Dual-energy computed tomography for improved delineation in postoperative brain-tumor patients

Purpose/Objective:

The clinical use of dual-energy CT (DECT) contributes to an improved accuracy in proton treatment planning compared to single-energy CT (SECT) as demonstrated in recent studies. A precise delineation of tumor volumes and organs at risk (OARs) is essential in particular for emerging high-conformal treatment techniques. Since DECT provides additional tissue information and allows for the generation of various tissue contrasts, we assessed its influence on the intra- and inter-observer delineation variability.
Material/Methods:
Two cohorts of 10 postoperative brain-tumor patients each, receiving either a 120kVp SECT or 80/140kVp DECT scan with identical total dose, were evaluated. Pseudo-monoenergetic CT (MonoCT) datasets of 50, 60, 70 and 79keV, representing several tissue contrasts, were derived from DECT scans processed in syngo.via (Siemens Healthineers). Three radiation oncologists with different levels of experience in neuro-oncology delineated the postoperative tumor bed volume (TBV) and OARs (brainstem, parotid and lacrimal glands, eyes, lenses, optic nerves, and chiasm) on each dataset, at least two-weeks apart per patient. Relevant image information was blinded. The delineations on SECT datasets were repeated once to assess the intra-observer variability. Finally, the delineation was also performed on T1/T2 MR scans as clinical reference.
The contour conformity was quantified by the Jaccard index (JI) and Hausdorff distance (HD) between the intersection and union of the respective contours (Fig. 1).
Results:
The median inter-observer TBV conformity (Fig. 2A) was almost independent from the CT dataset (HD=6-9mm, JI=61-66%) and comparable to MR scans (HD=6-7mm, JI = 66-67%). The consistency of brainstem contours (Fig. 2B) was best at the lowest energy of MonoCT datasets (median HD=2.8mm, JI=81%). In contrast, the contour conformity of the parotid glands (Fig. 2C) gained slightly from an increased energy (median HD reduction of 0.6mm, JI increase of 1%) and also led to better results as MR scans. For these OARs, using the most suitable MonoCT instead of SECT resulted in smaller interobserver variations. No relevant differences between SECT and MonoCT were determined for the other OARs, potentially due to their small volume.
The intra-observer TBV variability obtained on SECT did not depend on clinical experience. However, the contouring of less experienced clinicians is more affected by different image contrasts introduced by MonoCT of different energies (Fig. 2D).
Conclusion:
For postoperative brain-tumor patients, DECT-derived MonoCT datasets can improve the intra- and inter-observer delineation conformity compared to the currently used SECT. Moreover, they in part led to similar or better results as the gold standard MR. The most suitable image contrast to meet individual delineation requirements of anatomical structures can be chosen after CT acquisition. Future studies need to show whether the advantages can also be translated to other tumor entities and body regions.