Is NME the Enemy of Breast DWI?

Dynamic contrast-enhanced (DCE) MRI of the breast has the highest sensitivity for lesion detection amongst breast imaging modalities, but also results in many benign biopsies. Enhancing lesions identified on DCE-MRI are characterized according to the BI-RADS guidelines¹ as a focus, mass or non-mass enhancement (NME; defined as areas of enhancement lacking features of a focus or mass). Diagnosis of NME lesions can be challenging as malignant NMEs are difficult to distinguish from background parenchymal enhancement and benign NME lesions¹. Most NMEs represent benign proliferative pathology, such as fibrocystic changes. However, up to one-third of NME lesions are malignant, commonly reflecting ductal carcinoma in situ or diffuse invasive cancers, such as invasive lobular carcinoma². With substantial overlap in morphological and DCE-MRI kinetic features between benign and malignant NME lesions, there currently are no clear criteria to de-escalate NMEs and obviate subsequent biopsy¹.

Over the past decade, diffusion-weighted imaging (DWI) has emerged as a supplemental imaging technique that can improve breast MR specificity and enable more accurate lesion diagnoses³. DWI is sensitive to the mobility of water and provides a quantitative measure of diffusion, the apparent diffusion coefficient (ADC). Numerous studies have demonstrated that malignant breast lesions exhibit lower ADC values compared to benign lesions^4,5, attributed to hindered diffusion in more cellular dense tissue. While DWI has shown promise to improve MR specificity, its performance for NME lesion diagnosis is limited. On average, malignant NMEs exhibit higher ADC values than malignant masses, overlapping with those of benign lesions and impacting diagnostic accuracy^6–9. These findings were confirmed in the ACRIN 6702 multicenter study, where DWI diagnostic performance was higher in masses than NMEs when using an ADC cutoff to avoid unnecessary biopsies while maintaining sensitivity (46% reduction in benign biopsies vs. 21%)¹⁰.

In this retrospective single-center study, Marino et al.¹¹ investigate the diagnostic performance of breast multiparametric MRI (including DCE-MRI and DWI sequences) for NME lesions. In 95 patients who underwent biopsy for a suspicious NME detected on DCE-MRI, two readers retrospectively assessed BI-RADS lesion characteristics on DCE-MRI and measured lesion ADC on DWI acquired in the same MRI exam. Multiple approaches were tested to evaluate the diagnostic performance of multiparametric MRI for NMEs, with differing ADC thresholds to distinguish benign versus malignant lesions, including those recommended by the European Society of Breast Imaging (EUSOBI) consensus¹² and ACRIN 6702 multicenter trial¹⁰ (1.3×10⁻³ mm²/sec and 1.5×10⁻³ mm²/sec, respectively). The authors compared the sensitivity and specificity of these different approaches, as well as the individual performances of DCE-MRI and DWI. In line with prior studies^6,7,9, the authors observed lower ADC values in malignant vs. benign NMEs. They reported that depending on the approach employed, multiparametric MRI could improve specificity for NME diagnosis compared to DCE-MRI alone, a finding supported by other studies^5,9,10,13, but always came at the cost of reduced sensitivity.

For their DWI acquisition, the authors used single-shot echo planar imaging (EPI), one of the most common DWI approaches that provides fast acquisition times but with substantially lower spatial resolution than DCE-MRI. The limited spatial resolution of DWI, along with lack of distinct boundaries between NME lesions and normal tissue, can cause partial volume effects that result in incorrect lesion characterization on DWI or poor to no lesion visibility. For example, in the ACRIN 6702 trial, 26% of NME lesions were found to be non-evaluable for ADC measurement¹⁴. Similarly, in the present study Marino et al. found 25% of NME lesions (N=24 total; 12 benign, 12 malignant) were not visible on DWI and therefore also excluded from their analyses¹¹. While in both studies breast DWI was performed in line with consensus guidelines¹², it is important to note that more advanced DWI techniques may improve lesion visibility.

Emerging technical developments in DWI sequences show promise in improving diagnostic performance for NME lesions³. High-resolution acquisition strategies including multi-shot and reduced field-of-view EPI could improve lesion conspicuity and produce sharper images, allowing for better assessment of tumor shape and margin. Moreover, simultaneous multislice imaging (SMS), which accelerates acquisition in the slice dimension, enables collection of a greater number of thinner slices and has been used in conjunction with multi-shot EPI in the breast to leverage benefits of both methods¹⁵. More recently, emerging reconstruction techniques leveraging artificial intelligence (AI) to assist in image denoising are being implemented by multiple scanner manufacturers and show promise to support even higher resolution DWI acquisitions¹⁶. A variety of DWI post-processing techniques that are not yet routinely implemented can also improve image quality by reducing magnetic field inhomogeneity-related EPI distortions and artifacts due to eddy-currents and motion, as well as correcting b-value inaccuracies due to gradient nonlinearities¹⁷. Overall, critical image quality improvements to increase DWI spatial resolution and alignment with DCE images would likely increase the utility of multiparametric MRI for evaluating NME by enabling direct voxel-wise comparison for all lesions, regardless of conspicuity on DWI. Additionally, as the authors acknowledge, ADC measurement is highly sensitive to region-of-interest (ROI) placement, impacting both accuracy and reproducibility. While the authors observed high inter-reader agreement across DCE-MRI measures, agreement across multiparametric MRI reads was poor, potentially due to variable ROI selection on DWI between readers. This limitation of DWI has also been noted in other studies^18,19, motivating the development of semiautomated segmentation tools to generate more consistent ADC measurements¹⁹.

In summary, caution should be exercised in the interpretation of DWI to evaluate NME lesions in a multiparametric MRI setting. As Marino et al. confirm, there remain challenges to overcome for the application of DWI in NME diagnosis to improve specificity without reducing sensitivity. However, emerging DWI technological developments hold potential to improve the accuracy and reproducibility of NME discrimination and warrant re-evaluation of NME diagnosis with multiparametric MRI in the near future.