ESTRO 2024 Congress Report

Advanced radiotherapy (RT) technologies can deliver increasingly complex treatments that precisely shape the high dose volume of RT dose distributions. Such advancements open the possibility for RT to tightly conform the high dose volume to the target and better spare healthy tissues, but it is impossible to eliminate all healthy tissue exposures. Therefore, effectively leveraging advanced technologies for gentler, yet effective treatments will require a deep understanding of which tissues are the most radiosensitive, and hence the most important to spare.

Analyses based on dose-volume histograms (DVHs) have traditionally been the workhorse of tissue radiosensitivity studies. The results of DVH analyses represent the degree to which the doses within an organ or tissue of interest are associated with a particular treatment effect. Although DVH analyses have contributed to many important breakthroughs in our understanding of radiation dose response, they have several critical limitations. First, DVH analyses require a priori selection of which organs and tissues to consider. Pre-selection of structures can bias analyses to pre-existing anatomic assumptions regarding which organs and tissues drive a particular toxicity. Second, the process of creating DVHs discards the detailed spatial information regarding how the dose was distributed within the pre-selected structure. Finally, DVH analyses require individual delineations of each pre-selected organ or tissue of interest for each patient. The delineation requirement hinders DVH analyses of organs and tissues that lack clear anatomic limits, including sub-regions of organs or tissues.

Voxel-based analysis (VBA) methods were developed for radiation oncology applications specifically to overcome the limitations of traditional DVH techniques. VBA methods agnostically analyse entire dose distributions, without the need for a priori structure selection and have shown promise for investigating spatial associations between radiation dose and treatment outcomes across several body sites, including lung, head-and-neck, and pelvis. The results of VBA represent the volume of tissue in which dose is associated with a particular treatment effect. It remained unclear, however, how to interpret new VBA findings in the context of prior DVH-based results. Therefore, we set out to investigate the concordance between the results of novel VBA and conventional DVH analyses.

We used data from a published study of cognitive outcomes following craniospinal photon therapy for medulloblastoma. The prior study used DVH techniques to identify brain substructures in which radiation dose was associated with declines in working memory or processing speed. We used the same treatment planning and follow-up data as the prior study, but instead applied VBA to identify spatial associations with radiation dose. We evaluated the concordance of results from the two methods by quantifying the volume overlap between the VBA-identified significant volume and the anatomic structures considered in the prior study.

Overall, our analyses revealed that VBA results agreed with over 75% of the prior DVH findings (45% significant association in both DVH and VBA, 33% no significant association in either DVH or VBA). Disagreements were uniformly because DVH analysis found significance in organs or tissues that were excluded from VBA significant volumes, suggesting a false-positive DVH-based result (22%). The presence of DVH false positives is plausible because repeated DVH analyses on organ substructures are vulnerable to statistical artefacts while the VBA method is specifically designed to conservatively protect against spurious results due to multiple comparisons.

Our results are encouraging because they suggest that the modern VBA technique produces results that are largely consistent with those of well-accepted DVH methods. Concordance between the two methods will facilitate the integration of novel VBA tissue radiosensitivity findings into the existing, DVH-based body of knowledge. It is important to remember, however, that both DVH and VBA methods evaluate correlations only, and so further work is needed to establish causal links (e.g., causal inference, pre-clinical studies, etc.). Furthermore, clinical applications of VBA should always include robustness evaluations using several reference patients, and results should be confirmed in independent validation cohorts. Moving forward, VBA results can help researchers refine our understanding of tissue radiosensitivities, which can then be fed into modern RT workflows to provide patients with treatments that achieve a cure without radiation-related toxicities.

Lydia J Wilson, PhD, DABR
Thomas Jefferson University, Philadelphia, PA, USA
Lydia.Wilson@Jefferson.edu
@DrLydiaJWilson