Effect of contrast dose on diagnostic performance in DCE‐MR breast imaging

Abstract Objective To assess the diagnostic performance of breast magnetic resonance (MR) imaging as a function of gadolinium contrast dose using a retrospective reader study. Material and Methods IRB approval was obtained prior to the start of this study and was HIPAA compliant. One‐hundred and fifty MR breast examinations were included that were acquired between January 2001 and December 2006. Seventy‐five patients received contrast doses (gadopentetate dimeglumine) by weight of 0.10 mmol/kg and 75 patients were imaged using fixed volumes of 20 ml. The images were assessed by two radiologists with performance calculated for each reader as well as a combined assessment. Dose response was measured by comparing performance between cases binned by dose: <=0.10; >0.10; and >0.13 mmol/kg. Statistical significance was calculated using a one‐sided Z‐test for differences in proportions with interobserver agreement calculated using Cohen's kappa statistics. Results In the combined reader assessment with equivocal lesions classified as negative, sensitivity rose from 66% (19/29) to 92% (24/26, P < 0.01) and 95% (18/19, P < 0.01) with the specificity also increasing from 65% (32/49) to 87% (40/46, P < 0.01) and 86% (32/37, P = 0.01) corresponding to doses <=0.10, >0.10, >0.13 mmol/kg. With equivocal lesions classified as positive, sensitivity rose from 79% (23/29) to 92% (24/26, P < 0.10) and 95% (18/19, P < 0.10) Specificity also increased from 53% (26/49) to 72% (33/46, P < 0.05) and 70% (26/37, P = 0.05) with increasing dose. Interobserver agreement also improved at the higher doses.

across studies, institutions, and manufacturers, particularly for quantification of contrast kinetics. At this point most imaging sites started calculating the dose by weight using 0.1 mmol/kg. Although the American College of Radiology recommends this dose for breast studies 1 it is not clear how this dose was determined. In the few dose-response studies that were performed, there is some disparity in conclusions.
Heywang-Kobrunner et al compared performance using 0.16 mmol/kg gadopentetate dimeglumine versus a low dose of 0.1 mmol/kg of body weight. 2 The conclusion of this study was that conspicuity of breast lesions was much improved using the higher dose. In a study of the relationship between contrast dose with uptake kinetics using three dose groups (<0.122 mmol/kg; 0.123-0.155 mmol/kg; and> 0.155 mmol/kg), Jansen et al reported that initial and peak enhancement increased with contrast dose for in-situ and invasive cancers. 3 However, Knopp et al did not find any improvement in diagnostic accuracy using 0.2 mmol/kg gadopentetate dimeglumine over 0.1 mmol/kg and concluded that a dose of 0.1 mmol/kg was "probably sufficient". 4 The reported difference in dose response may be due to different inclusion criteria and diagnosis occurrence rates in each study.
For example, in Knopp's study, subjects were recruited who had an "abnormality highly suspicion of being breast cancer" with 75% containing malignancies of which only 5% were DCIS and 8% were ILC. 4 Jansen's study used only malignant lesions but 33% of the lesions were classified as DCIS. 3 This suggests that lesions assessed as highly suspicious by mammogram and/or ultrasound will have sufficient conspicuity on MR at low doses of contrast to allow a confident assessment. However, not all cancers respond in the same manner, particularly those with lower angiogenic activity. 5 In the role that DCE-MR breast imaging currently fills, (confirmation, staging, and localization), the low dose may be sufficient. However, in screening studies for women at high risk or with dense breasts, in which the highest conspicuity is desired, the optimal dose may be higher than what is currently recommended.
The concern over the safety of gadolinium-based contrast agents (GBCAs) is an important issue that continues to influence the choice of contrast dose. Nephrogenic systemic fibrosis (NSF) was first described by Grobner et al 6 and Markmann et al 7 who reported that patients developed the disease after being exposed to GBCAs. However, upon review of adverse events, the FDA determined that there were no confirmed cases of NSF in patients who had normal kidney function or mild-to-moderate kidney insufficiency. Although the report recommended caution when using these agents, GBCAs are safe when patients are adequately screened. 8 More recently, there has been concern raised over the detection of brain deposits in patients who have received repeated MRI contrast studies [9][10][11] although the risks, if any, from these deposits is currently unknown.
These are important issues and may justify the use of lower GBCA doses. However, before coming to that conclusion, the trade-off in diagnostic performance should be understood in order to make an informed risk/benefit decision. The purpose of this study was to assess diagnostic performance of breast MR imaging as a function of gadolinium contrast dose using a retrospective reader study of examinations that were acquired during a period of time when the contrast dose was changed from a fixed volume to a weight-based calculation.

2.A | Patients
IRB approval was obtained prior to the start of this study and was HIPAA compliant. One-hundred and fifty MR breast examinations were included that were acquired between January 2001 and December 2006, which covered a period before and after the transition from a fixed volume contrast dose to weight-based calculation of dose. These examinations were chosen because they used similar imaging protocols and had pathologic correlation or 2-yr follow-up.  Table 1 shows the distribution of diagnoses for all cases.

2.B | MR imaging
Images were acquired at 1.5 T using a General Electric Signa (43 examinations) or a Siemens Sonata (107 examinations). Subjects were placed in the prone position with the breasts gently compressed within a dedicated bilateral breast coil. All imaging series were performed in the sagittal plane. T1 and T2 weighted images were acquired followed by a dynamic contrast-enhanced series. The DCE series was acquired using a radial fast 3D spoiled gradient-recalled sequence using 512 data samples with 384 projections and 32 slices of 3 mm thickness. 12 Although this was an IRB-approved investigational sequence at the time the patient studies were acquired, the protocol can be reproduced with commercial sequences from either manufacturer. Seventy-five examinations were performed unilaterally, and the remaining 75 cases were bilateral examinations. Scan parameters were as follows: TR = 10 ms; TE = 4 ms; flip angle = 20°-45°. The fat signal was suppressed using a spectral inversion pulse played-out twice per slice group. A high-resolution baseline volume was acquired followed by three postcontrast volumes acquisitions over the following 6-minute period. Contrast Among the 107 patients (mean age 49 yr, mean weight 73 kg) performed on the Sonata scanner, 37% (40/107) had malignant findings.   Note that the two higher dose groups were formed by setting the dose threshold; therefore, cases in the highest dose group were also included in the >0.10 mmol/kg analysis. Stratifying the dose groups into more specific bins was not feasible since there would be too few positive cases in each group to achieve significance. Statistical significance was calculated using a one-sided Z-test for differences in proportions between the low-dose group and each of the higher dose groups. The one-sided test was justified since there are no reports that diagnostic performance decreases with increased contrast dose. Statistics were analyzed using Microsoft Excel 2013.  and 86% (32/37) at the higher doses. The increased sensitivity for reader 1 was not statistically significant (P > 0.2); however, the improved performance for reader 2 showed significance of P < 0.10 for the increase in sensitivity and P < 0.01 for the increase in specificity. The combined reader assessment showed the sensitivity rising from 79% (23/29) to 92% (24/26, P < 0.10) and 94% (17/18, P < 0.10) with the specificity also increasing from 53% (26/49) to 72% (33/46, P < 0.05) and 70% (26/37, P < 0.10).

2.C | Image assessment
With the indeterminate lesions classified as negative (  The first reader classified more lesions as indeterminate than did the second reader and how those were interpreted had a large effect on performance measures. In the true clinical setting, equivocal lesions would not be simply classified as a group to be benign or malignant but could instead be followed to assess changes. Additional information from other examinations could also make a definitive diagnosis possible in many cases. Therefore, many of these would subsequently be correctly classified as benign or malignant and the performance measures would be higher. However, regardless of whether the indeterminate lesions were classified as positive, negative to removed altogether, reader performance improved at the higher contrast doses. older compounds at 0.1 mmol/kg. This study suggests that there would be a benefit to taking advantage of the higher CNR that can be achieved using these agents.

| DISCUSSION
The safety of GBCAs may be the overriding factor in choosing the dose, but the trade-off in diagnostic performance should be understood in order to make an informed decision. This study showed that the diagnostic performance of DCE-MR breast imaging was greater when subjects received a higher gadolinium dose than the current standard of 0.10 mmol/kg.

CONF LICT OF I NTEREST
None of the authors has any conflict of interest. There is no industry support for this project.