Comparing methods to improve cone-beam computed tomography for dose calculations in adaptive proton therapy

Abstract

Background and purpose: Proton therapy requires dose monitoring, often performed based on repeated computed tomography (reCT) scans. However, reCT scans may not accurately reflect the internal anatomy and patient positioning during treatment. In-room cone-beam CT (CBCT) offers a potential alternative, but its low image quality limits proton dose calculation accuracy. This study therefore evaluated different methods for quality-improvement of CBCTs (synthetic CTs; sCTs) for use in adaptive proton therapy of head-and-neck cancer patients.

Materials and methods: Thirty-five CBCTs from twenty-four head-and-neck cancer patients were used to assess four sCT generation methods: an intensity-correction method, two deformable image registration methods, and a deep learning-based method. The sCTs were evaluated against same-day reCTs for CT number accuracy, proton range accuracy through single-spot plans, and dose recalculation accuracy of clinical plans via dose-volume-histogram (DVH) parameters.

Results: All four methods generated sCTs with improved image quality while preserving the anatomy relative to the CBCT. The differences in absolute median proton range between sCT methods were small and generally less than the difference between sCT and reCT, which had median differences of 1.0-1.1 mm. Similarly, differences in DVH parameters were generally small between the sCT methods. While outliers were identified for all four methods, these outliers were often consistent for all sCT methods and could be attributed to anatomical and/or positional discrepancies between the CBCT and reCT.

Conclusions: All four sCT methods enabled accurate proton dose calculation and preserved the anatomy, making them of value for adaptive proton therapy.

Keywords: Adaptive proton therapy; Cone-beam computed tomography; Head-and-neck cancer; Proton dose calculation; Synthetic computed tomography.

Fig. 1

A) Axial slices of the repeated CT (reCT), CBCT, and the four types of synthetic CTs (cCBCT, aCT, vCT, and DLsCT) are shown for two patients; one with good anatomical agreement between the reCT and CBCT (Patient 1) and one with poor agreement (Patient 2). The green arrows points to anatomical differences between the reCT and CBCT for Patient 2: Additional tissue is present on the reCT (red) compared to the CBCT (light blue). The red arrow points to streaking artifacts on the cCBCT also seen on the CBCT, and the yellow arrows point to residual artifacts on the DLsCT. For the aCT, the overriding with air and soft tissue is shown by magenta and yellow structures, respectively. The green insert shows a zoom of a region overridden with soft tissue on the aCT (CT number override of 40 HU), which is disconnected from the body outline. B) Sagittal and coronal views of Patient 2. The CBCT field-of-view and scan length is indicated on the synthetic CTs by a red box. The orange arrows point to a discontinuity resulting from the stitching technique used for the aCT and DLsCT. The cyan insert on the vCT shows a zoom of a region where voxels from the cCBCT have been used for overriding.

Fig. 2

A) Color-blend of the CBCT (blue) and planning CT or synthetic CTs (red) for a patient with nasogastric tube inserted after the planning CT acquisition. On the aCT, the CT number override with air and soft tissue is shown by magenta and yellow structures, respectively. The cCBCT and DLsCT correctly replicate the nasogastric tube. The aCT fails to restore the nasogastric tube (green arrows), while the vCT partially recovers the nasogastric tube through overriding with the cCBCT (orange arrow). B) Sagittal views of the CBCT, cCBCT, aCT, vCT, and DLsCT for a patient where the usage of cCBCT in seemingly mismatching low-density regions for the vCT leads to the introduction of large intensity artifacts (blue arrows).

Fig. 3

Violin plots of the difference in mean CT number (left) and standard deviation (std; right) for the regions-of-interest placed in muscle (top row), fat (middle row), and bone (bottom row) tissues on the repeated CT and synthetic CTs. The white dot on each violin plot indicates the median value.

Fig. 4

Violin plots of the difference in dose-volume-histogram (DVH) parameters (in Gray (Gy) or percent point (pp)) for the worst-case scenario of the robust evaluation (0% and ±3.5% range uncertainty; top) and nominal scenario (bottom) re-calculated on the repeated CT and synthetic CTs. The white dot on each violin plot indicates the median value.