ESTRO 2024 - Abstract Book

S3880

Physics - Image acquisition and processing

ESTRO 2024

Clinical studies have shown that tumor cells preferentially invade muscles along the muscle fiber orientation [1]. However, radiotherapy practice does not take into account the specific muscle fiber orientation when delineating the clinical target volume (CTV) for sarcoma patients with suspected invasion into adjacent muscles. We propose a method to construct the CTV using intramuscular structural information. The method is tested on a patient diagnosed with chondrosarcoma, a malignant bone tumor surrounded by muscle tissue located in the sacrum.

Material/Methods:

Unlike CT images, which have poor soft tissue contrast, high-resolution cryosection photographic images of a cadaver, available at the Visible Human Project [2], can provide muscle fiber orientation through the contrast of muscle fascicles with fat and connective tissue. Step 1: We perform deformable image registration (DIR) between the 3D reconstructed cryosection image and the patient CT image; and warp the anatomical cryosection image data to the patient. Specifically, the muscles surrounding the gross tumor volume (GTV), gluteus maximus, gluteus minimus, gluteus medius, and piriformis, are mapped to the patient image. Step 2: A structure tensor (second moment matrix derived from the image gradient in each direction) is computed in each voxel of the warped cryo image to quantify muscle fiber orientation. The tensor encodes the orientation of the muscle fibers, and the ratio of the tensor eigenvalues can model the extent of tumor infiltration in the dominant fiber direction versus other directions. The tensor eigenvalues can then be adjusted by user-specific values to reflect clinical considerations [3]. Step 3: The shortest path distance in each voxel around the GTV is computed considering muscle fiber direction. The shortest path calculation takes as input the GTV, the cryo-image-derived structure tensor in each muscle, and barrier structures that are considered impermeable to tumor invasion. The resulting “anatomy-aware” CTV margin is defined as a level set in the shortest path distance map. Fig. 1 illustrates the workflow for a chondrosarcoma patient with the warped cryosection image resulting from DIR and resulting shortest path distance map. Fig. 2 compares a classical CTV generated with a generic 1 cm and 2 cm margin for muscle and non-muscle tissue, to three anatomy-aware CTVs (CTV1-3) of the same volume (in cc) but with different values of the eigenvalue ratio. All contours are overlayed on the patient's CT image. CTV1, CTV2, and CTV3 were generated with an eigenvalue ratio of 0.25, 0.05, and 0.01, meaning that the preferred spread along the muscle fiber direction was modeled to be larger by a factor of 4, 20, and 100, respectively. As indicated by the white arrows, the CTV margin within the muscle increases inversely with the eigenvalue ratio, especially in the gluteus maximus muscle. Conversely, for CTV2 and CTV3, the CTV margin is significantly reduced in areas outside the muscle tissue, such as the areas superior and inferior to the GTV. Compared to the classical CTV, the target volume reduction in non muscle tissue was 4% and 14% for CTV2 and CTV3, respectively. Results: