Clinical Interpretation of Diffusion MRI, ROI Assessment, Common Errors, Pitfalls and Artifacts, Challenges in Acquisition

Clinical Sequences

When interpreting full diagnostic protocol MRI of the breast including DWI, it is recommended to review the MRI together with conventional imaging (mammograms, tomosynthesis, and whole breast or handheld ultrasound). Contrast-enhanced T1-weighted images (ideally including subtractions of unenhanced from enhanced images and maximum intensity projections (MIPs) are usually the first series to be reviewed, as they allow evaluation of the degree of background parenchymal enhancement and whether or not contrast-enhancing lesions can be identified on the scan. Once enhancing lesions are identified, they need to be further characterized by evaluating their morphology and their enhancement kinetics on dynamic-contrast enhancement. Ideally, kinetic maps are created automatically and overlaid on T1-weighted images to display the type of enhancement for each voxel reaching a set enhancement threshold (type 1, continuous enhancement; type 2, plateau enhancement; type 3, washout; type 3 is the most suspicious; see Chapter 2 ). Additional review of T2-weighted images is recommended to evaluate the signal intensity of enhancing lesions and to assess for nonenhancing lesions such as cysts. DWI is then reviewed for further lesion characterization. The recommended image review process is summarized in Table 13.1 .

Typically, the low and high diffusion sequences will be saved to the picture archiving and communication system (PACS), ready for clinical interpretation. Ideally the slice thickness and location will match the T2W sequence, although often the slice thickness will be greater than the T2W sequence. ADC will be calculated and the ADC maps saved onto the PACS. The quality of the DWI images should be assessed for artifact and movement and a decision made as to whether the images are interpretable. With this relatively new sequence, the DWI is often overlooked by the technicians and is sometimes not repeated when required. The radiologist starts with the high signal on a high b value image (with the greatest diffusion weighting) to try and identify abnormalities and correlate them with the other sequences. Generally, all high signal area in DWI greater than 5 mm should be interrogated. Using the ADC map, regions of lowest signal are also identified and further interrogated. For lesions identified on contrast-enhanced sequences or (if omitted as in unenhanced MRI protocols) on DWI, an ROI should be drawn within the boundaries of the lesion, either encompassing the largest possible region but avoiding any surrounding fat or fibroglandular tissue, or covering a small subregion with the lowest ADC. Meticulous comparison made with DCE images is recommended to ensure the lesion of concern is being included. One must be aware that the spatial position may not spatially map to the DCE sequence, as DWI sequences suffer from distortion, therefore requiring to pick up on anatomical features to make a match. The ROI can either be drawn on the high b value DWI images and then transferred onto the ADC map or on the ADC map directly. The ADC should be noted and if a marginal value is obtained then a repeat measurement should be undertaken.

Lesion Classification

DWI is increasingly used in the classification of suspicious breast lesions on MRI as an adjunct to DCE MRI to reduce false positive results and avoid unnecessary biopsies. In tumors, the tight cellular packing hinders or restricts diffusion, resulting in signal hyperintensity compared with normal breast tissue on high b value diffusion-weighted images, and significantly lower ADC values are measured in malignant breast tumors compared with benign lesions and fibroglandular tissue ( Fig. 13.2 ). A meta-analysis of the diagnostic performance of DWI using the ADC demonstrated a pooled sensitivity of 89% (95% CI 87–91) and specificity of 82% (95% CI 78–85) in 65 studies (6408 cancers in 5892 patients). A threshold ADC value can be used to classify lesions that appear hyperintense on DWI. These thresholds are often found empirically or using receiver operating characteristic curve analysis on a cohort of breast cancer patients. Reported ADC threshold values range from 0.87 × 10 ⁻³ mm ² /s to 2 × 10 ⁻³ mm ² /s across 65 studies. Variations in MRI technique and lack of standardization to date have prevented the establishment of a generalized ADC threshold. The choice of b value will affect the measured ADC, with ADC decreasing with increasing b value, though there is no evidence that choice of b value affects diagnostic performance. However, this suggests that ADC thresholds must be adjusted depending on chosen acquisition parameters or, better, that standardization fixes b values to be used so as to homogenize ADC cut-off values across protocols and sites. A sufficiently high ADC threshold can also be chosen to achieve 100% sensitivity, though this may lead to unnecessary biopsies. DWI should always be interpreted alongside supporting morphological and functional information (DCE MRI and unenhanced T2-weighted images) to enable accurate lesion classification. Example DWI images for a range of breast lesions are shown in Fig. 13.3 .

Morphological Assessment

According to the ACR BI-RADS, lesions can be classified as either foci, masses, or nonmass enhancement (NME). Masses can be ascribed descriptors for shape (round, oval, irregular) and internal signal pattern (homogeneous, heterogeneous, rim), and nonmass lesions can be described in terms of distribution (focal, regional, linear, segmental) and internal signal pattern (homogeneous, heterogeneous). The qualitative assessment of the location, size, and morphology of lesions is possible using DWI. However, the low spatial resolution of DWI limits morphological assessment compared with other high resolution anatomical and contrast-enhanced sequences. Morphology as assessed on DWI can be reported when it is discrepant with other sequences.

The majority of DWI literature has focused on enhancing masses. DWI is less sensitive for NMEs, with ROI placement significantly affecting the diagnostic accuracy of ADC measurements. NME lesions also have a lower conspicuity on DWI, with one study finding a third of lesions (29/94 NME lesions) could not be evaluated on DWI due to either nonvisibility or poor quality of DWI, suggesting that DCE MRI is required for the detection of all NME lesions.

A number of false positive and false negative findings on DWI can be attributed to the underlying tumor histology of different breast cancer histopathological subtypes. The most commonly reported examples of false positive lesions on DWI are complicated cysts and fibroadenomas, likely due to high cellularity (creating a more restricted environment for diffusion), fibrosis (providing boundaries or obstacles to the free diffusion of water), and chronic inflammatory elements. An inverse correlation has been found between ADC and the degree of fibrosis. Atypical ductal hyperplasia and intraductal papilloma are also often misclassified due to increased cellularity and duct ectasia, which has a low measured ADC. Diffusion can also be restricted in regions of high viscosity or intra- or extracellular edema, such as in mastitis, abscesses, and coagulated blood or proteinaceous debris within ducts and cysts.

On the other hand, mucinous carcinoma is a common false negative finding given that its measured ADC is significantly higher than other malignant tumors, due to the low cellularity relative to the abundant mucin and the high extracellular water content. Higher ADC values are also measured in papillary carcinomas due to the distribution of tumor cell batches within stromal spaces, allowing for more free diffusion in the interstitium, and scirrhous adenocarcinoma has a low cellularity and is often misdiagnosed. Lesions with extensive necrosis will also have a higher ADC due to more free diffusion in the necrotic core. Common false positive and false negative findings are summarized in Table 13.2 .

Problem-Solving

Whereas DCE-MRI is a highly sensitive technique for the detection of breast cancer, DWI is often used as an adjunct to contrast-enhanced MRI diagnosis of breast cancer to reduce false positive results, with a meta-analysis of 14 studies finding a combined sensitivity and specificity of 92% and 86%, respectively. Breast cancers are generally detected through the observation of suspicious enhancement after the administration of gadolinium contrast, identifying areas of increased vascularity and perfusion. However, benign lesions may also show suspicious enhancement on DCE MRI. The addition of complementary information regarding tissue cellularity provided by DWI can improve the ability of MRI to distinguish between malignant and benign breast lesions. Results from the multicenter ACRIN 6702 trial have shown that the use of ADC values from DWI can significantly reduce the number of benign biopsies prompted by breast MRI without reducing sensitivity.

DWI is also able to overcome some of the shortfalls of mammography. DWI has shown promise for detecting cancer in women with dense breasts for whom mammography has a reduced sensitivity and poor lesion visibility. In the detection of mammographically occult cancers, DWI has been shown to be superior to MRI-guided focused ultrasound and comparable to DCE MRI. Although DWI is increasingly investigated in common types of breast lesions, further investigation is needed for the types of lesions that are of interest for problem-solving where DCE MRI is inconclusive, including small lesions, multifocal and multicentric lesions, and NMEs (where no correlate can be found on mammograms or ultrasound, which require MRI guided biopsy).

Lymph Nodes

Early detection of axillary lymph node metastasis may improve breast cancer staging and selection of treatment. Sentinel node biopsy is current standard practice in assessing lymph node involvement with little morbidity compared with the more accurate surgical assessment, which is associated with morbidity including seroma, hematoma, lymphedema, and paresthesia.

On MRI, most normal lymph nodes will exhibit certain imaging features, such as fatty hila, uniform shape, and high T2 signal, which may enable a radiologist to identify them as benign. Abnormal lymph nodes can be enlarged either due to tumor involvement or reactive change. Loss of the kidney-bean oval shape to a rounded appearance, loss of the fatty hilum, and increased or irregular cortical thickness are classic signs of malignancy together with an increase in short axis diameter.

Although mean ADC values of metastatic lymph nodes have been shown to be significantly lower than those of nonmetastatic lymph nodes in a recent meta-analysis, and several studies reported high sensitivity and specificity of DWI in diagnosing metastatic axillary lymph nodes, contrasting results have been reported as well, with the diagnostic accuracy of DWI reported as inferior to T1- and T2-weighted sequences. These results indicate that further research is needed to evaluate the value of DWI for axillary lymph node assessment. At present, MRI does not match the accuracy of sentinel lymph node biopsy.

Unenhanced Screening

There is increasing interest in DWI as a stand-alone screening tool (see Chapter 6 ). DWI has been investigated independently for screening or for the noncontrast detection of cancers through blinded reader studies, reporting sensitivities >85% , or modest sensitivity (50%–77%) and high specificity >90%. , However, the performance of DWI is reduced for small cancers less than 1 cm in size. The in-plane spatial resolution of DWI (2 × 2 mm ² ) and slice thickness (3–5 mm) are limiting factors in the detection of these small lesions, as well as the limitations of the overall image quality of DWI. For DWI to be clinically useful in a screening setting, it must be able to detect and characterize a range of cancers at least comparable to that of contrast-enhanced MRI, particularly small, early stage cancers. Further evidence from larger prospective and multicenter studies in a true screening setting is required before the adoption of unenhanced DWI screening into clinical practice.

Monitoring Treatment Response

Neoadjuvant chemotherapy induces changes at a cellular level, significantly altering the histopathologic appearance of tumors (see Chapter 5 ). Studies investigating cellular changes to tumor specimens after chemotherapy found a significant reduction in tumor cellularity and increased nuclear atypia, with one study finding it difficult to distinguish between residual tumor cells and chemotherapy-induced atypia. Combined with a measurement of residual tumor size, the assessment of response using a measure of tumor cellularity can be more accurate than using tumor size alone.

Changes in tumor ADC as measured using DWI reflect these changes in cellularity. Response to treatment has been associated with changes in ADC as measured after one cycle, two cycles, at midtreatment, and at the end of treatment. Increases in ADC after chemotherapy have also been associated with the presence of necrosis and increasing cell lysis.

DWI can be a useful tool in the early assessment of response to therapy, as changes in ADC have been shown to occur before reduction in tumor size. Studies by Park and colleagues and Iacconi and colleagues found that tumors with lower pretreatment ADC and high cell density responded better to neoadjuvant chemotherapy. A meta-analysis by Chu and colleagues found a pooled sensitivity of 88% and specificity of 79% for ADC in the prediction of pathological treatment response (pCR) but with a higher pooled AUC when using the change in ADC during treatment than pretreatment ADC (0.80 vs. 0.63).