br The purpose of this work was to compare interobserver
The purpose of this work was to compare interobserver agreement in uterine segmentation and observer confidence in segmentation on CBCT, US, and CBCT-US fused images to determine the optimal imaging method for target local-ization during cervix RT.
Methods and Materials
Patients and treatment
Eleven patients with biopsy-proven diagnosis of locally advanced cervix cancer were included in this National Health Services Research Ethicseapproved study (refer-ence: 15/LO/1438). Fe´de´ration Internationale de Gyne´co-logie Obste´trique stage distribution was as follows: IIA Z 1, IIB Z 9, IIIB Z 0, IVA Z 1. The mean patient age was 51 ( 16) years. Patients were treated with radical chemoradiotherapy from February 2016 to May 2017. Pa-tients were instructed to drink 350 mL water after complete Poly(I:C) voiding 1 hour before treatment as per institutional protocol to maintain interfractional bladder volume and consistent setup. There were no bowel preparation in-structions. Kilovoltage CBCT images were acquired immediately before treatment for online correction based on bony registration on days 1 to 3 and weekly thereafter unless there was a systematic error of >5 mm or clinical indication, in which case they were acquired more frequently. 3D US of the uterus were acquired using the Clarity system (Elekta Ltd, Stockholm, Sweden) immedi-ately before CBCT acquisition at 4 to 6 treatment sessions after patient setup on the RT treatment couch.
3D US scans (5 MHz center frequency, mechanically swept probe) were acquired using the Clarity system. The Clarity
system is described in detail elsewhere,26 but briefly, it is a standard US imaging system that is integrated into the RT clinic via infrared tracking, whereby the position of the probe (and the corresponding US images) with respect to the isocenter of the treatment room is known with sub-millimeter accuracy. US operators (either a clinical oncol-ogist or a therapeutic radiographer) applied a thick layer of US gel to the probe and scanned the uterus trans-abdominally using the smallest probe pressure possible to minimize soft-tissue deformation while still obtaining clear visualization of the uterus.
CBCT imaging (Elekta Ltd) was performed immediately after US scanning, with no more than a 5-minute interval between US and CBCT scans. CBCT imaging parameters were 120 kVp and 80 mAs with 350 projections and a bowtie filter.
Sixty-four US-CBCT image pairs (128 images in total) were obtained as part of grasslands biome study. Two image pairs were excluded because of US operator errors in the probe cali-bration step of the Clarity QA, which caused misregistra-tion of US images to the treatment room isocenter. Five image pairs were excluded because of failure to save US or CBCT scans. Of the 57 remaining image pairs, 40 were randomly selected for analysis, leaving the remainder (17 image pairs) exclusively for observer training purposes.
CBCT images were registered to the planning computed to-mography (CT) scan using the Synergy bone match algo-rithm. Translational and rotational error was summarized as a translational couch shift. After export to the Clarity work-station, these translational shifts were applied to each CBCT scan to replicate match to planning CT. The infrared tracking technology provided by the Clarity system enabled spatial registration between the US images and the CBCT images. In the offline Clarity workstation, the same translational moves were applied to the corresponding US images.
A software application was written in Matlab (MathWorks, Natick, MA) to enable presentation of CT-CBCT, CT-US, and CBCT-US registration for assessment by observers. The pixel size (in the format of [superoinferior direction, anteroposterior direction]) was [2.5, 1] mm for CBCT images, [0.58, 0.58] mm for US, and [3, 1] mm for CT. Registered images were superimposed over one another, with user-adjustable sliders for 3D slice selection, trans-parency, and windowing in both the sagittal and axial ori-entations. This functionality was achieved by interpolating the image in the registration with the larger pixels to a grid matching the sampling density of the image with the smaller pixels. Note that any references to “CBCT” or “US” hereafter imply registration with the planning CT, as