Clinical Radiology 69 (2014) 378e384
Contents lists available at ScienceDirect
Clinical Radiology journal homepage: www.clinicalradiologyonline.net
Diffusion-weighted breast imaging at 3 T: Preliminary experience ~o, E. Matos, R.G. Nunes, H.A. Ferreira, J. Loureiro, L. Nogueira*, S. Branda I. Ramos ˇ
~o Joa ~o, Alameda Prof. Hernani Monteiro and School of Health MRI Unit, Department of Radiology, Hospital de Sa Technology of Porto/Polytechnic Institute of Porto (ESTSP/IPP), Rua Valente Perfeito, Porto, Portugal
article in formation Article history: Received 15 June 2013 Received in revised form 29 October 2013 Accepted 7 November 2013
AIM: To evaluate the performance of diffusion-weighted imaging (DWI) at 3 T for the detection and characterization of breast lesions. MATERIALS AND METHODS: Magnetic resonance imaging (MRI) of the breast, including DWI single-shot spin-echo echo planar images (SS-SE-EPI; eight b-values, 50e3000 s/mm2), were acquired in women with a clinical indication for breast MRI. The exclusion criteria were as follows: (1) previous breast surgery, radiotherapy and/or chemotherapy within the prior 48 months (14 women); (2) only cystic lesions (one woman); (3) no detectable enhancing lesion at dynamic contrast-enhanced (DCE)-MRI (15 women); and (4) breast implants (four women). MRI results were corroborated by histopathology or imaging follow-up. Apparent diffusion coefficients (ADCs) were estimated for lesions and normal glandular tissue. Differences in the ADC between tissue types were evaluated and the sensitivity and specificity of the method calculated by receiver operating characteristics (ROC) curves. RESULTS: The final cohort comprised 53 patients with 59 lesions. Histopathology was obtained for 58 lesions. One lesion was validated as benign on imaging follow-up. Mean ADCs of 1.99 0.27 103 mm2/s, 1.08 0.25 103 mm2/s, and 1.74 0.35 103 mm2/s were obtained for normal tissue, malignant, and benign lesions, respectively. Mean ADCs of malignancies were significantly lower than those of benign lesions (p < 0.001) and normal tissue (p < 0.0001). The sensitivity and specificity for stratifying lesions, considering an ADC threshold of 1.41 103 mm2/s, were 94.3% and 87.5%, respectively; accuracy was 91.5%. CONCLUSION: DWI proved useful for the detection and characterization of breast lesions in the present sample. ADC values provide a high diagnostic performance for differentiation between benign and malignant lesions. Ó 2013 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.
Introduction Breast cancer is among the most common diseases affecting women worldwide, carrying a high mortality rate.1 Early detection and treatment may increase survival and * Guarantor and correspondent: L. Nogueira, MRI Unit, Department of Radiology, Hospital de S~ ao Jo~ ao, Alameda Prof. Hernani Monteiro, 4200-319 Porto, Portugal. Tel.: þ351 914 117 547. E-mail addresses: [email protected], [email protected] (L. Nogueira).
improve quality of life,2 which is why diagnostic accuracy is critical. Magnetic resonance imaging (MRI) has emerged as an important method for breast cancer detection, based on its high sensitivity for detection and characterization of breast disease.3 Currently, it is the method of choice to detect invasive carcinoma, and it is routinely used for screening women at high risk of breast cancer.4 Although the sensitivity of MRI for breast cancer detection is in the range of 85e100%,5,6 it presents a specificity ranging from only 37e88%.7,8 Various studies have investigated ways to overcome this limitation, and the use of diffusion-weighted imaging (DWI) has shown
ˇ
0009-9260/$ e see front matter Ó 2013 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.crad.2013.11.005
L. Nogueira et al. / Clinical Radiology 69 (2014) 378e384
promise.9e11 This technique focuses on the random motion of water molecules within tissues of the human body. Changes in mobility are reflected as differences in the signal intensity (SI) and can be quantified using the apparent diffusion coefficient (ADC), expressed as 103 mm2/s.12 Different studies have used DW-MRI to examine breast tissue, although mostly at 1.5 T.13e15 In keeping with histological features, ADCs of malignant and benign lesions have been shown to differ, with values comparatively lower for malignant lesions.9,10,16,17 However, few prospective quantitative studies have been undertaken using 3 T equipment, which may provide added benefits. The increased signal-tonoise ratio (SNR) at 3 T enables increased spatial resolution, allowing the use of higher diffusion sensitization b-values with an adequate SNR. The present study was conducted to evaluate the clinical performance of DW-MRI at 3 T in the characterization of breast lesions. The aims of the present study were as follows: (1) to determine mean ADCs of malignant and benign lesions, as well as normal glandular tissue; (2) to establish a threshold ADC for differentiating between malignant and benign lesions; and (3) to evaluate the diagnostic performance of DWI in breast lesion characterization.
Materials and methods This prospective study was conducted in women with a clinical indication for breast MRI between September 2010 to April 2011. The study was approved by the institutional review board, and written informed consent was obtained from all patients after the procedure had been explained. All patients referred for MRI during this period and who agreed to participate were enrolled in the study, resulting in 87 patients being enrolled. The referred group included women who underwent breast MRI for preoperative staging, high-risk screening, suspicious lesions on mammography and/or ultrasound and therapeutic monitoring. The following exclusion criteria were applied to patients: (1) breast surgery, radiotherapy and/or chemotherapy within the prior 48 months (14 women); (2) with only cystic lesions (one woman); (3) without detectable enhancing lesions at dynamic contrast-enhanced (DCEMRI) (15 women); and (4) breast implants (four women). Inclusion criteria for lesion DWI analysis were: (1) a minimum size of 0.7 cm at DCE-MRI; (2) definitive diagnosis via histology (core needle biopsy or excised at surgery), or a minimum 2-year follow-up using mammography, ultrasound, or MRI examinations. For premenopausal women, MRI examination was performed in the second week of the menstrual cycle. When breast MRI examination was performed after biopsy, a minimum time interval of 8e10 days between biopsy and MRI was considered to reduce the potential for signal increase due to haemorrhage.
Acquisition protocol MRI examinations were performed using a 3 T MRI machine (TimÒ Trio, Siemens, Erlangen, Germany). A dedicated
379
four-channel breast radiofrequency (RF) phased-array coil was used. Patients were examined in a resting prone position. The study protocol included the following sequences: bilateral axial T2-weighted (W) turbo spin-echo (TSE); sagittal unilateral T1W three-dimensional (3D) fast low angle shot (FLASH) without fat suppression; unilateral sagittal T2W TSE sequence using short time inversion recovery (STIR). Unilateral sagittal DWI images were obtained separately for each breast, with free breathing, using a single-shot spinecho echo planar imaging (SS-SE-EPI) sequence with STIR for fat suppression. Sensitizing diffusion gradients were applied sequentially in the x-, y-, and z-directions, with bvalues of 50, 200, 400, 600, 800, 1000, 2000, and 3000 s/ mm2 In this protocol, eight b-values with a minimum echotime (TE) of 108 ms, to include ultra-high b-values (2000 and 3000 s/mm2). These were used to explore non-Gaussian models of diffusion, to evaluate whether these can provide more information on lesion characterization (results to be reported separately). For healthy fibroglandular tissue at 3 T, a transverse relaxation time T2 of 71 6 ms has been reported.18 Using higher b-values meant prolonging the achieved TE. To compensate for this signal decrease, the number of excitations was increased to 3. DCE images were obtained with a T1W 3D FLASH sequence using spectrally adiabatic inversion recovery (SPAIR) for fat suppression; six dynamic phases were acquired using automatic injection of gadobenate dimeglumine (MultiHance; Bracco) via an antecubital vein at a rate of 2 ml/s (total dose 0.1 mmol/kg body weight), followed by a 20 ml saline flush. After the administration of contrast material, unilateral sagittal fat-suppressed (water excitation) T1W 3D FLASH images were repeated. Table 1 presents the imaging sequences in sequential order and their main parameters of acquisition. Total imaging time was approximately 40 min.
Image analysis All images were analysed using an image-processing platform [Siemens Medical Systems, Work in progress (WIP) V.17A]. The same radiologist, experienced in breast imaging (6 years), performed conventional MRI interpretations. Data were evaluated according to morphological and kinetic criteria described in Breast Imaging Reporting and Data System e Magnetic Resonance Imaging (BIRADS e MRI).19 Sizing of lesions was determined from dynamic images, using the ruler function. The longest dimension was considered. Two researchers retrospectively analysed the DWI-MRI images in consensus. The readers were blinded to the final diagnosis (histological results) and regions of interest (ROI) were marked according to the conventional MRI report descriptions. At DWI-MRI, both lesions and uninvolved breast parenchyma were evaluated quantitatively, using as the reference the visual inspection of T1- or T2W morphological, dynamic and contrast-enhanced images. ROIs were drawn on diffusion images (b ¼ 400 s/mm2), and then copied to the other DWI images and ADC map. The ROI sizes were set to 0.10 cm2 for lesions and normal tissue, and
380
L. Nogueira et al. / Clinical Radiology 69 (2014) 378e384
Table 1 Acquisition protocol for conventional magnetic resonance imaging (MRI) and diffusion-weighted imaging (DWI) pulse sequences. Parameters Sequence Orientation TR/TE (ms) TI (ms) Fat suppression FOV (mm2) Matrix Section thickness Number of sections NEX Bandwidth (Hz/pixel) Scan time (min) b-Values (s/mm2)
Conventional pre-contrast T2W TSE Axial bilateral 4990/88 _ _ 320 320 512 384 4 26 2 305 2:06 _
T1W 3D Flash Sagittal 17/4.9 _ _ 200 200 275 384 2 64 1 430 3:49 _
T2W TSE Sagittal 4920/67 210 STIR 200 200 448 314 4 26 2 248 4:26 _
DWI STIR
Dynamic
Contrast enhanced
Single-shot EPI Sagittal 4900/108 240 STIR 250 250 84 128 5 16 3 1628 5:58 50, 200, 400, 600, 800, 1000, 2000 and 3000
T1W 3D FLASH Axial bilateral 3.77/1.42 _ SPAIR 320 320 358 448 0.9 160 1 490 4:32 _
T1W 3D FLASH Sagittal 7.8/3.9 _ Water excitation 160 160 256 256 0.9 144 1 450 3:12 _
EPI, echo-planar imaging; FOV, field of view; NEX, number of excitations, SPAIR, spectrally adiabatic inversion recovery; STIR, short tau inversion recovery; T1 3D FLASH, three-dimensional gradient-echo fast low angle shot; TI, inversion time; TR/TE, repetition time/echo time; TSE, turbo spin echo.
the mean value and standard deviation (SD) of the SI were calculated. In the case of lesions, ROIs were drawn so as to include the area of highest hyperintensity. Normal glandular tissue was evaluated in women with both breasts and unilateral lesions. In these cases, measurements were performed also in the contralateral side. To attempt the same location for ROI placement, the ROIs were defined on the sections that included each nipple, and in regions without abnormal enhancement. The mean ADC value within the ROI was calculated using a mono-exponential fit with six b-values (50, 200, 400, 600, 800, and 1000 s/mm2), using the software provided by the system manufacturer. As this is the only analysis model available on typical machines, the two high b-value images were excluded from the ADC quantification, and used for another study.
Statistical analysis Sample characteristics were analysed accounting for patient’s age and lesion size. Mean ADCs for benign, malignant lesions, and normal tissue were calculated. Differences in mean ADCs between normal, benign, and malignant tissues were then further assessed using Student’s t-test. The ADC cut-off values for benign and malignant lesions were calculated using as a reference histological and imaging results. Sensitivity and specificity calculations incorporated the threshold for maximum sensitivity and specificity rendered by the receiver operating characteristics (ROC) curve. Lesions with mean ADCs below threshold were scored as malignant. All computations were performed using the PASW Statistics V17 software. Statistical significance was set at p < 0.05.
Results Subjects and lesion characteristics The final cohort included 53 women (26e82 years, mean age SD was 52.1 12.4 years). Fifty-nine lesions were
detected in both conventional and DWI-MRI datasets. Fiftyfive lesions (93.2%) could be seen as distinct masses and four (6.8%) behaved like non-masses on DCE imaging. The mean size of benign lesions was 12.91 5.85 mm (7e30 mm), while that of malignant lesions was 21.48 10.6 mm (7e60 mm). Histopathological confirmation was obtained for 58 lesions via excisional surgery and/or needle core biopsy, identifying 35 (59.3%) as malignant (5 ductal carcinoma in situ; 1 lobular carcinoma in situ; 20 invasive ductal carcinoma; 9 invasive lobular carcinoma) and 23 as benign (13 fibroadenomas; 4 epithelial proliferative lesions; 2 papillomas; 4 fibrocystic change). One additional lesion was presumed to be benign, based on morphological features and dimensional stability, as observed by mammography and ultrasound during a 2-year follow-up, leading to a total of 24 (40.7%) benign lesions. Fig 1 illustrates MRI images of a 38-year-old woman with malignant tumour (invasive ductal carcinoma) including different types of images: dynamic acquisition (Fig 1a), post-contrast T1W 3D FLASH (Fig 1b), DWI with a b-value of 1000 s/mm2 (Fig 1c), and mean ADC map (Fig 1d).
Quantitative evaluation of normal glandular tissue and lesions The mean ADC of normal glandular tissue was 1.99 0.27 103 mm2/s. In women with only one lesion, the mean ADC of normal glandular tissue did not differ significantly depending on lesion type (malignant versus benign, p ¼ 0.58). Table 2 depicts descriptive ADC analysis for normal glandular tissue, and benign and malignant lesions. The mean ADC of the malignant breast lesions was significantly lower than that of benign lesions (p < 0.001) and also compared to normal glandular tissue (p < 0.0001). Mean ADCs for both types of lesion are presented in Fig 2. Two of the malignancies evaluated showed mean ADCs that were 1.5 times beyond the interquartile range (outlier values): one of the lesions had an ADC below the first quartile (Q1 ¼ 0.91), whereas the ADC of the second lesion was above the third quartile (Q1 ¼ 1.27).
L. Nogueira et al. / Clinical Radiology 69 (2014) 378e384
381
Figure 1 A 38-year-old woman with suspected malignant lesion in the right breast. (a) Dynamic contrast-enhanced acquisition image shows a spiculate mass with heterogeneous enhancement. (b) Sagittal post-contrast high-resolution fat-suppressed T1W 3D FLASH. (c) DWI with a bvalue of 1000 s/mm2 image shows a hyperintense lesion. (d) The respective mean ADC map where the lesion shows hypointense signal and a central area of necrosis. Histological diagnosis: invasive ductal carcinoma, grade III.
The lowest mean ADC (0.46 0.25 103 mm2/s) corresponded to an invasive lobular carcinoma (20 mm) and the highest mean ADC (1.76 0.25 103 mm2/s) to an invasive ductal carcinoma (20 mm).
both histopathological results and imaging criteria and using ROC curve analysis (Fig 3). The area under the curve was 0.94 (0.87e1), enabling a mean ADC value of 1.41 103 mm2/s to be used for discriminating malignant
Diagnostic performance of ADC measurements To assess the accuracy of DWI for lesion characterization, the mean ADC threshold value was calculated, attending to Table 2 Descriptive apparent diffusion coefficient (ADC) statistics for normal glandular tissue, and benign and malignant lesions. ADC values (103 mm2/s)
Mean Standard deviation 95% Confidence intervals Median Minimumemaximum value
Normal tissue (n ¼ 42)
Benign lesions (n ¼ 24)
Malignant lesions (n ¼ 35)
1.99 0.27 1.89e2.08 1.98 1.56e2.72
1.74 0.35 1.59e1.89 1.80 1.05e2.3
1.08 0.25 0.99e1.16 1.09 0.46e1.76
Figure 2 Box plot show distributions of mean ADC (103 mm2s) calculated from six b-values (50e1000 s/mm2) for benign and malignant lesions.
382
L. Nogueira et al. / Clinical Radiology 69 (2014) 378e384
Figure 3 ROC curve used to differentiate benign and malignant lesions according to mean ADC, calculated with all b values. The area under the ROC was 0.94 (0.87e1). An ADC cut-off value of 1.41 103 mm2/s corresponded to a sensitivity of 94.3% and specificity of 87.5%.
versus benign lesions. Mean ADCs below this threshold were considered to correspond to malignant lesions. With this threshold, 12.5% (3/24) of benign lesions (three fibroadenomas with mean ADCs of 1.05 103 mm2/s, 1.12 103 mm2/s, and 1.27 103 mm2/s) and 5.7% (2/35) of the malignancies (one invasive lobular carcinoma with mean ADC 1.43 103 mm2/s and one ductal carcinoma in situ with mean ADC 1.76 103 mm2/s) would have been mislabelled. Sensitivity and specificity values of 94.3% and 87.5% were obtained. The positive predictive value was 91.7% and the negative predictive value was 91.3%. Altogether, 91.5% of the cases were correctly diagnosed.
Discussion Water diffusion in biological tissues is a process that reflects exchanges of different components between intraor extracellular compartments and vascular space. DWI focuses on the mobility of water molecules in the tissues, reflecting higher or lower mobility at a cellular level.12 In normal tissues, water movement is conditioned by cell membranes and macromolecules. When pathological processes are present, alterations in cellularity, tortuosity, and architectural changes occur. In more compact tissue, these changes cause reduction of the normal configuration of the extracellular space, leading to hindered or restricted water diffusion.20 The impact of these changes can be visualized via the ADC. This measurement parameter represents an indirect index of tissue complexity, assuming that malignant lesions have a more complex microstructure than benign or normal tissues. Previous studies have shown that breast lesions may be differentiated by mean ADC values with malignancies yielding lower values when compared to benign or normal tissues.14e16 Despite some differences in study protocols, the present results corroborate previously reported mean ADCs for normal glandular tissue and for malignant and benign lesions,11,17,21,22 confirming the feasibility of DWI-MRI to
discriminate between normal and diseased states. Mean ADCs of malignancies were significantly lower than those of benign lesions, as a likely consequence of differences in tissue microstructure (i.e., cell density) resulting in reduced extracellular space.23 Mean ADCs of 1.99 0.27 103 mm2/ s for normal glandular tissue, 1.74 0.35 103 mm2/s for benign lesions, and 1.08 0.25 103 mm2/s for malignancies were obtained; all of which were similar to absolute ADCs cited by other publications.21,24,25 Cystic lesions were excluded from the present study as they show high ADCs; their inclusion in the benign group would have biased the ADC value of benign lesions towards higher values, thus affecting the overall ADC quantification and interfering with the aim of this study. Although the present data are comparable to prior results obtained at 1.5 and 3 T, it is possible that the mean ADCs for malignancies may have been overestimated. The reason for this is that many lesions were examined (26/59) after biopsy, which could have altered their mean ADC due to the presence of residual haemorrhagic components. Nevertheless, Rahbar et al.26 reported no significant difference in mean ADC values before and after biopsy in ductal carcinoma in situ. However, further studies would be required to confirm whether this result is also applicable to other histological subtypes of breast carcinoma. In theory, using a higher number of b-values lowers the uncertainty associated with estimated ADC values, in comparison with only two b-values.20 As ADC estimates can depend on the number of b-values used, cut-off values for lesion classification could depend on this parameter. However, others groups have shown similar results when two or more b-values were used for ADC estimation.21,22 Establishing a threshold value for lesion classification (malignant versus benign) greatly affects the sensitivity and specificity of DWI. In a recent meta-analysis,27 threshold values ranged from 1.1 103 mm2/s to 1.6 103 mm2/s, placing the present mean ADC threshold of 1.41 103 mm2/ s within the acceptable range of variation. With this mean ADC threshold value, a sensitivity of 94.3% and a specificity of 87.5% were achieved. For the present study, accuracy was 91.5%, which is slightly higher than that of other studies ranging from 83.6e91%.11,13,28 Preliminary evidence suggests that the intense cellularity of some benign lesions may produce false-positive results.13,15,29 In the present study, three false-positives cases were found (all fibroadenomas). However, only these three lesions were misclassified from a total of 13 fibroadenomas. A possible explanation for the false-positive cases is a presence of a certain amount of fibrotic tissue that may restrict water movement, resulting in lower ADC values. These findings were reported by other groups22,30 and are related to the variable spectrum in tissue composition and degree of cellularity, which can be found for this type of lesion. Falsenegative results included one ductal carcinoma in situ and one extensively spread invasive lobular carcinoma. Possible explanations for these false-negative cases could be related to partial volume effects, the section thickness used (5 mm), or the inadvertent inclusion of areas of necrosis in the ROI, which caused ADC elevation.
L. Nogueira et al. / Clinical Radiology 69 (2014) 378e384
Lesion detection based on DWI is dependent on diffusion sensitization b-values, as well as lesion size and histological composition.31 The present study confirms the ability of the DWI sequence to distinguish breast lesions from surrounding tissue, depicting all as hyperintense areas in all the cases. Nevertheless, some authors have found malignant lesions that were not visible on DWI sequences (6e37.5%),24,30,32 which can be attributed to the protocol used, regarding the b-values selected, and/or the inherent histology of the lesions. Others have suggested that the pattern of non-mass lesions is less easily detected. In the present study, there were four such cases. Conceivably, this low figure of non-mass lesions may have biased lesion detection in the present results. Although conditioned to the present criteria for lesion analysis, others small lesions that were described in the conventional MRI report were also depicted in DWI images. It is expected that DWI at 3 T should enable better detection and characterization of lesions by providing a higher SNR and greater spatial resolution. Matsuoka et al.33 reported improved visualization at 3 versus 1.5 T for lesions
Contents lists available at ScienceDirect
Clinical Radiology journal homepage: www.clinicalradiologyonline.net
Diffusion-weighted breast imaging at 3 T: Preliminary experience ~o, E. Matos, R.G. Nunes, H.A. Ferreira, J. Loureiro, L. Nogueira*, S. Branda I. Ramos ˇ
~o Joa ~o, Alameda Prof. Hernani Monteiro and School of Health MRI Unit, Department of Radiology, Hospital de Sa Technology of Porto/Polytechnic Institute of Porto (ESTSP/IPP), Rua Valente Perfeito, Porto, Portugal
article in formation Article history: Received 15 June 2013 Received in revised form 29 October 2013 Accepted 7 November 2013
AIM: To evaluate the performance of diffusion-weighted imaging (DWI) at 3 T for the detection and characterization of breast lesions. MATERIALS AND METHODS: Magnetic resonance imaging (MRI) of the breast, including DWI single-shot spin-echo echo planar images (SS-SE-EPI; eight b-values, 50e3000 s/mm2), were acquired in women with a clinical indication for breast MRI. The exclusion criteria were as follows: (1) previous breast surgery, radiotherapy and/or chemotherapy within the prior 48 months (14 women); (2) only cystic lesions (one woman); (3) no detectable enhancing lesion at dynamic contrast-enhanced (DCE)-MRI (15 women); and (4) breast implants (four women). MRI results were corroborated by histopathology or imaging follow-up. Apparent diffusion coefficients (ADCs) were estimated for lesions and normal glandular tissue. Differences in the ADC between tissue types were evaluated and the sensitivity and specificity of the method calculated by receiver operating characteristics (ROC) curves. RESULTS: The final cohort comprised 53 patients with 59 lesions. Histopathology was obtained for 58 lesions. One lesion was validated as benign on imaging follow-up. Mean ADCs of 1.99 0.27 103 mm2/s, 1.08 0.25 103 mm2/s, and 1.74 0.35 103 mm2/s were obtained for normal tissue, malignant, and benign lesions, respectively. Mean ADCs of malignancies were significantly lower than those of benign lesions (p < 0.001) and normal tissue (p < 0.0001). The sensitivity and specificity for stratifying lesions, considering an ADC threshold of 1.41 103 mm2/s, were 94.3% and 87.5%, respectively; accuracy was 91.5%. CONCLUSION: DWI proved useful for the detection and characterization of breast lesions in the present sample. ADC values provide a high diagnostic performance for differentiation between benign and malignant lesions. Ó 2013 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.
Introduction Breast cancer is among the most common diseases affecting women worldwide, carrying a high mortality rate.1 Early detection and treatment may increase survival and * Guarantor and correspondent: L. Nogueira, MRI Unit, Department of Radiology, Hospital de S~ ao Jo~ ao, Alameda Prof. Hernani Monteiro, 4200-319 Porto, Portugal. Tel.: þ351 914 117 547. E-mail addresses: [email protected], [email protected] (L. Nogueira).
improve quality of life,2 which is why diagnostic accuracy is critical. Magnetic resonance imaging (MRI) has emerged as an important method for breast cancer detection, based on its high sensitivity for detection and characterization of breast disease.3 Currently, it is the method of choice to detect invasive carcinoma, and it is routinely used for screening women at high risk of breast cancer.4 Although the sensitivity of MRI for breast cancer detection is in the range of 85e100%,5,6 it presents a specificity ranging from only 37e88%.7,8 Various studies have investigated ways to overcome this limitation, and the use of diffusion-weighted imaging (DWI) has shown
ˇ
0009-9260/$ e see front matter Ó 2013 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.crad.2013.11.005
L. Nogueira et al. / Clinical Radiology 69 (2014) 378e384
promise.9e11 This technique focuses on the random motion of water molecules within tissues of the human body. Changes in mobility are reflected as differences in the signal intensity (SI) and can be quantified using the apparent diffusion coefficient (ADC), expressed as 103 mm2/s.12 Different studies have used DW-MRI to examine breast tissue, although mostly at 1.5 T.13e15 In keeping with histological features, ADCs of malignant and benign lesions have been shown to differ, with values comparatively lower for malignant lesions.9,10,16,17 However, few prospective quantitative studies have been undertaken using 3 T equipment, which may provide added benefits. The increased signal-tonoise ratio (SNR) at 3 T enables increased spatial resolution, allowing the use of higher diffusion sensitization b-values with an adequate SNR. The present study was conducted to evaluate the clinical performance of DW-MRI at 3 T in the characterization of breast lesions. The aims of the present study were as follows: (1) to determine mean ADCs of malignant and benign lesions, as well as normal glandular tissue; (2) to establish a threshold ADC for differentiating between malignant and benign lesions; and (3) to evaluate the diagnostic performance of DWI in breast lesion characterization.
Materials and methods This prospective study was conducted in women with a clinical indication for breast MRI between September 2010 to April 2011. The study was approved by the institutional review board, and written informed consent was obtained from all patients after the procedure had been explained. All patients referred for MRI during this period and who agreed to participate were enrolled in the study, resulting in 87 patients being enrolled. The referred group included women who underwent breast MRI for preoperative staging, high-risk screening, suspicious lesions on mammography and/or ultrasound and therapeutic monitoring. The following exclusion criteria were applied to patients: (1) breast surgery, radiotherapy and/or chemotherapy within the prior 48 months (14 women); (2) with only cystic lesions (one woman); (3) without detectable enhancing lesions at dynamic contrast-enhanced (DCEMRI) (15 women); and (4) breast implants (four women). Inclusion criteria for lesion DWI analysis were: (1) a minimum size of 0.7 cm at DCE-MRI; (2) definitive diagnosis via histology (core needle biopsy or excised at surgery), or a minimum 2-year follow-up using mammography, ultrasound, or MRI examinations. For premenopausal women, MRI examination was performed in the second week of the menstrual cycle. When breast MRI examination was performed after biopsy, a minimum time interval of 8e10 days between biopsy and MRI was considered to reduce the potential for signal increase due to haemorrhage.
Acquisition protocol MRI examinations were performed using a 3 T MRI machine (TimÒ Trio, Siemens, Erlangen, Germany). A dedicated
379
four-channel breast radiofrequency (RF) phased-array coil was used. Patients were examined in a resting prone position. The study protocol included the following sequences: bilateral axial T2-weighted (W) turbo spin-echo (TSE); sagittal unilateral T1W three-dimensional (3D) fast low angle shot (FLASH) without fat suppression; unilateral sagittal T2W TSE sequence using short time inversion recovery (STIR). Unilateral sagittal DWI images were obtained separately for each breast, with free breathing, using a single-shot spinecho echo planar imaging (SS-SE-EPI) sequence with STIR for fat suppression. Sensitizing diffusion gradients were applied sequentially in the x-, y-, and z-directions, with bvalues of 50, 200, 400, 600, 800, 1000, 2000, and 3000 s/ mm2 In this protocol, eight b-values with a minimum echotime (TE) of 108 ms, to include ultra-high b-values (2000 and 3000 s/mm2). These were used to explore non-Gaussian models of diffusion, to evaluate whether these can provide more information on lesion characterization (results to be reported separately). For healthy fibroglandular tissue at 3 T, a transverse relaxation time T2 of 71 6 ms has been reported.18 Using higher b-values meant prolonging the achieved TE. To compensate for this signal decrease, the number of excitations was increased to 3. DCE images were obtained with a T1W 3D FLASH sequence using spectrally adiabatic inversion recovery (SPAIR) for fat suppression; six dynamic phases were acquired using automatic injection of gadobenate dimeglumine (MultiHance; Bracco) via an antecubital vein at a rate of 2 ml/s (total dose 0.1 mmol/kg body weight), followed by a 20 ml saline flush. After the administration of contrast material, unilateral sagittal fat-suppressed (water excitation) T1W 3D FLASH images were repeated. Table 1 presents the imaging sequences in sequential order and their main parameters of acquisition. Total imaging time was approximately 40 min.
Image analysis All images were analysed using an image-processing platform [Siemens Medical Systems, Work in progress (WIP) V.17A]. The same radiologist, experienced in breast imaging (6 years), performed conventional MRI interpretations. Data were evaluated according to morphological and kinetic criteria described in Breast Imaging Reporting and Data System e Magnetic Resonance Imaging (BIRADS e MRI).19 Sizing of lesions was determined from dynamic images, using the ruler function. The longest dimension was considered. Two researchers retrospectively analysed the DWI-MRI images in consensus. The readers were blinded to the final diagnosis (histological results) and regions of interest (ROI) were marked according to the conventional MRI report descriptions. At DWI-MRI, both lesions and uninvolved breast parenchyma were evaluated quantitatively, using as the reference the visual inspection of T1- or T2W morphological, dynamic and contrast-enhanced images. ROIs were drawn on diffusion images (b ¼ 400 s/mm2), and then copied to the other DWI images and ADC map. The ROI sizes were set to 0.10 cm2 for lesions and normal tissue, and
380
L. Nogueira et al. / Clinical Radiology 69 (2014) 378e384
Table 1 Acquisition protocol for conventional magnetic resonance imaging (MRI) and diffusion-weighted imaging (DWI) pulse sequences. Parameters Sequence Orientation TR/TE (ms) TI (ms) Fat suppression FOV (mm2) Matrix Section thickness Number of sections NEX Bandwidth (Hz/pixel) Scan time (min) b-Values (s/mm2)
Conventional pre-contrast T2W TSE Axial bilateral 4990/88 _ _ 320 320 512 384 4 26 2 305 2:06 _
T1W 3D Flash Sagittal 17/4.9 _ _ 200 200 275 384 2 64 1 430 3:49 _
T2W TSE Sagittal 4920/67 210 STIR 200 200 448 314 4 26 2 248 4:26 _
DWI STIR
Dynamic
Contrast enhanced
Single-shot EPI Sagittal 4900/108 240 STIR 250 250 84 128 5 16 3 1628 5:58 50, 200, 400, 600, 800, 1000, 2000 and 3000
T1W 3D FLASH Axial bilateral 3.77/1.42 _ SPAIR 320 320 358 448 0.9 160 1 490 4:32 _
T1W 3D FLASH Sagittal 7.8/3.9 _ Water excitation 160 160 256 256 0.9 144 1 450 3:12 _
EPI, echo-planar imaging; FOV, field of view; NEX, number of excitations, SPAIR, spectrally adiabatic inversion recovery; STIR, short tau inversion recovery; T1 3D FLASH, three-dimensional gradient-echo fast low angle shot; TI, inversion time; TR/TE, repetition time/echo time; TSE, turbo spin echo.
the mean value and standard deviation (SD) of the SI were calculated. In the case of lesions, ROIs were drawn so as to include the area of highest hyperintensity. Normal glandular tissue was evaluated in women with both breasts and unilateral lesions. In these cases, measurements were performed also in the contralateral side. To attempt the same location for ROI placement, the ROIs were defined on the sections that included each nipple, and in regions without abnormal enhancement. The mean ADC value within the ROI was calculated using a mono-exponential fit with six b-values (50, 200, 400, 600, 800, and 1000 s/mm2), using the software provided by the system manufacturer. As this is the only analysis model available on typical machines, the two high b-value images were excluded from the ADC quantification, and used for another study.
Statistical analysis Sample characteristics were analysed accounting for patient’s age and lesion size. Mean ADCs for benign, malignant lesions, and normal tissue were calculated. Differences in mean ADCs between normal, benign, and malignant tissues were then further assessed using Student’s t-test. The ADC cut-off values for benign and malignant lesions were calculated using as a reference histological and imaging results. Sensitivity and specificity calculations incorporated the threshold for maximum sensitivity and specificity rendered by the receiver operating characteristics (ROC) curve. Lesions with mean ADCs below threshold were scored as malignant. All computations were performed using the PASW Statistics V17 software. Statistical significance was set at p < 0.05.
Results Subjects and lesion characteristics The final cohort included 53 women (26e82 years, mean age SD was 52.1 12.4 years). Fifty-nine lesions were
detected in both conventional and DWI-MRI datasets. Fiftyfive lesions (93.2%) could be seen as distinct masses and four (6.8%) behaved like non-masses on DCE imaging. The mean size of benign lesions was 12.91 5.85 mm (7e30 mm), while that of malignant lesions was 21.48 10.6 mm (7e60 mm). Histopathological confirmation was obtained for 58 lesions via excisional surgery and/or needle core biopsy, identifying 35 (59.3%) as malignant (5 ductal carcinoma in situ; 1 lobular carcinoma in situ; 20 invasive ductal carcinoma; 9 invasive lobular carcinoma) and 23 as benign (13 fibroadenomas; 4 epithelial proliferative lesions; 2 papillomas; 4 fibrocystic change). One additional lesion was presumed to be benign, based on morphological features and dimensional stability, as observed by mammography and ultrasound during a 2-year follow-up, leading to a total of 24 (40.7%) benign lesions. Fig 1 illustrates MRI images of a 38-year-old woman with malignant tumour (invasive ductal carcinoma) including different types of images: dynamic acquisition (Fig 1a), post-contrast T1W 3D FLASH (Fig 1b), DWI with a b-value of 1000 s/mm2 (Fig 1c), and mean ADC map (Fig 1d).
Quantitative evaluation of normal glandular tissue and lesions The mean ADC of normal glandular tissue was 1.99 0.27 103 mm2/s. In women with only one lesion, the mean ADC of normal glandular tissue did not differ significantly depending on lesion type (malignant versus benign, p ¼ 0.58). Table 2 depicts descriptive ADC analysis for normal glandular tissue, and benign and malignant lesions. The mean ADC of the malignant breast lesions was significantly lower than that of benign lesions (p < 0.001) and also compared to normal glandular tissue (p < 0.0001). Mean ADCs for both types of lesion are presented in Fig 2. Two of the malignancies evaluated showed mean ADCs that were 1.5 times beyond the interquartile range (outlier values): one of the lesions had an ADC below the first quartile (Q1 ¼ 0.91), whereas the ADC of the second lesion was above the third quartile (Q1 ¼ 1.27).
L. Nogueira et al. / Clinical Radiology 69 (2014) 378e384
381
Figure 1 A 38-year-old woman with suspected malignant lesion in the right breast. (a) Dynamic contrast-enhanced acquisition image shows a spiculate mass with heterogeneous enhancement. (b) Sagittal post-contrast high-resolution fat-suppressed T1W 3D FLASH. (c) DWI with a bvalue of 1000 s/mm2 image shows a hyperintense lesion. (d) The respective mean ADC map where the lesion shows hypointense signal and a central area of necrosis. Histological diagnosis: invasive ductal carcinoma, grade III.
The lowest mean ADC (0.46 0.25 103 mm2/s) corresponded to an invasive lobular carcinoma (20 mm) and the highest mean ADC (1.76 0.25 103 mm2/s) to an invasive ductal carcinoma (20 mm).
both histopathological results and imaging criteria and using ROC curve analysis (Fig 3). The area under the curve was 0.94 (0.87e1), enabling a mean ADC value of 1.41 103 mm2/s to be used for discriminating malignant
Diagnostic performance of ADC measurements To assess the accuracy of DWI for lesion characterization, the mean ADC threshold value was calculated, attending to Table 2 Descriptive apparent diffusion coefficient (ADC) statistics for normal glandular tissue, and benign and malignant lesions. ADC values (103 mm2/s)
Mean Standard deviation 95% Confidence intervals Median Minimumemaximum value
Normal tissue (n ¼ 42)
Benign lesions (n ¼ 24)
Malignant lesions (n ¼ 35)
1.99 0.27 1.89e2.08 1.98 1.56e2.72
1.74 0.35 1.59e1.89 1.80 1.05e2.3
1.08 0.25 0.99e1.16 1.09 0.46e1.76
Figure 2 Box plot show distributions of mean ADC (103 mm2s) calculated from six b-values (50e1000 s/mm2) for benign and malignant lesions.
382
L. Nogueira et al. / Clinical Radiology 69 (2014) 378e384
Figure 3 ROC curve used to differentiate benign and malignant lesions according to mean ADC, calculated with all b values. The area under the ROC was 0.94 (0.87e1). An ADC cut-off value of 1.41 103 mm2/s corresponded to a sensitivity of 94.3% and specificity of 87.5%.
versus benign lesions. Mean ADCs below this threshold were considered to correspond to malignant lesions. With this threshold, 12.5% (3/24) of benign lesions (three fibroadenomas with mean ADCs of 1.05 103 mm2/s, 1.12 103 mm2/s, and 1.27 103 mm2/s) and 5.7% (2/35) of the malignancies (one invasive lobular carcinoma with mean ADC 1.43 103 mm2/s and one ductal carcinoma in situ with mean ADC 1.76 103 mm2/s) would have been mislabelled. Sensitivity and specificity values of 94.3% and 87.5% were obtained. The positive predictive value was 91.7% and the negative predictive value was 91.3%. Altogether, 91.5% of the cases were correctly diagnosed.
Discussion Water diffusion in biological tissues is a process that reflects exchanges of different components between intraor extracellular compartments and vascular space. DWI focuses on the mobility of water molecules in the tissues, reflecting higher or lower mobility at a cellular level.12 In normal tissues, water movement is conditioned by cell membranes and macromolecules. When pathological processes are present, alterations in cellularity, tortuosity, and architectural changes occur. In more compact tissue, these changes cause reduction of the normal configuration of the extracellular space, leading to hindered or restricted water diffusion.20 The impact of these changes can be visualized via the ADC. This measurement parameter represents an indirect index of tissue complexity, assuming that malignant lesions have a more complex microstructure than benign or normal tissues. Previous studies have shown that breast lesions may be differentiated by mean ADC values with malignancies yielding lower values when compared to benign or normal tissues.14e16 Despite some differences in study protocols, the present results corroborate previously reported mean ADCs for normal glandular tissue and for malignant and benign lesions,11,17,21,22 confirming the feasibility of DWI-MRI to
discriminate between normal and diseased states. Mean ADCs of malignancies were significantly lower than those of benign lesions, as a likely consequence of differences in tissue microstructure (i.e., cell density) resulting in reduced extracellular space.23 Mean ADCs of 1.99 0.27 103 mm2/ s for normal glandular tissue, 1.74 0.35 103 mm2/s for benign lesions, and 1.08 0.25 103 mm2/s for malignancies were obtained; all of which were similar to absolute ADCs cited by other publications.21,24,25 Cystic lesions were excluded from the present study as they show high ADCs; their inclusion in the benign group would have biased the ADC value of benign lesions towards higher values, thus affecting the overall ADC quantification and interfering with the aim of this study. Although the present data are comparable to prior results obtained at 1.5 and 3 T, it is possible that the mean ADCs for malignancies may have been overestimated. The reason for this is that many lesions were examined (26/59) after biopsy, which could have altered their mean ADC due to the presence of residual haemorrhagic components. Nevertheless, Rahbar et al.26 reported no significant difference in mean ADC values before and after biopsy in ductal carcinoma in situ. However, further studies would be required to confirm whether this result is also applicable to other histological subtypes of breast carcinoma. In theory, using a higher number of b-values lowers the uncertainty associated with estimated ADC values, in comparison with only two b-values.20 As ADC estimates can depend on the number of b-values used, cut-off values for lesion classification could depend on this parameter. However, others groups have shown similar results when two or more b-values were used for ADC estimation.21,22 Establishing a threshold value for lesion classification (malignant versus benign) greatly affects the sensitivity and specificity of DWI. In a recent meta-analysis,27 threshold values ranged from 1.1 103 mm2/s to 1.6 103 mm2/s, placing the present mean ADC threshold of 1.41 103 mm2/ s within the acceptable range of variation. With this mean ADC threshold value, a sensitivity of 94.3% and a specificity of 87.5% were achieved. For the present study, accuracy was 91.5%, which is slightly higher than that of other studies ranging from 83.6e91%.11,13,28 Preliminary evidence suggests that the intense cellularity of some benign lesions may produce false-positive results.13,15,29 In the present study, three false-positives cases were found (all fibroadenomas). However, only these three lesions were misclassified from a total of 13 fibroadenomas. A possible explanation for the false-positive cases is a presence of a certain amount of fibrotic tissue that may restrict water movement, resulting in lower ADC values. These findings were reported by other groups22,30 and are related to the variable spectrum in tissue composition and degree of cellularity, which can be found for this type of lesion. Falsenegative results included one ductal carcinoma in situ and one extensively spread invasive lobular carcinoma. Possible explanations for these false-negative cases could be related to partial volume effects, the section thickness used (5 mm), or the inadvertent inclusion of areas of necrosis in the ROI, which caused ADC elevation.
L. Nogueira et al. / Clinical Radiology 69 (2014) 378e384
Lesion detection based on DWI is dependent on diffusion sensitization b-values, as well as lesion size and histological composition.31 The present study confirms the ability of the DWI sequence to distinguish breast lesions from surrounding tissue, depicting all as hyperintense areas in all the cases. Nevertheless, some authors have found malignant lesions that were not visible on DWI sequences (6e37.5%),24,30,32 which can be attributed to the protocol used, regarding the b-values selected, and/or the inherent histology of the lesions. Others have suggested that the pattern of non-mass lesions is less easily detected. In the present study, there were four such cases. Conceivably, this low figure of non-mass lesions may have biased lesion detection in the present results. Although conditioned to the present criteria for lesion analysis, others small lesions that were described in the conventional MRI report were also depicted in DWI images. It is expected that DWI at 3 T should enable better detection and characterization of lesions by providing a higher SNR and greater spatial resolution. Matsuoka et al.33 reported improved visualization at 3 versus 1.5 T for lesions
