This document contains definitions for the DICOM objects and attributes specific to the ACRIN 6657/I-SPY 1 MRI image data collection, and descriptions of associated data provide with this image data set.
Patient and study identification
All patients and studies are identified through standardized, deidentified attributes as shown in Table 1.
- Patient unique identification is provided through a 4 digit identifier set by the I-SPY TRIAL
- Study time-point is identified through the Clinical Trial Visit Codes: T0, T1, T2, and T3
- Imaging center discrimination has been retained (Clinical Trial Site ID) to allow for center-based analysis, but has been anonymized as codes "SITE A", "SITE B", etc…
- In general all private DICOM attributes considered HIPPA compliant by TCIA have been retained in the image sharing submission process; but it should be noted that some studies may have been anonymized prior to submission to the analysis and archiving centers, so the presence of scanner manufacturer private attributes that may be required for some analyses can not be guaranteed.
All objects have been deidentified to preserve patient privacy. If any evidence of non-HIPPA compliant patient PHI is found please notify the Image Analysis core lab at: UCSF Breast Imaging Research Program (BIRP) .
Table 1: Patient and study identification attributes included in all DICOM objects
Encoded patient ID (pppp), ACRIN Protocol ID, study name
Study name and patient ID
Clinical Trial Patient ID
Unique identification for patient in the trial
Clinical Trial Site ID
Code of the trial site:"SITE x"
Clinical Trial Visit Code
ID code for the trial visit
Clinical Trial Visit Description
Description for the trial visit:
Table 2 details the DICOM objects provided by the UCSF core lab in the I-SPY 1 / ACRIN 6698 shared data set on TCIA. Not all objects will be present in all cases due to unanalyzable studies. Original image objects are unprocessed except for necessary conversion to DICOM and deidentification. Derived image objects, including standardized early time-point percent enhancement (PE) and signal enhancement ratio (SER) maps, are provided for all volume-SER analyzable studies. DICOM segmentation objects representing PE analyzable breast tissue and SER volume are also provided. Parameters used in calculating the derived objects are stored in DICOM private group attribute set (0117,10xx) in each object as described below.
Table 2: DICOM Objects in shared data set
Object Group Name
SERIES ID *
Images and Derived Maps
Original DCE MRI Volumetric Image Sets
MRI pre contrast
pre-contrast image set
DICOM Image Files
MRI early post contrast
2'30" (nominally) post-injection image set
DICOM Image Files
MRI late post contrast
7'30" (nominally) post-injection image set
DICOM Image Files
Derived DCE Image Maps
Percent signal enhancement map at the early (nominally 2'30" post-injection) time-point relative to the pre-injection baseline image: PE=100.0 * MRI(early) / MRI(pre)
DICOM Image Files
(Sref*10000) + 1001
Signal enhancement ratio (SER) map between the early (nominally 2'30" post-injection) and late (nominally 7'30" post-injection) time-points: SER=PE(early) / PE(late)
DICOM Image Files
(Sref*10000) + 1000
Segmentation used for early post-contrast PE map
DICOM Segmentation objects
(Sref*10000) + 2001
PE thresholded SER mask
Segmentation used for SER map
DICOM Image Files
(Sref*10000) + 2000
** See Appendix A for SER derivation
The following information has been added to some or all of the DICOM objects in the data set:
- Image quality and protocol compliance assessment
- Timing information
- Tumor volume of interest (VOI)
- SER analysis parameters
- Functional tumor volume (FTV) results
All are contained in DICOM group 0117x, labeled with a private creator field:
(0117,0010) UCSF BIRP PRIVATE CREATOR 011710xx
Image quality and protocol compliance assessment
Image quality and protocol compliance were assessed by the UCSF core lab for all submitted image studies. DCE images were assessed for fat suppression, image quality, and artifacts; and then given an overall quality score. In addition, studies were evaluated for protocol violations that would prohibit volume SER analysis. QC ratings are stored in a DICOM sequence, attribute tag (0117,1024), with each separate QA rating contained in an item in this sequence, as described in Table 3. In addition, overall protocol compliance is stored in separate fields as listed in Table 3. Table 4 gives details for the different QA factors.
Table 3: DICOM fields for quality assessment
VR (VM) *
Sequence of items for each QC factor evaluated
> QC Type
Type of quality assessment. Defined terms:
> QC Factor
Quality factor evaluated
> QC value
Numerical quality assessment
> QC meaning
Meaning of quality assessment
> QC comment
Additional quality assessment comments
Protocol compliance sufficient for volume SER calculation
Protocol non-compliance reasons
Description of protocol compliance violation(s)
Table 4: Quality assessment factors for the I-SPY 1 / ACRIN 6698 data set
Quality of fat suppression. Integer scores:
Quality of images aside from fat suppression
Presence of imaging artifacts
Overall image quality for volume SER calculation
WARNING: Timing information was determined to the best of the core lab's ability based on the meta information in the original images submitted. Accuracy of the timing information cannot be guaranteed. In particular, all post-contrast times are based on the assumption that the injection and the start of the 1st post-contrast scan were simultaneous, which could not be confirmed.
Timing information fields are shown in Table 5. Timing information was added to all derived image and segmentation objects.
Table 5: Scan timing information fields for dynamic contrast-enhanced (DCE) MRI
Number of acquired time points (phases) including a single pre-contrast acquisition
Single phase acquisition duration
Acquisition start times
Starting time delay in seconds for each acquisition relative to the start of the 1st post-contrast acquisition
Assumed injection time per scanner clock
Effective acquisition delay
Effective post-injection delay for each acquisition. Non-centric phase encoding is assumed, placing the effective time half way through the acquisition
SER timing indices
Indices (0-origin) of the 3 acquisitions used in the SER calculation
Timing information method
Method used to determine the timing acquisition. Defined terms:
Timing information comments
Comments on determination of timing information
Tumor Analysis Volume - Volume of Interest (VOI) and OMIT Regions
A 3D rectangular VOI enclosing the enhancing tumor region was defined on all cases with acceptable quality and compliance for volume SER analysis. VOI are defined in the DICOM standard patient coordinate system, as defined by the Image Position Patient (0020,0032) and Image Orientation Patient (0020,0037) fields in the original DICOM image objects. Tumor VOI attributes are described in Table 6, and are included in all derived image and segmentation objects.
In cases where significant regions of non-tumor enhancement could not be excluded from the VOI without exclusion of tumor areas , "OMIT" regions of interest (ROI) were defined to mask out these regions. OMIT ROIs can currently be defined either as 3D rectangular VOI analogous to the analysis VOI, or as 2D irregularly shaped ROIs which were projected across the 3D image along one of the 3 orthogonal image axes. At the time of processing the I-SPY 1 / ACRIN 6657 data only the 2D irregular ROI OMIT form was available. OMIT regions are described in private attributes detailed in Table 7.
NOTE: The projected OMIT ROIs were defined on displayed orthogonal maximum intensity projection (MIP) images that had been interpolated to have isotropic voxel dimensions and were transposed where necessary to display in the standard radiologic orientations. Therefore, except for those projected along the z-axis (slice axis, projection axis (0117,1051) = 2) the stored X- and Y- vertices cannot be directly applied to the original images.
Table 6: DICOM Fields for rectangular VOI
Patient coordinate system specified rectangular VOI Sequence
> VOILPS Center
Center of the VOI
> VOILPS HalfWidth
1st half dimension vector of the VOI
> VOILPS HalfHeight
2nd half dimension vector of the VOI
> VOILPS HalfDepth
3rd half dimension vector of the VOI
> VOILPS Type
Use for the specified region. Defined terms:
(x,y,z) coordinates of the first voxel in the VOI
(x,y,z) coordinates of the last voxel in the VOI
- VOI_pixel_start and VOI_pixel_end are defined in cases where the Volume SER calculation was done, on the images that were used for the calculation. In out-of-protocol cases where images were acquired in the Axial plane these analyzed images will have been reformatted, cropped and/or resampled to isotropic resolution from the original images.
Table 7: DICOM Fields for description of OMIT regions: rectangular VOI and irregular projected 2D ROIs
OMIT region sequence. Each item contains either a 3D patient-coordinate system rectangular VOI or a 2D pixel-coordinate projection ROI
> VOILPS ROI flag
Type of VOI: enumerated values:
> VOILPS item
See Table 5 for attributes for rectangular VOI
> ProjectedROI npixels
Number of pixels for image used for ROI definition
> Projection axis
Image pixel axis of projection for the 2D ROI. Enumerated values: 0=x-axis, 1=y-axis, 2=z-axis
> ProjectedROI transpose flag
Flag indicating ROI coordinates are defined on a transposed image
> ProjectedROI X vertices *
X-axis pixel coordinates defining the irregular ROI
> ProjectedROI Y vertices *
Y-axis pixel coordinates defining the irregular ROI
> ProjectedROI Z range *
Z-axis (plane) range of projection of the ROI. If not present the ROI was projected across all planes in the image.
> ProjectedROI type
Type (usage) of ROI. Defined terms:
> ProjectedROI label
Label for display with the ROI
- ROI vertices are defined on the images that were used for the volume SER calculation. In out-of-protocol cases where images were acquired in the Axial plane these analyzed images will have been reformatted, cropped and/or resampled to isotropic resolution from the original images. Furthermore, for all ROI with projection axis 0 or 1 the transpose flag and npixels values must be used to convert the stored vertices into the original image coordinate system.
SER analysis parameters
Parameters used to specify the Volume SER calculation are stored in a DICOM sequence (0117,1010) described in Table 8. Table 9 lists the parameters used, with each parameter being described in one item in the sequence. See Appendix A for a description of the Volume SER calculation.
Table 8: DICOM sequence for storing analysis parameters
> Parameter type
Parameter type. Enumerated values: FLOAT, INTEGER, STRING
> Parameter name
> Parameter description
Description of parameter
> Floating parameter value
Value of floating point parameter
> Integer parameter value
Value of integer parameter
> String parameter value
Value of string parameter
Table 9: Parameters for Volumetric Signal Enhancement Ratio (VOLSER) Analysis of Dynamic Contrast-enhanced (DCE) MRI stored in Parameter sequence (0117,1010). Each item in the sequence describes one parameter.
Method used for pre-contrast selection of breast fibroglandular tissue regions. Defined terms:
Intensity threshold applied to pre-contrast T1 image to select fibroglandular tissue regions.
Background masking level percentage
PEthresh: early percent enhancement threshold
Kernel size for a minimum connectivity filter for SER analysis: voxels with fewer than this number of immediate neighbors passing the pre-contrast intensity and PE threshold tests were not included in the SER volume.
Flag indicating that SER values were adjusted for scan timing.
Parameters used for correction of SER values for acquisitions with significant protocol timing errors. Present if and only if ser_time_correct is present and equal to 1.
Functional tumor volume (FTV) results
Functional tumor volume (FTV = FTV(PEthresh, SERmin, SERmax) ) is defined as the volume of tissue within the tumor VOI, or otherwise segmented breast tissue region, with a PE greater than or equal to the early PE enhancement threshold (PEthresh) and an SER greater than a specified minimum SERmin and less than or equal to a specified maximum SERmax. SERmax is assumed to be infinite if not specified. Calculated FTV values are stored in the DICOM segmentation objects using the sequence described in Table 10. For the I-SPY 1 / ACRIN 6657 data set two FTV are reported: FTVPE (PEthresh, SERmin=0.0, SERmax=∞) and FTVSER (PEthresh, SERmin=0.9, SERmax=∞), where PEthresh was set empirically for each imaging center.
Table 10: DICOM sequence for storing functional tumor volume (FTV) results
MRI SER FTV results
> SER Minimum
Minimum value of SER
> SER Maximum
Maximum value of SER: assumed to be infinite if not specified
> Voxel count
FTV number of voxels
FTV in cc
Display label for FTV result
I-SPY Patient Clinical Data
A set of Excel files are provided giving a subset of the clinical data collected on the study subjects. Descriptions of these data fields are provided within those files and in the attached dictionary documents:
|Patient Clinical Data||DataDict - TCIA Shared Patient Clinical|
|Patient Outcome Data||DataDict - TCIA Shared Outcome|
Annotated Image Markup (AIM) Files
The FTV results will also be presented in AIM files accompanying the image data sets. [To be available at a future date.]
Appendix A: Functional Tumor Volume (FTV)1,2,3
Signal Enhancement Ratio (SER) is a combined enhancement/washout measure derived from dynamic contrast enhanced MRI scans. Three time-points are used: pre-contrast injection, early post-contrast, and late post-contrast. Each acquisition is a high spatial resolution, 3D, T1-weighted scan. Sequential (non-centric) phase encoding is used to ensure that the effective acquisition time for time-points 2 and 3 can be taken as the time from contrast injection to the midpoint of the MRI scan. This time is generally 0.75 to 2.5 minutes after injection for the early time-point, and 7.5 minutes or greater for the late time-point. Initial validation studies and the ACRIN 6657 protocol were done with MRI acquisition duration of 5 minutes, with post-contrast scan timings of 2.5 and 7.5 minutes.
Tumor vascularity can be characterized by the percent enhancement (PE) of a post-contrast time-point S1, from the pre-contrast time-point S0, which reflects contrast uptake in the tissue and is given by .SER, given by the ratio of the PE at the early post-contrast time to the PE at the late post-contrast time, adds a measure of the washout rate in the tissue. SER is given by: .SER is a three-point approximation of the contrast-enhancement curve that has previously been shown to correlate well with tumor microvessel density and tumor grade, with promising prognostic value for breast cancer. Both PE and SER are calculated on a per-pixel basis.
We calculate functional tumor volume (FTV) using a semi-automated tumor segmentation algorithm based on the PE and SER maps. To avoid including skin and chest wall enhancement and imaging artifacts, analysis is limited to an operator selected rectangular volume of interest (VOI). The VOI is usually drawn on a set of orthogonal maximal intensity projection (MIP) images taken either from the early post-contrast image or from a subtraction image S1-S0. For a minority of cases it is also necessary for the operator to draw one or more irregularly shaped exclusion regions to eliminate non-tumor enhancement regions that can not be excluded with the rectangular VOI. All further processing is fully automatic. A map consisting of the SER of each voxel is calculated using 3 levels of filtering: a pre-contrast intensity background mask level set to 60% of the 95th percentile intensity of the VOI is used to reduce spurious noise and to exclude low signal regions such as suppressed adipose tissue and strongly enhancing vessels; a PE threshold, typically 70%, at the early post-contrast time point is applied to segment malignant tissue from normal appearing tissue; a connectivity test is applied to the combined background and PE threshold mask, requiring a minimal number of connected neighboring voxels, to eliminate speckle noise. An SER color map is generated for qualitative assessment, showing areas of strong enhancement and washout (SER>0.9) in a gradation of colors from white to green, while enhancing but non-washing out tissue (SER<0.9) is shown in blue. FTVPE is calculated by summing the volumes of all voxels within the VOI passing all the filtering steps and having a positive SER. Inclusion of the low SER component of the map was found to be beneficial to getting a useable FTV measure in post-chemotherapy pre-surgery examinations where enhancement values are significantly depressed relative to pre-treatment values. FTVSER, measured similarly but with a lower limit of SER > 0.9, giving a volume measure of the washout regions of the lesions, was also investigated.
For further information see:
1.Partridge SC, Gibbs JE, Lu Y, et al: Accuracy of MR imaging for revealing residual breast cancer in patients who have undergone neoadjuvant chemotherapy. AJR Am J Roentgenol 179:1193-9, 2002
2.Hylton NM, Blume JD, Bernreuter WK, et al: Locally advanced breast cancer: MR imaging for prediction of response to neoadjuvant chemotherapy--results from ACRIN 6657/I-SPY TRIAL. Radiology 263:663-72, 2012
3.ACRIN PROTOCOL 6657 / CALGB 150007 http://www.acrin.org/6657_protocol.aspx Contrast-Enhanced Breast MRI for Evaluation of Patients Undergoing Neoadjuvant Treatment for Locally Advanced Breast Cancer