The Impact of Interactive MRI-Based Radiologist Review on Radiotherapy Target Volume Delineation in Head and Neck Cancer

DISCUSSION

MR imaging–based radiologist review resulted in amendments to 76% of tumor and 41.4% of nodal volumes, with 54.8% of GTVT and 66.6% of GTVN changes being classified as major. Most GTVT changes were related to the addition of disease in the deep mucosal extent. Some decisions made following collaborative discussion, such as inclusion of borderline nodes, may not have occurred during oncologist-only contouring, and their clinical relevance is uncertain, with a potential for overcontouring resulting in an increased dose to normal tissues. Conformity indices suggested good similarity between the pre- and postradiologist GTVT and GTVN. Because expansion from GTVT was often constrained by adjacent bone and air, there was less impact of the radiologist’s review on representative CTVT volumes, though the small differences in volume size were statistically significant. Conversely, differences among nodal volumes were amplified by expansion to the CTVN. This did not reach statistical significance, possibly due to the small number of nodal volumes available for analysis.

Among radiologists, HN imaging is recognized as a challenging subject requiring specialized training.¹⁸ Oncology training includes no formal diagnostic radiology education; therefore, radiologist-delineated volumes were presumed to be the criterion standard. In our study, most GTV changes were due to undercontouring. Radiologists routinely analyzed information from multiple sequences and had more familiarity with additional sequences such as DWI, which allows greater confidence in distinguishing pathology from normal tissues and in detecting lymph nodes. Review of the DWI, in particular, was implicated in 7 GTV adjustments on peer review. Oncologists agreed with radiologists’ recommendations to increase volumes to reduce the risk of a geographic miss, as primary tumor recurrence within the irradiated field is rarely salvageable. However, if the radiologic evaluation of superficial tumor extension was thought to be erroneous on the basis of mucosal clinical findings, then this issue was discussed and discounted.

Diagnostic reports were available at contouring; however, most lymph node additions were either misidentified pathologic nodes within the same cervical level as correctly delineated nodes or suspicious nodes that were not formally reported. In our institution, diagnostic reports state the existence of nodal disease per cervical nodal level without listing each individual node, thus increasing the likelihood of imaging misinterpretation by oncologists. In addition to detecting overlooked pathologically appearing (by size and morphological criteria) lymph nodes, there was a tendency for radiologists to suggest the inclusion of borderline enlarged nodes in the lymphatic drainage pathways of the primary tumor that were deemed suspicious, with which the oncologists agreed.

Compared with similar studies, our rate of change to gross volumes was high. However, the proportion of major changes was comparable. Braunstein et al⁷ studied the impact of neuroradiologist peer review on, primarily, CT-based RT planning volumes. They noted a 55% alteration rate, 61% of which was clinically significant. More recently, Chiu et al⁸ evaluated the effects of radiologist input using coregistered MR imaging scans and reported a 52% alteration rate, with major changes seen in 79%. Studies of oncologist-only peer review reported lower rates of any change (14%–39%) and major/clinically significant change (8.8%–13%).^9,10 The discrepancy between outcomes of radiologist and oncologist peer review may be reflective of the level of expertise in imaging interpretation, possibly exaggerated further when using MR imaging. Moreover, in our institution, radiologists attended interactive review sessions rather than postdelineation peer review meetings. This practice allowed detailed exploration for each case on a one-to-one basis.

Comparison of clinical significance across studies is difficult because definitions for major/clinically significant changes vary. Some articles, such as this one, classified major changes on the basis of volumetric results; therefore, rates were reported only among amended volumes (with prereview volumes saved).⁷ Others reported clinical significance among all reviewed volumes (including unaltered and unsaved volumes).^8⇓-10 Among all reviewed volumes in this study, 34% of GTVT and 14.6% of GTVN would have undergone major changes. However, the lack of volumetric analysis of 15 unsaved cases of GTVonc likely distorts those results.

Our conformity indices were very good and consistent with quality-assurance studies in clinical trials reporting a median JI of ≥0.7 and a DSC of ≥0.8,^19,20 and the recent peer-review study of Chiu et al.⁸ Chiu et al reported median GTVT values of DSC = 0.97, JI = 0.94, and HD = 3.6 mm. They also observed a larger HD for GTVN at 37 mm and a reduction in conformity on GTVN-to-CTVN expansion. While a GTVT is usually 1 continuous volume, GTVN often includes multiple separate nodes; therefore, the inclusion of additional nodes will have a greater impact on conformity and distance. In our study, nodes were often added superior and/or inferior to the GTVNonc, expanding the volume within the soft tissues before anatomic barriers were encountered.

Following publication of the study of Braunstein et al,⁷ the use of MR imaging–guided RT volume delineation in HNC has increased, with a projected future move toward delineation based on functional imaging and MR imaging–based synthetic CTs,²¹ hence the focus of our study on MR imaging–based volume delineation. In contrast to earlier, predominantly CT-based studies, we detected a predominance of undercontoured volumes by oncologists when using MR imaging.^7,22 This may reflect the lack of formal training in MR imaging interpretation for oncologists as well as a conservative approach to include areas/nodes considered suspicious to minimize the risk of a geographic miss. Therefore, involvement of input of specialist radiologists in GTV delineation may be more advantageous, given the increasing complexity of HN MR imaging interpretation, particularly when using functional sequences.

Few MRIs in our study were acquired in the RT treatment position; therefore, registration error exceeded both optimum TRE values in half of the scans. The effect of suboptimal registration on the degree of volume change is uncertain. Among amended volumes with TREmax > 5mm (n = 24), only 1 case corrected the GTVrad, but not GTVonc, on CT. In 3 cases, both GTVonc/rad were contoured on both CT/MR imaging and compared accordingly. In 5 cases, both GTVonc and GTVrad were produced on MR imaging only. In 16 cases, GTVonc and GTVrad were evidently corrected for anatomic mismatch to some degree (ie, both contours edited away from bone/air). In these cases, some soft-tissue discrepancies remained, which may have been due to registration error, but there was no clarifying documentation regarding this aspect. Nevertheless, the accompanying descriptive reasons for volume change for each case suggest that many soft-tissue volume discrepancies would have persisted irrespective of registration error.

There are limitations to this study. It describes a one-to-one comparison for a single oncologist and single radiologist at a given time, without analysis of interobserver agreement within the oncologist and radiologist groups. It is recognized that delineation of GTVs by each of the oncologists and radiologists would have allowed the calculation of agreement parameters for tumor volumes and would have given a better understanding of performance between and across the 2 disciplines. However, the inclusion of multiple different radiologists and oncologists in the review process reflects the range of real-world practice in our institution.

Another limitation includes a lack of longitudinal data to assess the presence of a learning curve, especially important as the more junior oncologists gain experience. The possibility that oncologists may have delayed complex contouring decisions in anticipation of the pending radiologist review, resulting in more volume amendments, cannot be excluded. However, the prospective documentation suggests that when oncologists viewed a case as complex, the radiologist tended to agree. Therefore, the peer review process facilitated decision-making for these difficult cases.

The use of a single observer to create study CTVs may have increased the variation between the GTVs and CTVs. Ideally the oncologist for each case would have produced a CTVonc before peer review; however, this step was impractical because radiologist review and oncology target delineation sessions occurred on the same day. Preoperative GTVTs were included in volumetric analysis because they still provided information on delineation accuracy of the oncologist. However, removing the consequent postoperative CTVTs may have introduced selection bias toward poorer CTVT conformity (no postoperative nodal volumes were amended). Nevertheless, CTVT conformity was still superior to GTVT conformity on both measures.

Additional comparison metrics assessing the extent of undercontouring (geographic miss index) and overcontouring (discordance index) may have provided a more specific assessment of contour similarity, uninfluenced by volume size. However, further metrics would not likely reduce the qualitative benefits of peer review highlighted in this study. Furthermore, as more centers adopt the intensity-modulated RT 5 + 5 consensus,²³ which reduces the margin for the high-risk CTV to 5mm, the need for optimized GTV delineation is increasing.

At our institution, HN oncology and radiology teams agreed on the need for expert radiology review of gross volumes when MR imaging delineation was introduced in the HNC RT workflow, given the complexity of the tumor site and the lack of specific MR imaging training of oncologists. This study was planned to assess its impact. Since its implementation, we have expanded the use of MRIs in the immobilization mask for patients having definitive intensity-modulated RT, to reduce registration errors. The efficiency and responsiveness of the radiology review process have been enhanced since 2020 by performing remote radiology review using an application equipped with shared screen functions. CTV and PTV volumes are subsequently peer reviewed within the oncology group. In this study, we have shown that the impact of radiology review of tumor and nodal volumes is sufficiently significant to warrant expert review of radiologists in MR imaging–based volume delineation in patients with HNC, something to be considered by institutions and funding bodies.