Test Page for Joe
- This site will be used to test wiki features and posting options only.
Quantitative Imaging Biomarker Alliance (QIBA) for Volumetric CT Image Analysis:
Subgroup 1C – Assessing Impact of Instrumental Variability on Volumetrics November 10, 2008
Discussion items:
1. Charge and scope of QIBA CT Volumetrics, Subgroup 1C, characterizing instrumental variability.
Conduct a designed experiment to analyze CT scans of a phantom imaged under a range of conditions to assess the impact of scanners on the performance relative to the requirements posed by the use of the measurements in clinically relevant scenarios. Use radiologist markup as another volume measure to be assessed along side of automated measurements. Use “physical” volume measures as reference “truth”. 2. Questions to be answered Measure the precision and bias of CT volume measurements on phantom data, as applied to two profiles: (1) the smaller nodules as typical in Stage I neoadjuvant or early diagnostic settings, and (2) larger tumors as typical in Stage IV that form the mainstay of most drug development trials. [Note that stages II and III would draw from this work.] 3. Instrument recruitment/selection A key contribution of this work is to address multi-center and multi-vendor variability, so it would be desireable to have broad manufacturer participation as well as a range of centers FDA has scanned the Shiba Soku phantom with a Siemens 64 and a Philips 16. 4. Experimental design The primary basis for the activity is drawn from a systems analysis of the sources of variability in volumetric CT represented in a matrix . Based on this analysis, sources of variability that are relevant to each considered profile would constitute “factors”, and levels for these factors would be derived to sample the appropriate range as might exist in practical real-world settings. The experimental design is organized so as to optimize the use of resources (time, equipment, organizational attention etc.) of stakeholders to satisfy necessary precursor questions needed to set the profile details. This essentially entails the use of fractional factorial designs that restrict the number of treatment combinations for factor subsets needed to characterize the variability as would be seen in applying the profile. Designs are constructed that limit the number of treatment combinations within subsets of factors. Such designs have been applied in a number of settings, and rigorous statistical work-up is pursued. For example, to provide a comprehensive analysis evaluating all of the sources of variability using a full factorial would exhaust the stakeholders, but it is possible to analytically determine what treatment combinations are actually necessary with respect to the specific clinical conditions as well as to understand in what ranges experiments need to be performed. Others have cited the benefits of such design efficiencies while retaining a comprehensive nature to the results. This subgroup effort documents and pursues the design for the stated profiles as well as establishing methodology that could be applied more broadly for other profiles in CT and/or other quantitative imaging settings. Examples of scanner Settings – which parameters are critical to control for constancy between scanners include slice thicknesses, non-ovelapping slices, mAs, kVp, FoV/voxel spacing. Fuller listing of scanner dependent parameters that may be studied. a. Potential differences in imaging protocol 1. reconstructed slice thickness and spacing/overlap 2. collimation 3. mAs 4. pitch 5. reconstruction kernel 6. kVp 7. Reconstructed Field of View 8. With larger detector row arrays providing extended z-axis coverage, sequential scans may be performed, and lesions may would span - or be at the interface of - several acquisitions FOVs
b. Potential differences between imaging hardware MDCT 1. number of detector rows 2. number and size of detector elements in each row 3. number of sources (e.g. dual source CT) 4. x-ray spectrum/beam filtration (e.g softer vs. harder x-ray) 5. software version: acquisition, correction or reconstruction
5. Phantoms for DICOM imaging - propose Image Set 1: FDA-type phantom with 3 to 20 mm phantom nodules. Phantom QA – NIST Pocket Phantom? Should QIBA acquire a phantom? 6. Algorithms: - At least one algorithm - Measured precision and bias - Intra and inter-reader variability - Precision across repeat scans.
7. Identification of required resources: Existing FDA image data sets acquired on multiple CT scanners are . Readers Other resources needed: algorithm, automatic format for study parameters.
8. Next steps Recruit/identify readers and reading tasks. Recruit participation from multiple scanners. Specify experimental conditions – source of phantom, scan parameters, Specify data interface for automated algorithm/CAD tool iinput. Proposal: adopt specification from CT Volumentrics 1B. 9. Notional complexity estimate: 5 scanners, 2 slice thicknesses, 4 nodule sizes/shapes, 3 repeat scans = 15 scans. x 2 reconstructions x 4 nodules = 120 nodule volumes to analyze.