Diffusion-weighted Imaging and ADC Mapping for Early Prediction of Pathologic Complete Response in Breast Cancer

See also the article by Le et al in the July 2026 issue.

Pratiksha Yadav, MD, PhD, is a professor of radiology at Symbiosis Medical College for Women and Symbiosis University Hospital and a fellow of the Indian College of Radiology. Her research work focuses on advanced MRI techniques and their clinical applications and artificial intelligence, particularly in breast imaging. She is involved in teaching and clinical work, along with peer review and academic work. She is a dedicated researcher with a strong commitment to advancing clinical care and research.

Breast cancer represents a major global health challenge, with increasing incidence and mortality. It is a heterogeneous disease that requires tailored treatment strategies and targeted therapies according to molecular subtype. Among current imaging modalities, dynamic contrast-enhanced (DCE) MRI is the most sensitive noninvasive technique for breast cancer detection. It provides functional information about lesions that helps differentiate benign from malignant disease. Although DCE MRI offers excellent sensitivity for breast cancer screening, its clinical utility is limited by variable specificity. In addition, the use of contrast media increases cost, prolongs imaging time, and carries a risk of adverse reactions. Multiparametric breast MRI with diffusion-weighted imaging (DWI) has improved the sensitivity and specificity for detecting breast malignancies (1,2). DWI demonstrates differences in the diffusion of water molecules between normal and pathologic tissues, providing additional information about tissue microstructure. Malignant lesions typically have high cellularity and demonstrate diffusion restriction on DW images, which can be assessed both qualitatively and quantitatively. The apparent diffusion coefficient (ADC) provides a quantitative measure of water diffusion that aids in differentiating benign from malignant lesions and in predicting tumor grade, proliferation rate, and molecular subtype. Previous studies have demonstrated an inverse relationship between tumor grade and ADC values (1). Other studies have explored the role of DWI and the predictive value of tumor ADC for assessing pathologic complete response (pCR) to neoadjuvant chemotherapy, attributed to treatment-induced changes in tumor cellularity and the tumor microenvironment associated with cell lysis and the cytotoxic effects of chemotherapy (3,4).

Thus, diffusion-weighted MRI has emerged as a valuable noncontrast adjunct for detecting malignant lesions, lesion characterization, and predicting response to neoadjuvant chemotherapy (5). However, the accuracy of ADC analysis depends on placement of the region of interest (ROI), and manually positioning the ROI within the areas of lowest ADC is time-consuming and susceptible to human error. Automated calculation of ADC values represents an important advancement by providing a reproducible, noninvasive quantitative measure that may help predict chemotherapy response. Automated approaches can extract multiple metrics, including minimum ADC, mean ADC, and whole-lesion histogram features. Furthermore, they reduce processing time and minimize interobserver variability.

In the July 2026 issue of Radiology: Imaging Cancer, Le et al (6) investigated an automated method for ADC measurement of breast tumors on DWI MR images for early assessment of response to neoadjuvant chemotherapy. The authors performed a retrospective analysis of data from the American College of Radiology Imaging Network (ACRIN) 6698 multicenter trial conducted at 10 institutions between August 2012 and January 2015 (4). A total of 226 participants with breast cancer who underwent neoadjuvant chemotherapy were included in the analysis. Multiparametric breast MRI, including DCE MRI and DWI, was performed during treatment. Imaging was obtained at baseline (T0), early treatment (T1, 3 weeks after treatment initiation), interregimen (T2, approximately 12 weeks after treatment initiation), and presurgery (T3, approximately 24 weeks after treatment initiation). For the present study, only the T0 and T1 examinations were analyzed.

Functional tumor volume was derived from DCE MRI as the volume of contrast-enhancing tissue, using a predefined threshold previously validated for predicting response to neoadjuvant chemotherapy in multicenter trials (4). The tumor ROI was transferred from DCE MRI to DWI using image registration, allowing automated ROI generation. Image registration quality was assessed qualitatively by visual inspection. The authors also evaluated the reproducibility of automated ADC measurements in a test-retest subcohort. For this analysis, two DWI examinations performed with identical protocols during a single visit, preferably at baseline (T0) or, if unavailable, at the early treatment time point (T1), were analyzed. ADC values for the test-retest cohort were calculated using two b values (b = 0 and 800 sec/mm²).

The study used the same ADC maps as those in the previously published primary ACRIN 6698 analysis; however, automated rather than manually defined ROIs were used. The automated ROI was further divided into overlapping subregions, containing voxels shared by both the automated and manual ROIs, and nonoverlapping subregions, containing voxels present only in the automated ROI. Seventy-one participants (31.1%) achieved pCR. The change in ADC (ΔADC) at 3 weeks predicted pCR with an area under the receiver operating characteristic curve (AUC) of 0.61 (P = .008). Predictive performance differed between the overlapping and nonoverlapping subregions. Receiver operating characteristic analysis demonstrated an AUC of 0.52 (95% CI: 0.44, 0.61) for the overlapping subregion and 0.63 (95% CI: 0.55, 0.70) for the nonoverlapping subregion, although the difference between AUCs was not statistically significant (P = .25).

In the test-retest subcohort, predictive performance was higher in human epidermal growth factor receptor 2 (HER2)-positive than in HER2-negative tumors. ΔADC derived from the nonoverlapping subregion demonstrated the strongest predictive performance. However, the HER2-positive subgroup was relatively small, and image registration errors were present. The authors proposed that automated ΔADC measurement is reproducible and can predict pCR early during treatment, particularly in HER2-positive tumors. The study demonstrated good reproducibility, with an agreement index of 0.82, supporting its potential clinical utility in monitoring response to neoadjuvant chemotherapy. Notably, the stronger predictive performance observed in the nonoverlapping region at the tumor periphery highlights the potential prognostic value of peritumoral tissue.

Despite this potential, widespread adoption of DWI and ADC measurements faces several challenges. ADC measurements often vary across MRI scanners and acquisition protocols, highlighting the need for greater standardization (7,8). Furthermore, automated ADC calculation has limitations for small breast lesions (≤10 mm), increasing the risk of false-negative findings in subcentimeter masses. Additionally, quantitative metrics designed to measure intratumoral heterogeneity often fail to provide added diagnostic value in small lesions. Although minimum ADC is highly effective for larger, heterogeneous lesions containing necrosis, subcentimeter lesions rarely exhibit macroscopic necrosis; consequently, mean ADC often outperforms minimum ADC in these smaller lesions. Deep learning has the potential to optimize clinical workflows by triaging screening examinations and classifying lesions. Standardized imaging protocols are essential to ensure that ADC measurements are repeatable and reproducible, thereby establishing ADC as a reliable noninvasive biomarker for diagnosis, treatment monitoring, and prediction of response to neoadjuvant chemotherapy. However, to fully utilize the potential of ADC as a predictive biomarker, a thorough understanding of the factors influencing these measurements, together with global standardization of imaging techniques, remains essential. Evaluating the reproducibility of ADC measurements is particularly important because these metrics are susceptible to variation related to scanner hardware, acquisition protocols, and patient positioning. Whole-lesion ADC measurements have demonstrated slightly higher accuracy than ROI-based ADC measurements for predicting radiologic complete response. Le et al observed that automated ΔADC measured from the whole-tumor ROI predicted pCR at the early treatment time point, a finding that was not observed in the primary ACRIN 6698 trial (4).

Recent breast imaging research has evolved beyond basic lesion differentiation (benign vs malignant) toward more advanced tumor characterization and prognostication. By capturing tissue microstructure, quantitative ADC metrics derived from DWI can differentiate tumor grades, proliferation rates, and breast cancer subtypes (6). In addition, this approach may help predict the surgical upgrading of pure ductal carcinoma in situ to invasive cancer and may help assess the risk of axillary lymph node involvement.

In summary, the study by Le et al provides a clinically relevant evaluation of automated ADC measurement using image registration. This approach may facilitate interpretation of diffusion changes during neoadjuvant chemotherapy and help predict pCR early during treatment. The authors lay important groundwork for establishing ADC as a reliable noninvasive biomarker of early chemotherapy response that may ultimately help guide treatment decisions in patients with breast cancer. The study also supports the development of predictive models that integrate functional tumor volume and ADC for assessment of treatment response. Future efforts should focus on improving automated ADC calculation and advancing acquisition and reconstruction techniques to optimize the use of DWI for breast cancer diagnosis and treatment monitoring.

Disclosures of conflicts of interest

Please see ICMJE form(s) for author conflicts of interest. These have been provided as supplemental materials.