Accurate positioning for head and neck cancer patients using 2D and 3D image guidance

Abstract

Our goal is to determine an optimized image-guided setup by comparing setup errors determined by two-dimensional (2D) and three-dimensional (3D) image guidance for head and neck cancer (HNC) patients immobilized by customized thermoplastic masks. Nine patients received weekly imaging sessions, for a total of 54, throughout treatment. Patients were first set up by matching lasers to surface marks (initial) and then translationally corrected using manual registration of orthogonal kilovoltage (kV) radiographs with DRRs (2D-2D) on bony anatomy. A kV cone beam CT (kVCBCT) was acquired and manually registered to the simulation CT using only translations (3D-3D) on the same bony anatomy to determine further translational corrections. After treatment, a second set of kVCBCT was acquired to assess intrafractional motion. Averaged over all sessions, 2D-2D registration led to translational corrections from initial setup of 3.5 ± 2.2 (range 0-8) mm. The addition of 3D-3D registration resulted in only small incremental adjustment (0.8 ± 1.5 mm). We retrospectively calculated patient setup rotation errors using an automatic rigid-body algorithm with 6 degrees of freedom (DoF) on regions of interest (ROI) of in-field bony anatomy (mainly the C2 vertebral body). Small rotations were determined for most of the imaging sessions; however, occasionally rotations > 3° were observed. The calculated intrafractional motion with automatic registration was < 3.5 mm for eight patients, and < 2° for all patients. We conclude that daily manual 2D-2D registration on radiographs reduces positioning errors for mask-immobilized HNC patients in most cases, and is easily implemented. 3D-3D registration adds little improvement over 2D-2D registration without correcting rotational errors. We also conclude that thermoplastic masks are effective for patient immobilization.

Figure 1

The percentile distributions of the 3D lengths of the final shifts: manual 2D‐2D followed by manual 3D‐3D (black) and the couch shifts determined by manual 2D‐2D (blue) are shown (a). The percentile distributions of the translational couch shifts determined by manual 2D‐2D registration (blue) from the initial patient position, and further shifts from the following manual 3D‐3D registration (red) for all 54 sessions in left‐right, X (b), posterior‐anterior, Y (c), superior‐inferior, Z (d) and the 3D length of the error vectors (e).

Figure 2

Average translational (a) and rotational (b) setup errors calculated through automatic six DoF registration at treatment position in X (black), Y (red) and Z (green) for all 54 imaging sessions. The bars are the range of the errors.

Figure 3

Sagittal (a) and transverse (b) views of patient #1 at treatment position (CBCT, cyan) on one imaging day superimposed with the simulation position (simulation CT, red). The patient's chin was dropped from the simulation position, and patient's anatomy was deformed. Note that the FOV of CBCT covered only part of the whole target volume for this patient.

Figure 4

Average translational (a) and rotational (b) intrafractional errors calculated with automatic six DoF registration in X (black), Y (red) and Z (green) for all 47 post‐treatment kVCBCT. The bars are the range of the errors.