A Prototype Image Quality Assurance System for Accelerated Quantitative Breast DCE-MRI
Yuan Le1, Aneela Afzal2, Xiao Chen3, Bruce Spottiswoode4, Wei Huang2, and Chen Lin1

1Radiology and Imaging Science, Indiana University School of Medicine, Indianapolis, IN, United States, 2Advanced Imaging Research Center, Oregon Health and Science University, Portland, OR, United States, 3Siemens Healthcare, Princeton, NJ, United States, 4Siemens Healthcare, Chicago, IL, United States


The goal of this work is to build a prototype quality assurance (QA) system for the quantitative pharmacokinetic (PK) analysis of breast DCE-MRI acquired with accelerated imaging techniques. A 3D digital tumor model with two sub-regions was constructed by segmenting patient images. The dynamic contrast enhanced images were synthesized according to the Tofts and Shutter Speed models with the TWIST technique. The QA system shows how the TWIST technique impacts the estimated pharmacokinetic parameters, and therefore allows necessary adjustments to be made to control the error.

Target Audience

Radiologists, and MRI scientists.


Quantitative pharmacokinetic (PK) analysis from DCE-MRI have been shown to provide more reliable evaluation of breast cancer response to neoadjuvant chemotherapy (NAC) than tumor size measurement based on Response Evaluation Criteria In Solid Tumors (RECIST) guidelines (1-3). In order to achieve the temporal resolution required for accurate PK parameter evaluation without sacrificing spatial resolution, accelerated imaging methods are often employed in breast DCE-MRI (4, 5). However, these techniques can result in infidelity of the DCE signal intensity time course (6-9) and consequently errors in estimated PK parameters. A quality assurance (QA) system, which evaluates DCE-MRI image quality/accuracy for PK analysis when accelerated techniques are implemented, is an unmet need. Here we report preliminary findings from such a prototype QA system that we have developed and used on images acquired with TWIST technique, a k-space view sharing technique commercially available on the Siemens platform (5).


Figure 1 gives an overview of the implemented prototype QA system. A background phantom filled with 0.2mM Gd-BOPTA, resembling un-enhanced tissue, was scanned on a 3T scanner (MAGNETOM Skyra, Siemens Healthcare, Erlangen) with the TWIST sequence but fully sampled (5), and TR=4ms, Flip angle=10°. A 3D tumor model was retrospectively constructed from patient data, with two sub-regions, one with wash-out and one with persistent enhancement (Figure 2). Tumor k-space data were generated using separate 3D Fourier transforms of the two sub-regions zero filled to match background phantom k-space data. Enhancement curves were created for the two sub-regions using both the Tofts model (10) and the Shutter-Speed Model (SSM) (3) with PK parameters derived from breast cancer patient data (3). For wash-out and persistent regions, Tofts model parameters were Ktrans=0.14 and 0.05 (min-1) and ve=0.32 and 0.28, respectively; and SSM parameters were Ktrans=0.21 and 0.07 (min-1), ve=0.60 and 0.59, and τi (mean intracellular water lifetime, SSM only parameter) = 0.37 and 0.37, respectively. These parameters also serve as the “ground truth” for evaluating any error due to acquisition or reconstruction. The tumor and background k-space data were re-organized according to the data acquisition scheme of the TWIST sequence for percentage of center k-space pA = 10%, 15%, 20% and 33%, and sampling density of peripheral k-space pB = 20% and 50%. Tumor k-space views were multiplied with the enhancement at the corresponding time point (Figure 1). The enhanced tumor data were then added to the background data. In addition, an ‘Faithful’ data set was reconstructed not only with fully sampled k-space data, but also with all the k-space data for one measurement acquired at the same time. The temporal resolution was set to 16 sec, the minimum temporal resolution required for accurate PK analysis of breast DCE-MRI data (11). After the images were reconstructed from the synthesized k-space data (with different pA and pB combinations) using Siemens Image Calculation Environment, the PK parameters for a ROI in the wash-out sub-region were estimated using both the Tofts model and SSM and compared with the ground truth.


Figure 3 shows the average enhancement in the wash-out sub-region from view-shared images and the ‘ground truth’. TWIST view sharing increased the ‘temporal foot print’ of each measurement, and sometimes caused ‘enhancement’ at or even before the arrival of the contrast (pA=10% and pB=50%). Table 1 lists the Ktrans, ve and τi parameters estimated with the two PK models. These parameters were closest to the ‘ground truth’ when using the ‘Faithful’ images. When images with view sharing, the PK parameters deviated further from the ‘ground truth’ to different extents. Figure 4 shows examples of the SSM Ktrans, ve, and τi color maps of the selected ROI, obtained from an ‘Faithful’ and a view-shared data set (pA=10% and pB=20%). With view-sharing, the PK parameter maps also became less uniform.


We have developed a novel prototype QA system for breast DCE-MRI to evaluate the accuracy of estimated PK parameters. TWIST sequence was used as an example, but this QA system can be easily adapted to other fast imaging methods that use k-space under-sampling. Future studies will include testing the QA system with additional tumor models, enhancement types, imaging sequences, and acquisition parameter options. The ultimate goal is to implement such QA systems on scanners for real-time evaluation of the impact of imaging parameters on PK analysis accuracy (or bias)so the users can select the imaging parameters appropriately.


Grant Support: NIH grant U01 CA154602


1. D. K. Woolf, et al., Breast Cancer Res Treat, 2014, 2(147), 335-43 2. V. Dialani, et al., Ann Surg Oncol, 2015, 5(22), 1416-24 3. W. Huang, et al., Transl Oncol, 2014, 1(7), 153-66 4. K.-H. Herrmann, et al., In: ISMRM, 2010, 2503 5. R. Janka, et al., In: ISMRM, 2010, 4553 6. T. Song, et al., Magn Reson Med, 2009, 5(61), 1242-8 7. Y. Le, et al., In: ISMRM, 2012, 2404 8. Y. Le, et al., In: ISMRM, 2013, 3375 9. Y. Le, et al., In: ISMRM, 2014, 6436 10. P. S. Tofts, et al., J Magn Reson Imaging, 1999, 3(10), 223-32 11. E. Henderson, et al., Magn Reson Imaging, 1998, 9(16), 1057-73


Fig 1. Diagram of the prototype quality assurance system. Acceleration 1, 2 represent imaging parameters related to the image acquisition acceleration, such as pA and pB for TWIST.

Fig 2. Tumor model based on patient data, with wash-out (brighter) and persistent (darker) sub-regions.

Fig 3. Curves from ‘Faithful’ or view shared images vs. ‘ground truth’ curve based on pre-set PK parameter values.

Fig 4. SSM based PK maps obtained from the simulated tumor model data. (a, c, e) results from ‘Faithful’ images; (b, d, f) results from images with pA=10% and pB=20%. (a,b) Ktrans map; (c, d) ve map; (e, f) τi map.

Table 1. Estimated PK parameters from tumor model.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)