi:10.23698/aida/megre.
THE DATA IS PROVIDED “AS IS” AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS DATA INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR CHARACTERISTICS OF THIS DATA.