Enhancing Lung Tumor Visibility on 2D X-Ray Imaging Utilizing Prior CT Information
Abstract
Purpose
Lung tumors are poorly visible on conventional on-board 2D X-ray imaging due to limited soft-tissue contrast and overlapping anatomy, posing a challenge for patient setup on proton therapy systems equipped with 2D X-ray imaging. Our goal is to develop a novel approach to enhance lung tumor visibility on 2D X-ray by leveraging prior information from simulation CT. This approach has potential to enable real-time, marker-less lung tumor tracking.
Methods
We developed a subtraction imaging pipeline and validated it using 4DCT and on-board X-ray images. In radiotherapy for lung cancer, tissues outside the thoracic cavity remain relatively static during treatment and are referred to as non-essential (NE) tissues, which were extracted from simulation CT. A patient-specific neural network–based 3D/2D registration was employed to align simulation CT with treatment-day X-ray images. Digitally reconstructed radiographs (DRRs) of the NE tissues were generated to simulate their signal contributions to 2D X-ray images, which were subtracted from treatment-day X-ray to create subtraction 2D X-ray images. Tumor visibility was quantitatively evaluated using tumor-to-background contrast (TBC) and contrast-to-noise ratio (CNR).
Results
Subtraction 2D X-ray significantly enhanced tumor visibility. A Feret diameter 6.4 cm tumor in the upper left lung was totally invisible with conventional X-ray from either of the 2 X-ray sources (A and B). The tumor was clearly visible on the subtraction images with boundary clearly delineated. For X-ray Source A, TBC increased from −0.12 to 2.04 and CNR increased from −0.39 to 1.33, corresponding to absolute improvements of +2.16 and +1.72, respectively. For X-ray Source B, TBC and CNR improved from 0.09 to 2.23 and from 0.36 to 1.22, yielding absolute improvements of +2.14 and +0.86, respectively.
Conclusion
This study demonstrates the clinical potential of subtraction 2D X-ray imaging to provide direct tumor visualization on 2D X-ray, enabling real-time marker-less lung tumor tracking.