Author + information
- Received December 28, 2010
- Revision received April 26, 2011
- Accepted May 3, 2011
- Published online August 30, 2011.
- Rebecca S. Beroukhim, MD⁎,⁎ (, )
- Ashwin Prakash, MD⁎,
- Emanuela R. Valsangiacomo Buechel, MD†,
- Joseph R. Cava, MD, PhD‡,
- Adam L. Dorfman, MD§,
- Pierluigi Festa, MD∥,
- Anthony M. Hlavacek, MD¶,
- Tiffanie R. Johnson, MD#,
- Marc S. Keller, MD⁎⁎,
- Rajesh Krishnamurthy, MD††,
- Nilanjana Misra, MD‡‡,
- Stephane Moniotte, MD, PhD§§,
- W. James Parks, MD∥∥,
- Andrew J. Powell, MD⁎,
- Brian D. Soriano, MD¶¶,
- Monvadi B. Srichai, MD##,
- Shi-Joon Yoo, MD⁎⁎⁎,
- Jing Zhou, MS⁎ and
- Tal Geva, MD⁎
- ↵⁎Reprint requests and correspondence:
Dr. Rebecca S. Beroukhim, Department of Cardiology, Children's Hospital Boston, 300 Longwood Avenue, Boston, Massachusetts 02115
Objectives The aim of this study was to report the results of an international multicenter experience of cardiac magnetic resonance imaging (MRI) evaluation of cardiac tumors in children, each with histology correlation or a diagnosis of tuberous sclerosis, and to determine which characteristics are predictive of tumor type.
Background Individual centers have relatively little experience with diagnostic imaging of cardiac tumors in children, because of their low prevalence. The accuracy of cardiac MRI diagnosis on the basis of a pre-defined set of criteria has not been tested.
Methods An international group of pediatric cardiac imaging centers was solicited for case contribution. Inclusion criteria comprised: 1) age at diagnosis ≤18 years; 2) cardiac MRI evaluation of cardiac tumor; and 3) histologic diagnosis or diagnosis of tuberous sclerosis. Data from the cardiac MRI images were analyzed for mass characteristics. On the basis of pre-defined cardiac MRI criteria derived from published data, 3 blinded investigators determined tumor type, and their consensus diagnoses were compared with histologic diagnoses.
Results Cases (n = 78) submitted from 15 centers in 4 countries had the following diagnoses: fibroma (n = 30), rhabdomyoma (n = 14), malignant tumor (n = 12), hemangioma (n = 9), thrombus (n = 4), myxoma (n = 3), teratoma (n = 2), and paraganglioma, pericardial cyst, Purkinje cell tumor, and papillary fibroelastoma (n = 1, each). Reviewers who were blinded to the histologic diagnoses correctly diagnosed 97% of the cases but included a differential diagnosis in 42%. Better image quality grade and more complete examination were associated with higher diagnostic accuracy.
Conclusions Cardiac MRI can predict the likely tumor type in the majority of children with a cardiac mass. A comprehensive imaging protocol is essential for accurate diagnosis. However, histologic diagnosis remains the gold standard, and in some cases malignancy cannot be definitively excluded on the basis of cardiac MRI images alone.
Cardiac tumors in children are rare, with a reported incidence ranging from 0.027% to 0.08% (1,2). Therefore, the experience of individual centers is relatively small. Case reports and small case series have suggested that cardiac magnetic resonance imaging (MRI) is useful in the delineation of the location, extent, and signal characteristics of cardiac tumors and masses in children (3–7). On the basis of our initial experience with 11 cases and a review of the published data, we developed cardiac MRI-based diagnostic criteria that predict tumor type (4). This project aimed to evaluate the accuracy of the diagnostic criteria against a cohort of cases with histologic diagnosis of cardiac tumors as the reference standard and to create a registry of pediatric cardiac masses with cardiac MRI data and histologic diagnosis generated from a multi-institutional experience. The term “cardiac tumor” will be used throughout the report to encompass all cardiac masses.
This was a retrospective, multicenter study of cardiac tumors. Investigators at the core laboratory (Children's Hospital Boston) solicited participation from an international group of pediatric cardiac imaging centers. Cases that met the following criteria were included: 1) patient age at diagnosis ≤18 years; 2) cardiac MRI evaluation of cardiac tumor performed between 1998 and November 2010; 3) histologic diagnosis or definitive diagnosis of tuberous sclerosis (on the basis of either revised clinical diagnostic criteria established by the Tuberous Sclerosis Consensus Conference  or genetic diagnosis of tuberous sclerosis ); and 4) submission of de-identified clinical, cardiac MRI, and pathology data. The following cases were excluded: 1) technically inadequate cardiac MRI data, including inability to analyze data at the core laboratory; and 2) inconclusive histologic diagnosis or the case did not meet criteria for tuberous sclerosis diagnosis. The study was approved by the Scientific Review Committee of the Department of Cardiology and by the Children's Hospital Boston Committee on Clinical Investigation. Contributing centers followed the policies of their respective institutional review boards.
All patient identifying information was removed at the contributing institutions, and each case was assigned a unique study number. The submitting institution completed a case information form with demographic and clinical data. This form, a copy of the cardiac MRI and pathology reports, and a digital copy of the cardiac MRI study were sent to the study coordinator for review. The following demographic and clinical information were recorded: 1) center name and location; 2) patient demographic data at the time of cardiac MRI (age, sex, weight, height); 3) presenting signs/symptoms; 4) mode of initial diagnosis (e.g., chest radiograph, echocardiogram); 5) pre-cardiac MRI diagnosis of likely tumor type; 6) date of cardiac MRI; 7) cardiac MRI technical data (e.g., scanner type, radiofrequency coil, sedation type, MR sequences, technical problems); 8) adverse events attributable or possibly attributable to cardiac MRI; 9) cardiac MRI diagnosis of tumor type at contributing institution; 10) source of tissue (e.g., tumor resection, open biopsy, transcatheter biopsy); 11) histologic diagnosis; 12) treatment; 13) date of latest follow-up; and 14) clinical status at latest follow-up.
Cardiac MRI image analysis
Cardiac MRI images were first analyzed by a single investigator (R.S.B.) at the core laboratory who was blinded to the histologic diagnosis. The following tumor characteristics were recorded: 1) size; 2) location; 3) presence of pericardial or pleural effusion; 4) hemodynamic impact of tumor; and 5) signal characteristics on cine steady state free precession (SSFP) imaging, pre- and post-contrast T1-weighted turbo (fast) spin echo imaging (T1-TSE), T2-weighted turbo (fast) spin echo imaging (T2-TSE), first pass myocardial perfusion imaging (FPP), and myocardial delayed enhancement imaging (MDE). On each of these imaging sequences the following characteristics were recorded: 1) signal intensity ratio of the tumor relative to adjacent, uninvolved myocardium as described by Kiaffas et al. (4); 2) homogenous or heterogeneous appearance; 3) whether the tumor appeared well-circumscribed and/or intramyocardial; 4) complete or incomplete examination for diagnosis of tumor type (complete examination was defined as including the necessary imaging sequences required to determine tumor type as defined in Table 1); and 5) image quality score (very good [sharply defined borders—excellent quality diagnostic information], good [minimal blurring—good quality diagnostic information with definite diagnosis], fair [structures visible with moderate blurring—able to establish diagnosis], poor [structures visible but markedly blurred—diagnosis suspected but not definite]) determined by an experienced observer (R.S.B.). Nondiagnostic quality images were excluded from analysis. However, images with fair or poor quality were included if deemed interpretable. Echo time (TE) and repetition time (TR) were recorded for black blood images.
Diagnosis of tumor type
First, a diagnostic criteria table was developed on the basis of review of the published data and prior experience. The table was then tested on each case by a consensus of 3 reviewers at the core laboratory who were blinded to the histologic diagnosis (R.S.B., A.P., T.G.). The suspected tumor type, presence or absence of imaging sequences necessary for diagnosis, and image quality were recorded for each case. After completion of the review process and data analysis, minor modifications were made to the table, and the final set of criteria are presented in Table 1.
Continuous variables were summarized as median and range, and categorical variables were described with counts and percentages. Using chi-square or Fisher exact test, the percentages of accurate diagnosis of tumor type were compared to assess how image quality and missing cardiac MRI image data influenced diagnostic accuracy. All tests were performed with 2-sided type I error rate of 0.05. Data analyses were performed with SAS software (version 9.2, SAS Institute, Inc., Cary, North Carolina).
Patients and clinical presentation
Cases (n = 90) were submitted from an international group of 15 imaging centers from 4 countries. Of those, 12 cases were excluded either due to lack of histologic diagnosis (no biopsy or resection performed, n = 9) or nondiagnostic images (n = 3). The remaining 78 cases comprised the study cohort, and their characteristics are summarized in Table 2. Because of the relatively young age of the cohort (median age 3 years at time of cardiac MRI), 73% of scans were performed under general anesthesia or with sedation (Table 3). Several of the cases included in this study have been previously published (4,5,10).
Mode of initial presentation included the following: arrhythmia (n = 29; 37%), fetal echocardiogram (n = 13; 17%), murmur (n = 12; 15%), incidental finding (n = 11; 14%), and clinical symptoms, including respiratory distress (n = 15; 19%), chest pain (n = 6; 8%), fatigue (n = 4; 5%), dizziness (n = 2; 3%), and cyanosis (n = 2; 3%). Most (97%) diagnoses were confirmed by tumor histology (biopsy or resection) with the exception of 2 cases of rhabdomyoma in patients who met diagnostic criteria for tuberous sclerosis. Among the patients with benign tumors (n = 66), most (87%) underwent partial or complete tumor resection, whereas the others underwent biopsy only (4 fibromas, 1 hemangioma, and 5 rhabdomyomas). Follow-up information was available on 46 patients for variable lengths of time (Table 2). Most patients with benign tumors were asymptomatic at last follow-up. However, there were 3 deaths among patients with benign tumors: 1 fibroma (perioperative death); 1 hemangioma (perioperative death); and 1 rhabdomyoma (died awaiting heart transplantation). In addition, residual postoperative morbidity included ventricular dysfunction (n = 3), arrhythmias (n = 3), and valve regurgitation (n = 3). Of the 9 patients with malignant tumors and follow-up information, 7 either had residual symptoms (n = 1), tumor growth (n = 1), or were deceased (n = 5).
Technical aspects of cardiac MRI
Image quality was good or very good in 81% of cases, and most cases included interpretable cine SSFP, T1-TSE, and T2-TSE imaging of the mass. However, fewer cases (particularly for those performed before 2004) included FPP (62%) or MDE sequences (58%) (Table 3). Imaging parameters for TSE sequences varied among institutions; median values (range) for T1-TSE were: TE 15 ms (5.6 to 43 ms), TR 960 ms (411 to 1,967 ms); for T2-TSE: TE 80 ms (56 to 120 ms), TR 1,463 ms (727 to 5,625 ms). The most common causes of fair or poor image quality were: 1) blurring of cardiac structures and tumor as well as suboptimal suppression of signal from the blood pool on TSE images, often seen when half-Fourier acquisition single-shot turbo spin echo sequence was employed; and 2) MDE imaging with incorrect inversion time resulting in inadequate suppression of signal from normal myocardium. The latter was most often seen when inversion time mapping or phase sensitive inversion recovery techniques were not employed.
The incidence of reported technical problems and adverse events was low (4%). All 3 patients with adverse events during the cardiac MRI examination were infants with either large fibroma or rhabdomyoma who underwent cardiac MRI under general anesthesia. Two of the infants had transient hypotension requiring fluid and/or dopamine administration after induction of anesthesia. The other patient, who had a large left ventricular rhabdomyoma resulting in hypoplastic left heart physiology, required cardioversion for symptomatic supraventricular tachycardia during the cardiac MRI. This patient ultimately died while waiting for a heart transplant.
Tumor type and location
Tumors were found in all cardiac chambers and extracardiac locations (Fig. 1). The most common tumor locations were within the ventricular myocardium (right or left ventricular free wall or ventricular septum). Tumor types are listed in Table 2.
With pre-defined diagnostic criteria, the 3 reviewers who were blinded to the histologic diagnosis classified the tumor type in each case. Of the 78 cases, 76 (97%) were correctly diagnosed, with 43 (55%) having a single correct diagnosis, and 33 (42%) having the correct diagnosis as part of a differential diagnosis. Of the 34 cases with a differential diagnosis, 16 had 2 possible diagnoses, and 18 had ≥3 possible diagnoses.
There were 2 cases of incorrect diagnosis, and both had an atypical appearance on cardiac MRI images. The first misdiagnosis was a fibroma in a right atrial location, incorrectly diagnosed as teratoma (Fig. 3). The location of the tumor was atypical, and it was difficult to distinguish right atrial from mediastinal location on the basis of the images. Cine SSFP imaging, T1-TSE, T2-TSE, and FPP imaging were performed with good image quality. However, MDE image quality was nondiagnostic. The second misdiagnosis was a rhabdomyoma that was thought to be either malignancy or thrombus (Online Fig. 1). This tumor had the appearance of a small, indistinct mass at the junction between the inferior right ventricular free wall and the septum. Cine SSFP, T1-TSE, FPP, and MDE imaging were performed with good image quality, but T2-TSE imaging was not performed.
More complete examination and better image quality were 2 factors that improved the likelihood of a single accurate diagnosis. The majority (40 of 47; 85%) of cases with complete examination had a single correct diagnosis as compared with only 3 of 31 (10%) of those with an incomplete study (p < 0.0001). With respect to image quality, 19 of 25 (76%) of the studies with very good image quality had a single correct diagnosis as compared with 17 of 38 (45%) of studies with good image quality and 7 of 15 (47%) of studies with ≤fair image quality (p = 0.04). However, a higher image quality grade was not associated with better diagnostic accuracy when controlling for completeness of examination. Of the 47 complete studies, 34 of 41 (83%) with ≥good image quality versus 6 of 6 (100%) studies with ≤fair image quality had a single accurate diagnosis (p = 0.57). Furthermore, of the 31 incomplete studies, 2 of 22 (9%) with ≥good image quality versus 1 of 9 (11%) with ≤fair image quality had a single accurate diagnosis (p = 1.0).
Of the 12 malignant tumors in this series, 3 were primary cardiac malignancies, and the remainder had a history of extracardiac malignancy. All malignant cardiac masses were correctly diagnosed as malignancy (n = 11) or possible malignancy (n = 1).
Cardiac MRI characteristics of tumors
The cardiac MRI characteristics that best define the following tumors are listed herein.
For fibroma (Fig. 4), characteristics include: 1) intramyocardial location involving the ventricular septum or free wall (with the exception of the right atrial fibroma shown in Fig. 3); 2) well-defined borders with a thin rim of myocardium; 3) strong hyperenhancement on MDE imaging with or without a hypoenhancing (dark) core; 4) hypointense on FPP or magnetic resonance angiogram; and 5) heterogeneous appearance on T1- and T2-weighted TSE sequences with variable areas of slightly hypo- or hyperintense areas.
For vascular tumors (Fig. 5), characteristics include: 1) clearly positive (iso- or hyperintense) FPP sequence; 2) variable, often weak enhancement (iso- or mildly hyperintense) on MDE imaging; and 3) variable location. Note that imaging sequences currently available by cardiac MRI might not allow distinction among benign vascular tumors (e.g., hemangioma), malignant vascular tumors (e.g., angiosarcoma), vascular malformations, and tumors with ample vascular supply (e.g., paraganglioma). However, our single case of angiosarcoma (included in the category of malignant tumors) was hypo-perfused on first pass perfusion imaging.
For rhabdomyoma (Online Fig. 2), characteristics include: 1) intramyocardial or intracavitary location, attached to the myocardium; 2) hypointense on FPP sequence; 3) isointense on MDE imaging; 4) mildly hyperintense on T2W-TSE imaging; and 5) homogenous appearance on all sequences.
For teratoma (Fig. 6), characteristics include: 1) intrapericardial, often compressing the superior vena cava and/or right atrium; 2) multilocular, bosselated mass with solid and cystic areas; and 3) hypointense on FPP. Note the similarity of the intrapericardial teratoma shown in Fig. 6 with the hemangioma shown in Fig. 5. One distinguishing characteristic was the presence of avid perfusion in the hemangioma and absence of detectable perfusion in the teratoma.
For pleuropericardial cyst, characteristics include: 1) pleuropericardial location, often in the right cardio-phrenic angle; 2) smooth-walled and well-defined; 3) hyperintense on cine SSFP; 4) strongly hyperintense on T2-TSE; and 5) hypointense on FPP.
For myxoma (Online Fig. 3), characteristics include: 1) irregular border, pedunculated, and mobile; 2) hyperintense on T2-TSE; 3) hypointense on FPP; 4) heterogeneous enhancement (iso- or hyperintense) on MDE imaging; and 5) location in any cardiac chamber (commonly left or right atrium, or atrial septum).
For fibroelastoma, characteristics include: pedunculated, mobile, endocardial, or valvular mass. Other signal characteristics have not been well defined.
Malignant Cardiac Tumors
For malignant cardiac tumors (Online Fig. 4), characteristics include: 1) history of extracardiac malignancy; 2) infiltrative appearance, defined as a) crossing an annular or tissue plane within the heart, b) involving both cardiac and extracardiac structures, or c) appearance of linear growth through a large vessel such as the superior or inferior vena cava; and 3) variable (hypo-, iso-, or hyperintense) appearance on cine SSFP, T1-TSE, T2-TSE, FPP, and/or MDE imaging. If hyperintense on MDE imaging, the appearance is heterogeneous (in contrast to the fibroma, which is strong and well defined).
Although echocardiography is the conventional screening method for intracardiac masses and tumors, cardiac MRI can provide better imaging of tumor size and location, adjacent mediastinal structures, and tumor signal characteristics (11). Despite the rare occurrence of cardiac tumors, cases evaluated by cardiac MRI have been reported in the published data (3,4,7,12,13). On the basis of this initial experience, we found that certain characteristics on cardiac MRI are likely predictive of pathologic diagnosis; however, this impression has not been tested rigorously. The present study provides the largest database of pediatric cardiac tumors with cardiac MRI and tissue diagnoses and is also the first to examine the utility of current cardiac MRI techniques in this uncommon condition.
In nearly all of our cases, the correct diagnosis was either selected or included in a differential diagnosis on the basis of the criteria outlined in Table 1. Importantly, no malignant tumor was misdiagnosed as benign on the basis of the pre-defined criteria. However, malignancy was included as part of the differential diagnosis for several benign tumors, demonstrating that malignancy might be difficult to exclude, even in cases with excellent image quality. In addition, cardiac MRI does not distinguish between specific types of highly vascular tumors such as hemangioma, vascular tumors with malignant potential (e.g., angiosarcoma), some vascular malformations, and some neuroendocrine tumors (e.g., paraganglioma). Because of the possibility of malignant potential, consideration should be given to resection of any tumor with cardiac MRI evidence of a strong vascular supply (iso- or hyperintense on FPP).
The 2 factors that best predicted diagnostic accuracy were completeness of the examination and image quality. A comprehensive examination that includes all key imaging sequences was the most important determining factor with regard to arriving at a correct single diagnosis. This observation is explained, at least in part, by the study design, which excluded nondiagnostic images. Therefore, a systematic approach to the cardiac MRI examination of pediatric patients with cardiac tumors is essential to maximize the diagnostic potential of the test. On the basis of the experience with this cohort, the following imaging sequences are recommended in pediatric patients evaluated for a cardiac tumor or mass:
1 Electrocardiography (ECG)-gated cine SSFP sequences to characterize tumor location, relation to neighboring structures, and to assess for potential interference with blood flow or ventricular function. Suggested imaging planes include axial and oblique planes across the tumor as well as assessment of biventricular size and function.
2 T1-TSE with and without fat suppression. The imaging plane should include the tumor as well as uninvolved ventricular myocardium.
3 T2-TSE with fat suppression.
4 Contrast FPP.
5 MDE imaging.
When an intracardiac thrombus is suspected, post-contrast imaging with an MDE sequence with a long inversion time (600 ms) is recommended (14). We have not found additional benefit from post-contrast T1- or T2-TSE sequences. Additional imaging sequences (e.g., flow velocity mapping, magnetic resonance angiography, coronary artery imaging to assess relationship to tumor) might be obtained to address specific additional diagnostic questions.
From a technical perspective, pitfalls were noted most often with TSE and MDE sequences. In some cases TSE images suffered from low signal/noise ratio and from blurring of the boundaries between tumor tissue and adjacent normal tissue. These problems were most frequently encountered in single-shot TSE sequences that were not ECG-gated and had relatively low spatial resolution. Superior results were usually obtained by performing TSE with ECG-gating and the use of the double inversion recovery technique to null the signal from blood. A TE of approximately 10 ms should be used for T1-weighting and approximately 70 ms for T2-weighting. The TR, which is confined to multiples of the cardiac cycle duration (R-R interval), is best set to 1 or 2 R-R intervals at heart rates <100 beats/min and 2 to 3 R-R intervals at faster heart rates to allow sufficient time for adequate signal recovery. Minimizing slice thickness while maintaining adequate signal/noise ratio and orientation of the imaging plane to maximize through-plane blood flow will also optimize image quality. With regard to MDE imaging, the most prevalent problem was inadequate nulling of signal from normal myocardium due to improper selection of the inversion time. Use of an inversion time mapping sequence and a phase-sensitive MDE sequence can minimize this problem.
Previous studies in adult patients have evaluated characteristics of cardiac and paracardiac masses by cardiac MRI with pathology correlation. In the largest study of its kind, Hoffman et al. (15) reviewed cardiac MRI characteristics of masses in 55 adults with confirmed tumor histology. They found that tumor location, tissue composition (homogenous vs. heterogeneous), and presence of pericardial or pleural effusion were predictors of malignancy. Two independent observers were accurate in predicting malignancy with an area under the receiver-operator characteristic curve between 0.88 and 0.92. However, although there were several good predictors of tumor type, no single predictor or combination of predictors could provide 100% sensitivity and specificity. Thus, although cardiac MRI provides important clinical information on cardiac tumors, malignancy cannot always be definitively excluded, requiring tissue diagnosis in certain cases—similar to our experience.
Another noteworthy observation is that the frequencies of cardiac tumors in this cohort do not reflect their overall incidence in the general population of pediatric patients. For example, in a recent review of the database at Children's Hospital Boston, rhabdomyoma represented 61% of cardiac tumors diagnosed from 1980 through 2005 (16). However, rhabdomyoma was present in only 18% of cases in our series. This apparent discrepancy is explained by referral bias, because in the majority of cases rhabdomyoma is easily diagnosed by echocardiography on the basis of the typical appearance of multiple tumors and/or a diagnosis of tuberous sclerosis (17,18). Consequently, patients with typical rhabdomyomas are not referred for cardiac MRI. The rhabdomyomas included in this series were atypical in that they were solitary, large tumors that were difficult to distinguish from other cardiac tumors by echocardiography.
Although we obtained pathology reports on all cases, the histology was not reviewed at a core laboratory. The retrospective nature of the study precluded a uniform imaging protocol, resulting in variable image quality and completeness of examinations. For example, variations in TE or TR might have resulted in variable T1- and T2-weighting of TSE images. Also, the timing of the cardiac MRI with respect to diagnosis and treatment might have impacted some of the data in specific cases. For example, a few patients had a partial tumor resection before cardiac MRI examination; thus the size of the native tumor, its hemodynamic impact, or the presence of a pericardial or pleural effusion might have been altered. Finally, as noted in the preceding text, the distribution of frequencies of tumors included in this cohort is not representative of that in the general population of children.
With a set of pre-defined diagnostic criteria, cardiac MRI imaging of cardiac tumors in children is predictive of tissue diagnosis. More complete examinations and better image quality result in improved diagnostic accuracy.
For supplementary figures, please see the online version of this article.
Characterization of Cardiac Tumors in Children by Cardiovascular Magnetic Resonance Imaging: A Multicenter Experience
This study was supported in part by the Higgins Family Noninvasive Cardiac Imaging Research Fund.
The authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- first pass myocardial perfusion
- myocardial delayed enhancement
- magnetic resonance imaging
- steady state free precession
- T1-weighted turbo (fast) spin echo
- T2-weighted turbo (fast) spin echo
- echo time
- repetition time
- turbo (fast) spin echo
- Received December 28, 2010.
- Revision received April 26, 2011.
- Accepted May 3, 2011.
- American College of Cardiology Foundation
- Moniotte S.,
- Geva T.,
- Perez-Atayde A.,
- Fulton D.R.,
- Pigula F.A.,
- Powell A.J.
- Yan A.T.,
- Coffey D.M.,
- Li Y.,
- et al.
- Roach E.S.,
- Gomez M.R.,
- Northrup H.
- Weinsaft J.W.,
- Kim H.W.,
- Shah D.J.,
- et al.
- Moss A.J.,
- Allen H.D.