Differentiation of clinically benign and malignant breast lesions using diffusion-weighted imaging




To evaluate the value of diffusion-weighted imaging (DWI) in distinguishing between benign and malignant breast lesions.

Materials and Methods

Fifty-two female subjects (mean age = 58 years, age range = 25–75 years) with histopathologically proven breast lesions underwent DWI of the breasts with a single-shot echo-planar imaging (EPI) sequence using large b values. The computed mean apparent diffusion coefficients (ADCs) of the breast lesions and cell density were then correlated.


The ADCs varied substantially between benign breast lesions ((1.57 ± 0.23) × 10−3 mm2/second) and malignant breast lesions ((0.97 ± 0.20) × 10−3 mm2/second). In addition, the mean ADCs of the breast lesions correlated well with tumor cellularity (P < 0.01, r = −0.542).


The ADC would be an effective parameter in distinguishing between malignant and benign breast lesions. Further, tumor cellularity has a significant influence on the ADCs obtained in both benign and malignant breast tumors. J. Magn. Reson. Imaging 2002;16:172–178. © 2002 Wiley-Liss, Inc.

DIFFUSION IS THE TERM USED TO describe the microscopically visualized thermally induced behavior of molecules moving in a random pattern. This so-called Brownian movement is quantified by means of an apparent diffusion coefficient (ADC) that depends largely on the presence of barriers to diffusion within the water microenvironment, namely, cell membranes and macromolecules. Consequently, compartments within different cellular structures may exhibit dissimilar ADCs and hence be distinctly identified on ADC maps.

Magnetic resonance imaging (MRI) mammography has recently evolved as an extremely promising tool in the differential analysis of breast lesions. Further, it has also become evident that Gadolinium (Gd)-enhanced MRI depicts breast cancer with exquisite sensitivity. In addition, with the exception of occasional cases of isolated ductal carcinoma in situ (DCIS), virtually all pathologically proven breast cancers have been detected in several large trials (1–3). As expected, the classification of breast lesions obtained by combining three independent classes of features (boundary descriptors, uptake parameters, and texture features) has shown a larger increased specificity than ever before documented (4). While this is so, a significant overlap in the enhancement patterns of malignant breast lesions and fibroadenomas still exists and is therefore an area of major concern.

Several studies presently available in the literature reveal that the measurement of ADCs has significant potential in the characterization of liver and brain lesions (5–9). In particular, preliminary studies that measured the ADCs in both the normal breast and in breast lesions (10–12) demonstrated that benign pathology and normal breast tissue have larger ADCs than those obtained from malignant breast tumors. These findings suggest that the measurement of extracellular water content may be an additional feature that could further increase the specificity of the classification of breast lesions. Such studies involving diffusion-weighted imaging (DWI) have already been performed with small b values (234 seconds/mm2), though they include only a few cases. Interestingly, a significant correlation has also been demonstrated between the ADCs of viable tissues and cell density in brain tumors and in human melanoma xenografts (9, 13). While it is therefore very likely that cell density might play an important role in the different ADCs obtained from benign and malignant breast lesions, there has been no experimental evidence to date in support of this concept.

In the present study, we performed DWI of the breast with a single-shot echo-planar imaging (EPI) sequence using large b values, calculated the mean ADCs of the breast lesions, and compared the derived ADCs with cell density. The aims of this study therefore were 1) to determine the ADCs for benign and malignant breast lesions, 2) to evaluate the ability of ADCs in distinguishing benign from malignant breast lesions, and 3) to assess the impact of cellularity on different ADCs obtained from benign and malignant breast lesions. To the best of our knowledge, this represents the first study that compares the histopathological grading with measures of diffusion to predict the nature of visualized breast lesions.



Fifty-two females (mean age = 58 years, age range = 25–75 years) with 55 breast lesions in all were included in the present study. There were 24 benign lesions (22 patients), including 18 fibroadenomas, 1 intraductal papilloma, 1 ductal ectasia, and 4 cysts (1 patient with 3 cysts). The mean size for solid benign lesions was 2.64 cm (range = 0.5–9 cm), and the mean size for breast cysts was 1.6 cm (range = 0.7–2.1 cm). There were 31 malignant lesions (30 patients), including 25 infiltrating carcinomas, 2 cases of DCIS, 3 medullary carcinomas (1 patient with 2 lesions), and 1 scirrhous adenocarcinoma. The mean size for malignant lesions was 2.75 cm (range = 0.4–6 cm). The final diagnosis was established on the basis of histopathological examination results of surgically excised specimens in 47 lesions, histological examination results of needle biopsy specimens in 5 lesions (1 infiltrating carcinoma and 4 adenomas), and breast US in 3 cysts. Benign lesions, which were diagnosed by needle biopsy or breast ultrasonography, were followed up for more than 12 months. Cases with breast adenosis were excluded from the final data analysis because the lesions were not visible on DWI.


All the breast MR examinations were performed with a 1.5-T clinical MR system (Signa Horizon, General Electric Medical Systems, Milwaukee, WI). A GPFlex surface coil was utilized and patients lay in the prone position throughout the duration of the scan. Prior to DWI, multisection T2-weighted spin-echo imaging of both the breasts was performed in the transverse plane in each participant that was aimed at detecting and selecting bilateral breast lesions considered suitable for DWI. T2-weighted images were obtained with a fat suppression technique that used the following imaging parameters: TR/TE= 2000/100 msec, acquisition matrix = 256 × 256, NEX = 2, field of view (FOV) = 32 cm, slice thickness = 5 mm, and interslice gap = 1 mm.

Diffusion-weighted MR images were acquired using a multisection spin-echo-type single-shot EPI sequence in the sagittal plane. Sensitizing diffusion gradients were applied sequentially in the x-, y-, and z- directions with b values of 0, 250, 500, 750, and 1000 seconds/mm2 for 10 patients, and 0 and 1000 seconds/mm2 in the remaining 42 subjects. Sequential sampling of the k-space was used with the following parameters: TE = 99 msec, bandwidth = 60 kHz, acquisition matrix = 128 × 128, FOV = 24 cm, section thickness = 5 mm, interslice gap = 1mm, and NEX = 1. Both breasts were imaged within a single DWI sequence of 40 seconds duration. In order to locate the solid portion of the lesions accurately, Gd-enhanced Enhanced Fast Gradient Echo (EFGRE) 3D scans were obtained with a breast coil after DWI. Further, as a control measure aimed at the validation of our MRI system and the pulse sequence we used for measuring ADCs, phantom studies of bottles filled with water and acetone were conducted, and the ADCs for those two substances were calculated as well.

Analysis of ADCs

All the ADCs were determined with a liner regression analysis of the natural log of the signal intensity vs. the gradient factor b (14). To measure the signal intensity of the lesions, a region of interest (ROI) that was minimally smaller than the actual solid portion of the breast lesion was carefully placed to ensure that cystic or necrotic areas were not included, and the mean ADCs were then obtained. In those cases with a breast cyst, an ROI drawn that was slightly smaller than the actual cyst was considered acceptable. Mean ADCs for each group (benign and malignant breast lesions) were then calculated and compared with each other (benign vs. malignant) using a t-test. A P value of less than 0.05 was considered significant.

Analysis of Tumor Cellularity

Analysis of tumor cellularity was performed by utilizing a method essentially similar to the one used by Sugahara et al (9). Initially, five slides (counterstained with hematoxylin-eosin) were obtained from different regions of the lesion in each tumor. Following this, two FOVs (original magnification × 200) were randomly chosen from each slide and photographed for the eventual analysis of tumor cellularity. The films of the specimens were then scanned into a personal computer. Adobe Photoshop 4.0 software (1996 Adobe Systems Incorporated) on the desktop was then opened, and the scanned color film was converted into gray scale by setting the image mode to gray scale. The threshold was next defined as the intensity value below which the image would be categorized as tumor cell nuclei. Tumor cell nuclei were identified based on their low signal intensities. However, as a precaution to minimize any individual variations that may have occurred in their assessment, the intensity value below which the tissue would be categorized as tumor cell nuclei was carefully chosen thrice with each film. Finally, the mean tumor cellularity was calculated according to the predefined threshold by dividing the total area of tumor cell nuclei by the area of the histology section (film). Forty-four lesions (47 surgically excised specimens, except 1 cyst, 1 DCIS, and 1 small adenoma not visible on DWI) underwent analysis for tumor cellularity.


Comparisons between tumor cellularity and the mean ADCs were performed using simple linear regression analysis, and a P value of less than 0.01 was considered to indicate statistical significance. To determine whether the ADCs measured in breast tumors helped in the differentiation of benign breast lesions from malignant ones, a one-sided upper limit of 95% permissible interval of ADC (mean + 1.66 × SD) was adopted as the point of separation between malignant and benign breast lesions.


Phantom Studies

The calculated ADCs of water and acetone were 2.29 × 10−3 mm2/second and 3.98 × 10−3 mm2/second, respectively. These values are consistent with those previously reported in the literature (14–15).

Feasibility of Measurement of ADCs in Lesions on DWI

Of the 20 solid benign breast lesions studied, 3 adenomas that were less than 1 cm in diameter were not demonstrable on DWI (with b values of 0 and 1000). On the other hand, of the 31 malignant breast lesions evaluated, only 1 case of DCIS, 4 mm in diameter, was not visualized (with b values of 0, 250, 500, and 1000). In general, malignant breast tumors demonstrated a higher signal intensity on DWI (Figs. 1 and 2).

Figure 1.

a: An axial contrast-enhanced image of a breast. A strong enhanced nodule with smooth margin is visible in the breast. b: Saggital DWI (b = 1000 seconds/mm2) of the breast. The nodule showed obviously high signal intensity. ADC = 0.74 × 10−3 mm2/second. c: Histological specimen of the nodule demonstrated a high cellularity (histological cellularity = 15.4%). The nodule was a medullary carcinoma of the breast.

Figure 2.

a: An axial contrast-enhanced image of a breast. A strong enhanced nodule with smooth margin is visible in the breast. b: Saggital DWI (b = 1000 seconds/mm2) of the breast. The nodule showed a slightly higher signal intensity than the glandular tissue. ADC = 1.47 × 10−3 mm2/second. c: Histological specimen of the nodule demonstrated a low cellularity (histological cellularity = 5.1%). The nodule was a fibroadenoma of the breast.

Comparison of ADCs in Benign and Malignant Breast Lesions

The mean ADC obtained from malignant breast lesions was statistically different from that observed with solid benign lesions (t = 9.526, P = 0.000) (Table 1). The mean ADC of four cystic lesions was (2.35 ± 0.08) × 10−3 mm2/second. Figure 3 demonstrates the scatterplots of the observed ADCs. An overlap still existed between the ADCs of malignant and benign breast lesions. If the one-sided upper limit of 95% permissible interval of malignant ADCs (mean + 1.65 × SD = 1.30 × 10−3 mm2/second) is used as a threshold, the following diagnostic indices emerge: sensitivity, 93% (28/30); specificity, 88% (15/17), and overall accuracy, 91% (43/47). According to the diagnostic criteria employed here, two benign lesions (a duct ectasia and an intraductal papilloma) and two malignant lesions (a lobular-scirrhous carcinoma and a DCIS) were misclassified.

Table 1. The Distribution of ADC in Breast Cancers, Benign Solid Lesions, and Cysts
 Age (years)Size (cm)ADC (10−3mm2/s)
Breast cancers50.3 ± 11.62.75 ± 1.180.97 ± 0.20
Benign solid lesions38.3 ± 7.12.64 ± 2.191.57 ± 0.23
Cysts54.5 ± 5.01.60 ± 0.622.35 ± 0.08
Figure 3.

Scatterplots of the ADCs obtained in benign and malignant breast lesions and cysts.

Influence of Tumor Cellularity on the Mean ADCs

The relationship of tumor cellularity with the mean ADC is well demonstrated by Figure 4. The mean ADC of the breast lesions (both benign and malignant) correlated well with tumor cellularity (P = 0.000, r = −0.528).

Figure 4.

Correlation of tumor cellularity vs. the ADCs (P < 0.01, r = −0.542).


Breast cancer is now a significant cause of worldwide morbidity and mortality. Further, the increasing rate of breast cancer continues to be a major area of concern for both clinicians and researchers. Heightened awareness in the affected population leading to more frequent physical examinations and diagnostic imaging procedures can be expected to result in an earlier diagnosis and hence improved prognosis.

Breast MRI is now increasingly used to accurately diagnose both primary and recurrent breast cancers, particularly in cases in which mammography and breast US are inconclusive or yield discrepancies (16–18). Of perhaps even greater significance is the fact that breast MRI may improve the local staging of breast cancer by revealing multifocal tumor growth in patients scheduled for conservative breast surgery (19). While the excellent sensitivity of breast MRI has proven particularly advantageous in the preoperative patient, its limited specificity continues to be a significant problem, particularly in patients referred for further clarification and delineation of an inconclusive finding obtained by conventional breast imaging (20).

To overcome problems in breast MRI and to improve its specificity, two strategies have now evolved: high spatial resolution and dynamic MRI. In the high-spatial-resolution approach, the morphology of breast lesions is evaluated on fat-nulled, 3D volume images (1, 21, 22). In dynamic breast MRI, on the other hand, the signal intensity of lesions is evaluated as a function of time with imaging that is rapidly repeated during intravenous administration of a bolus of contrast material (4, 23–25). Despite these significant advances, both approaches continue to have their drawbacks in the accurate classification of breast lesions. The boundary characteristics of malignant breast lesions (spiculated and irregular) generally differ from those of benign breast lesions (smooth and lobular). However, the resolution of breast MR images may not clearly distinguish these two types of boundaries. Further, an overlap in the enhancement properties of some benign and malignant breast lesions (attributed to increased vasculature) may also result in an inaccurate differentiation and therefore improper classification. Combining morphological and pharmacokinetic methods has led to an increasing specificity (4), but with attendant compromises pertaining to both spatial and temporal resolution.

Our results in the present study show that the ADC is useful in differentiating benign from malignant breast lesions. This finding suggests that the measurement of extracellular water content could provide an additional feature that may further increase the specificity of the classification of pathological breast lesions.

DWI has been primarily limited to the central nervous system because many technical problems (mostly related to gross physiologic motion) affect it (26, 27). EPI techniques have now been developed that have made DWI of the abdomen and breast feasible because their fast imaging capability minimizes the effect of gross physiologic motion (5–7, 10–12, 14). Although image distortion due to the susceptibility associated with single-shot EPI tends to become more amplified with a larger b value, DWI without breath holding is acceptable when the maximum b value of 1000 is used.

According to the diagnostic criteria employed here, all the fibroadenomas and invasive carcinomas were appropriately classified. These results indicate that ADCs would be effective in distinguishing between fibroadenomas and invasive carcinomas, while duct ectasia and intraductal papilloma, which were misclassified by ADC here, could be recognized by their obvious cystic component. Scirrhous carcinoma and DCIS (extremely troublesome to classify based on their morphologic properties and enhancement profiles alone) were still difficult to correctly classify by DWI.

Most of the malignant tumors showed higher signal intensity on DWI, compared with benign breast lesions. One DCIS that was 4 mm in diameter was not visualized with DWI. Ductal ectasia and patchy form enhancement were identified on T2-weighted and Gd-DTPA-enhanced MRI, respectively, in this case. Three adenomas that were less than 1 cm in diameter were not demonstrable with DWI. We believe that this was so because of the poor resolution obtained with DWI, the only slightly higher signal of some fibroadenomas, compared with the normal breast parenchyma.

In biological tissues, microscopic motion includes both the molecular diffusion of water and blood microcirculation in the capillary network. Both diffusion and perfusion thus affect the ADCs obtained from biological tissues. The pseudodiffusion coefficient of capillary microcirculation is typically many times greater than the diffusion coefficient of pure water. However, the volume of blood flowing in the perfused capillaries is only a small percentage of the total water content in normal brain tissue (28). In breast tumors, although higher microvessel counts were recorded for malignant than for benign pathology (29), they showed lower mean ADCs than did benign lesions. This suggests that molecular diffusion of water has a major impact on the ADCs seen in breast lesions.

Our results showed that the mean ADC of the breast tumors correlated well with tumor cellularity. Malignant breast tumors had a higher cellularity and a lower ADC than benign breast tumors. Hence, a malignant tumor with low cellularity, such as in the case of scirrhous adenocarcinomas in our study, may demonstrate a high ADC and consequently be misdiagnosed as benign in nature. Conversely, a benign tumor with high cellularity, such as in the case of papillomas in our study, may demonstrate a low ADC and be potentially misdiagnosed as malignant. In the case of ductal ectasia, both the solid and cystic parts demonstrated small ADCs. An outflow of white mucous from the cysts was observed during surgery, and a severe inflammatory reaction was found in the specimen. It should also be noted that because the ADCs are strongly affected by perfusion in the case of small b values and tend to be larger when measured only with small b values (30), large b values were used in our study.

In conclusion, the ADC would be an effective parameter in distinguishing between malignant and benign breast lesions. It should be well understood that DWI is not to be used as a stand-alone diagnostic criterion, but that it might be useful if integrated into the process of differential diagnosis of breast lesions. It is also clearly evident that to avoid nonvisualization of small breast lesions DWI should be performed in conjunction with contrast-enhanced MRI. Further, DWI seems most useful in the differential diagnosis of focal solid tumors. As expected, tumor cellularity plays a significant role in the different ADCs obtained from malignant and benign breast tumors. In summary, despite the obvious limitations of a significant age difference between the population groups of benign and malignant breast cancer patients, as well as the predominance of malignant breast pathology in this series, measuring the ADCs is still an effective method for characterizing focal breast masses.