Defining Meaningful Apparent Diffusion Coefficient Change Using MR-Linac in Head and Neck Cancer

See also article by McDonald et al.

Ramona-Alexandra Todea, MD, is a neuroradiologist specializing in neuro-oncology, focusing on MRI evaluation of brain tumors and advanced imaging in multidisciplinary care. Certified by the Swiss Board in Radiology and Neuroradiology and the European Board in Neuroradiology and Pediatric Neuroradiology, she is also an educator at the Universities of Zürich and Basel. A dedicated researcher in neuro-oncologic imaging, she is passionate about advancing the field and actively contributes as a reviewer.

Zsolt Kulcsar, MD, PhD, specializes in diagnostic and interventional neuroradiology, focusing on complex neurovascular conditions such as cerebral arteriovenous malformations, DAVFs, aneurysms, acute stroke, and pediatric vascular malformations like vein of Galen AVMs. He is recognized for expertise in endovascular treatments and pulse-synchronous tinnitus. Additionally, he is skilled in diagnostic neuroradiology and imaging-guided spine pain therapy, with a strong commitment to advancing clinical care and research in these fields.

Head and neck cancer represents a substantial global health burden, ranking as the seventh most common cancer worldwide, with approximately 888 000 new cases and 453 000 deaths annually. Incidence is markedly higher in men, with a male-to-female ratio of approximately 3:1, and epidemiologic trends vary by geographic region and tumor subtype. Although tobacco- and alcohol-related head and neck cancers have declined in some high-income countries, human papillomavirus–associated oropharyngeal cancers continue to rise. Incidence increases are particularly notable in cancers of the tongue, tonsil, and oropharynx—approximately 2.3% per year—reflecting the growing prevalence of human papillomavirus–driven disease, with associated mortality increasing by roughly 2% annually (1,2).

Head and neck squamous cell carcinoma is commonly treated with radiation therapy, often combined with chemotherapy, yet patient outcomes vary widely. Many individuals experience substantial treatment-related toxicities, including xerostomia, dysphagia, and mucositis. This clinical heterogeneity underscores the need for individualized treatment strategies that maximize tumor control while minimizing adverse effects. In this context, reproducible imaging biomarkers have become increasingly important. Diffusion-weighted MRI (DWI), particularly the apparent diffusion coefficient (ADC), has emerged as a promising quantitative biomarker, with early treatment-related changes in ADC showing potential for predicting response in head and neck squamous cell carcinoma.

Hybrid MR-linac systems have further transformed radiation oncology by enabling online adaptive radiation therapy with integrated high-quality MRI. MR-guided radiation therapy allows direct visualization of the tumor and surrounding anatomy before and during treatment, facilitating adaptation to day-to-day anatomic changes. This capability raises the prospect of biologically guided adaptive workflows based on quantitative imaging biomarkers, provided that such measurements are sufficiently stable and reproducible.

The Elekta Unity system (Elekta AB) combines a 1.5-T MRI scanner with a linear accelerator, enabling serial DWI acquisitions in the treatment position throughout the course of radiation therapy (3). This design minimizes registration errors and enhances the feasibility of longitudinal imaging biomarker assessment. However, several technical modifications necessary for radiation delivery—including weaker and slower gradients, a specialized coil design that allows beam passage, and fewer receiver channels than conventional diagnostic MRI systems—introduce challenges for diffusion imaging. These factors can affect image quality, prolong acquisition times, and influence ADC quantification. Although optimized DWI protocols may yield consistent results, absolute ADC values may differ from those obtained on diagnostic scanners, highlighting the importance of system-specific reproducibility studies.

To ensure that DWI functions reliably as a quantitative imaging biomarker, the Quantitative Imaging Biomarker Alliance recommends test-retest studies to characterize in vivo repeatability and reproducibility (4). Against this backdrop, the study by McDonald et al (5) in Radiology: Imaging Cancer provides an important technical assessment of ADC reproducibility in head and neck squamous cell carcinoma using a 1.5-T MR-linac system equipped with an echo-planar imaging (EPI) DWI sequence.

In this study, patients with histologically confirmed head and neck squamous cell carcinoma underwent EPI DWI at two or more time points between pretreatment MR simulation and the first day of radiation therapy, with a median interscan interval of 10 days. Imaging was performed in radiation therapy immobilization masks using T2-weighted sequences and single-shot EPI DWI, with ADC maps reconstructed from b values of 150 and 500 sec/mm² in accordance with MR-Linac Consortium recommendations (6). Primary tumors and pathologic lymph nodes were contoured on b0 images registered to T2-weighted images to mitigate geometric distortion, and these volumes were propagated to ADC maps. Mean and median ADC values, along with lesion volumes, were calculated voxelwise from three-dimensional contours.

Reproducibility was quantified using Quantitative Imaging Biomarker Alliance–recommended metrics, including within-subject deviation, reproducibility coefficient, within-subject coefficient of variation, and percent reproducibility coefficient. The final cohort included 37 patients, comprising 34 primary tumors and 55 metastatic lymph nodes. Median lesion volumes were 2.9 cm³ for primary tumors and 3.6 cm³ for lymph nodes, with corresponding median mean ADC values of 1.19 × 10⁻³ mm²/sec and 1.02 × 10⁻³ mm²/sec, respectively. No systematic differences in ADC were observed between time points for primary tumors, whereas lymph nodes demonstrated greater variability. The reproducibility coefficient for mean ADC was 0.365 × 10⁻³ mm²/sec (29.9%) for primary tumors and 0.355 × 10⁻³ mm²/sec (31.1%) for lymph nodes (5).

These findings are particularly important because they define the minimum ADC change that can be confidently interpreted as biologic rather than technical. Given that early ADC changes of 10%–20% have been associated with treatment response in prior studies, a reproducibility threshold of approximately 30% on the MR-linac may limit sensitivity for detecting early responders (7,8). This has direct implications for MR-guided adaptive workflows: only relatively large diffusion changes can be interpreted with confidence, whereas smaller but potentially meaningful early changes may fall within measurement noise.

Interestingly, reproducibility metrics were similar for primary tumors and lymph nodes, contrary to expectations that smaller nodal volumes would exhibit greater variability. This may reflect the fact that most primary tumors were small (T1–T2) and located near air–tissue interfaces prone to susceptibility artifacts, whereas lymph nodes were often situated in anatomically more stable regions. The study also demonstrated minimal differences between mean and median ADC reproducibility, suggesting that either metric may be suitable for longitudinal assessment despite the non-Gaussian distribution of ADC values in head and neck lesions. Variability was only modestly influenced by scan interval or lesion volume, although very small lesions (<10 cm³) tended to show higher within-subject coefficient of variation values.

As with all test-retest imaging studies, limitations warrant consideration. These include single-observer segmentation, heterogeneity in lesion subsites, and the inherent distortion susceptibility of EPI-based DWI on MR-linac platforms. The use of vendor-standard reconstructions enhances clinical relevance but may limit performance relative to advanced research techniques. These findings underscore the need for continued technical optimization, including alternative DWI readouts such as split acquisition of fast spin-echo signals for diffusion imaging, turbo spin-echo DWI, or multishot EPI, as well as artificial intelligence–based motion and distortion correction strategies.

In summary, the study by McDonald et al provides a clinically relevant evaluation of ADC reproducibility on a 1.5-T MR-linac system, establishing essential performance benchmarks for interpreting diffusion changes during radiation therapy in head and neck squamous cell carcinoma. By defining the magnitude of ADC change required to exceed measurement noise, the authors lay critical groundwork for biomarker-driven adaptive treatment strategies. Future efforts should focus on improving DWI sensitivity through advanced acquisition and reconstruction techniques, harmonizing protocols across MR-linac centers, and refining quantitative workflows to fully realize the potential of diffusion MRI for early response assessment and biologically guided radiation therapy.

Disclosures of conflicts of interest

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