ESTRO 2024 - Abstract Book

S3902

Physics - Image acquisition and processing

ESTRO 2024

Tumour tracking in the lung is challenging due to the intrinsic poor contrast of X-ray images and obstruction from bony anatomy [1]. Dual Energy (DE) imaging is a method that entails imaging at two different energies, followed by a weighted logarithmic subtraction to increase the soft tissue contrast [2]. This study aims at evaluating the performance of an experimental platform for DE imaging with integrated infrared tracking, in order to benchmark DE tracking accuracy vs. ground truth motion information at sub-millimetre resolution.

Material/Methods:

The mobile ImagingRing (IRm) designed by medPhoton and a dynamic phantom (CIRS 008M MRI-Linac Phantom) were used to acquire DE cone-beam CT (CBCT) projections, by operating the X-ray generator in pulsed mode between 80keV (15 mA, 17 ms) and 120keV (15 mA, 12 ms) (Figure 1A). The infrared tracking device integrated in the ImagingRing gantry, calibrated to the X-ray imaging coordinate system, was used to track a rigid body attached to the moving target of the phantom (Figure 1B), with optical detection synchronized to the acquired X-ray images. X-ray frames with alternating energies at consecutive pulses were acquired, with pork ribs attached to the phantom to simulate a human anatomy over a rotation in 120 seconds (165 ms between consecutive frames). In order to check the reliability of the integrated optical tracking device, three DE-CBCT scans were acquired with the target set to move automatically with a cos4 motion trace with ±10mm amplitude at two different breathing periods (2.55 and 4 s), along with a third static scan. To investigate the performance for different materials, the CIRS phantom was adjusted to accommodate for six different target materials: adipose, breast, muscle, bone, liver, and cortical bone tissues. A total of twelve DE-CBCTs were acquired: for each target, two DE-CBCT scans were taken, with the second scan capturing motion of the target approximately 2 cm longitudinally apart from the first scan, in order to test different overlap with respect to the ribs. Two frames which contained the maximum overlap between the ribs and the target were chosen from each experiment. To study the effect of different types of registration to correct the misalignment between frames, images were processed by considering deformable, rigid registration and a no registration approach. Multiple weighting factors, in the range [0.45, 0.9] were tested to perform the images weighted logarithmic subtraction.

The statistical significance of the result was calculated using the paired samples Wilcoxon test, applied to the resulting contrast to noise ratio (CNR) of the DE-enhanced weighted images.