Cellular Characterization of Breast Cancer Using Microstructural Diffusion MRI

See also the article by Wang and Ba et al in this issue.

Savannah Partridge, PhD, is a professor in the breast imaging section of the Department of Radiology at the University of Washington. Her research focuses on optimization and clinical validation of advanced breast MRI and diffusion-weighted imaging techniques. She is an NIH-funded principal investigator and a distinguished investigator of the Academy for Radiology and Biomedical Imaging Research, and has led breast MRI–focused multicenter trials through the ECOG-ACRIN Cancer Research Group.

Junzhong Xu, PhD, is an associate professor and director of cancer imaging research in the Department of Radiology and Radiological Sciences at Vanderbilt University Medical Center. His research interests focus on the development, validation, and implementation of advanced MRI methods in cancer. Dr Xu is a distinguished investigator of the Academy for Radiology and Biomedical Imaging Research, serves on NIH study sections, and is principal investigator for several NIH grants.

As oncology moves toward personalization of therapies, improved methods for disease stratification and prediction of response to neoadjuvant chemotherapy are critically needed to optimize treatment regimens. For breast cancer, imaging can have an important role in addressing this clinical need by noninvasively providing information about tumor biologic characteristics that can be used to tailor therapy. Among breast imaging modalities, multiparametric breast MRI is particularly effective for visualizing treatment-induced pathologic changes in breast tumors. Whereas dynamic contrast-enhanced MRI is the standard method for evaluating breast cancer based on alterations in tissue vascularity, diffusion-weighted imaging (DWI) has been shown in numerous studies to provide additional and complementary information about tissue cellularity and microstructure that can be used to characterize breast tumors and monitor their response to treatment (1).

For example, the American College of Radiology Imaging Network 6698 multicenter trial, known as ACRIN 6698, confirmed the value of the apparent diffusion coefficient (ADC) obtained from DWI as an early marker of treatment response. In 242 patients undergoing neoadjuvant chemotherapy, increase in tumor ADC reflecting cytotoxic effects midway through treatment predicted pathologic complete response when combined with tumor subtype (area under the receiver operating characteristic curve [AUC], 0.72; 95% CI: 0.61, 0.83) (2). Findings from the I-SPY 2 trial suggested DWI may be especially helpful in evaluating the effectiveness of immunotherapy-containing regimens. In this trial, early change in tumor ADC after 3 weeks outperformed dynamic contrast–enhanced MRI measures in predicting pathologic complete response in patients who were administered neoadjuvant chemotherapy with pembrolizumab (AUCs, 0.73 and 0.61, respectively) (3). Although these results are encouraging, efforts are underway to further increase the predictive performance and usefulness of DWI for tailoring therapies.

Although ADC, derived from the simplest monoexponential signal decay diffusion model, is a frequently used DWI metric for cancer applications, it is nonspecific and may have reduced sensitivity due to competing influences of concurrent pathologic changes. More advanced DWI models offer compelling advantages to distinguish the underlying microenvironmental factors of tissue cellularity, vascularity, heterogeneity, and microstructure contributing to the diffusion signal. So-called time-dependent diffusion MRI techniques, which incorporate time dependence of the diffusion signal, probe microstructural complexity at the cellular level. Approaches range from simple time-dependent ADCs (ie, ADCs calculated at different diffusion times with fixed b values) that reflect change in overall diffusion properties with time to multicompartmental methods that incorporate varying diffusion times and b values to estimate cytomorphologic tissue properties. This includes cell size and density, which cannot be measured at conventional DWI (4), opening exciting avenues for biologic characterization of cancer and other diseases.

In this issue of Radiology, Wang and Ba et al (5) investigate the clinical utility of a time-dependent diffusion MRI method for predicting molecular subtype and response to neoadjuvant chemotherapy in breast cancer. They implemented a recently developed time-resolved diffusion technique (imaging microstructural parameters using limited spectrally edited diffusion, known as the IMPULSED method) to estimate microstructural properties (6), along with calculating time-dependent ADCs and relative ADC change (with effective diffusion times of 5–30 msec) (7). The time-dependent ADC approach has long been used to characterize biologic tissues in neuroimaging and cancer imaging, but its reliance on a single-compartment model limits its specificity for pathologic features. However, the IMPULSED method combines (a) an advanced diffusion acquisition schema with a broader diffusion time range and (b) a multicompartmental tissue-modeling framework to enable mapping cell size (effective diameter), intracellular fraction, extracellular diffusivity, and cellularity in vivo (6).

In the prospective study by Wang and Ba et al (5), diffusion MRI was performed at 3-T MRI using a combination of oscillating and pulsed gradient spin-echo sequences to capture the pretreatment time-dependent diffusion characteristics in 408 participants with breast cancer (mean age, 51.9 years ± 9.1 [SD]). Of these, 221 participants underwent neoadjuvant chemotherapy. The time-dependent diffusion MRI–derived parameters and the corresponding combination models achieved predictive performance for identifying different breast tumor subtypes, such as luminal A (AUC, 0.70), luminal B (AUC, 0.78), triple negative (AUC, 0.72), and human epidermal growth factor receptor 2 enriched (AUC, 0.85), outperforming conventional ADC measurements (all P < .05). Cellularity was a crucial marker among time-dependent diffusion MRI microstructural parameters and showed the highest performance for predicting pathologic complete response in response to neoadjuvant chemotherapy (AUC, 0.84). A model combining progesterone receptor, human epidermal growth factor receptor 2, and cellularity achieved an even higher performance (AUC, 0.88), better than conventional DWI and clinical-pathologic models (P < .001). Moreover, the time-dependent diffusion MRI–derived parameters showed good correlations with pathologic measurements of cell size, cellularity, and intracellular fraction (n = 100; r = 0.67−0.81; P < .001). These results suggest that time-dependent diffusion MRI, including IMPULSED and time-dependent ADCs, has the potential to aid in prognostic profiling of breast tumors and in personalizing treatment strategies.

This study is an important step in translating time-dependent diffusion MRI, particularly IMPULSED, into clinical breast cancer imaging. Although time-dependent ADC has long been explored in patient studies, there is increasing interest in using more advanced microstructural diffusion MRI techniques, including the IMPULSED method and vascular extracellular and restricted diffusion for cytometry in tumors (known as VERDICT) method (4,8), to characterize tumors at the cellular level. However, there is a lack of large-scale patient studies to evaluate the clinical potential of these methods in breast cancer. Although IMPULSED previously demonstrated clinical potential for the assessment of tumor response to chemotherapy, radiation therapy, and immunotherapies (9), most of these studies were performed using preclinical animal models. The findings are yet to be confirmed in clinical populations. This study is, to our knowledge, the first implementation of the IMPULSED method in a large-scale cohort of participants with breast cancer to evaluate its ability to predict breast tumor subtypes and pathologic complete response. The promising results suggest this technique provides superior predictive performance compared with conventional DWI, consistent with preclinical findings, warranting further efforts toward clinical translation.

The main limitation of the study by Wang and Ba et al (5) originates from the typical multicompartmental diffusion MRI modeling (including IMPULSED), which ignores the influences of transcytolemmal (crossing cell membrane) water exchange. Cell membrane permeability changes during treatment (eg, apoptosis), resulting in nonnegligible influences on the estimation of microstructural parameters. More comprehensive biophysical models and/or more advanced sequences are needed to accurately estimate changes in cellularity as a marker of treatment response (9). Another limitation is that the pathologic analysis for validation used hematoxylin-eosin staining, which provides quantification on nuclear size, not cell size. A scaling factor of 1.8 was used to convert nuclear size to cell size, which is problematic because nuclear-to-cytoplasmic ratios vary during treatment. A promising solution has been demonstrated in preclinical studies (9) using sodium–potassium adenosine triphosphatase (sodium–potassium pump) staining to view cell boundaries and quantify pathologic cell sizes.

The ability to probe tissue microstructure depends on the diffusion time, which controls the spatial scale of the relevant tissue structures. At conventional DWI with pulsed gradient spin-echo MRI sequences on commercial MRI scanners, the range of diffusion time is limited (30–50 msec is typical), hampering translatability of time-dependent diffusion MRI techniques. However, the rise of clinical MRI scanners equipped with powerful gradient systems makes it feasible to achieve shorter times with oscillating gradient spin-echo sequences in a clinical setting, facilitating the application of microstructural MRI in imaging a wide range of clinical challenges.

There are several opportunities to expand this work and support further study of the clinical value of noninvasive microstructural breast imaging. With progress in the development of high-performance gradient systems used with MRI scanners, the approach by Wang and Ba et al (5) could be extended with a shorter echo time and a broader diffusion time range. This would improve image quality and fitting precision, and enable possibilities to probe more detailed microstructural information, such as cell size distribution and membrane permeability (9). Rigorous pathologic validation incorporating whole-lesion, spatially resolved analysis is needed to confirm the accuracy of the imaging-derived metrics in heterogeneous breast tumors. More studies are needed, designed in collaboration with pathologists and oncologists, to critically assess the added benefit of microstructural imaging to improve breast cancer diagnosis and treatment and to motivate its translation as a clinical tool.

Beyond technical improvements and further clinical validation, practical challenges must also be overcome to transition microstructural diffusion MRI from a research tool to routine practice in breast imaging. A recent survey conducted by the European Society of Breast Imaging International Breast Diffusion-weighted Imaging working group revealed that despite growing academic interest in advanced DWI techniques and evidence of potential clinical benefit, there has been limited adoption of advanced DWI techniques among its members (10). Their findings highlight the need for standardized protocols and parameter guidelines specific to advanced breast DWI and the importance of user-friendly software for clinical implementation. There is a crucial need for collaborative efforts between researchers, clinicians, and industry experts to achieve these goals and bridge the gap between research and clinical applications.