Clinical Validation of Ivim Parameter Estimation for Treatment Response Assessment In Head-and-Neck Cancer Radiotherapy
Abstract
Purpose
To clinically validate an implicit neural representation (INR)-based fitting approach for intravoxel incoherent motion (IVIM) diffusion MRI parameter estimation against conventional non-linear least squares (NLLS), and to evaluate a robustness metric (R-index)—a linear combination of IVIM parameters designed to mitigate collinearity and reduce estimation uncertainty—in head-and-neck cancer patients imaged before and during radiotherapy. This study aims to leverage IVIM-derived biomarkers to characterize individual treatment response and support biology-driven adaptive radiotherapy decision-making.
Methods
Six patients underwent IVIM-MRI prior to and during treatment (~1-month interval). Apparent diffusion coefficient (ADC) and IVIM parameters, including perfusion fraction (fp), tissue diffusion coefficient (Dt), and R-index (~2fp + Dt), were estimated using both INR and NLLS methods. Parameter values were compared between time points, relative changes (%) were calculated, and individual treatment responses were analyzed and interpreted based on synergistic patterns of parameter changes.
Results
Compared with NLLS, INR consistently yielded lower fp (-13% to -65%) and higher Dt (+14% to +49%) across patients, consistent with prior digital phantom results. R-index demonstrated mean inter-method differences of 1.3% and 2.7% at 1st and 2nd scans respectively, well below 5%. ADC also showed minimal inter-method difference (<2%). Both fitting approaches showed concordant temporal trends in 5 of 6 patients. Importantly, ADC and R-index reflected combined perfusion and diffusion effects; their interpretation as biomarkers requires caution. For example, an ADC increase (+17%) was associated with concurrent increases in both Dt and fp in one patient, whereas an ADC decrease (-11%) occurred despite increased Dt due to marked reduction in fp in another.
Conclusion
This first clinical validation confirms superior accuracy of INR-based IVIM parameter estimation. The findings highlight the importance of disentangling diffusion and perfusion contributions when interpreting composite biomarkers for treatment response. Future studies correlating IVIM parameter dynamics with clinical outcomes are warranted to establish prognostic value.