ESTRO 2025 - Abstract Book

S3068

Physics - Image acquisition and processing

ESTRO 2025

tumour tracking precision and improve the accuracy of radiotherapy planning. However, a huge amount of cine-MR data needs to be analysed. This study aims to evaluate the effectiveness of a computer vision algorithm, Scale Invariant Feature Transform (SIFT), for tracking tumour motion using cine-MR images, compared to traditional image registration methods, such as rigid registration (RR) and deformable registration (DR). Material/Methods: Cine-MR images were collected from two patients undergoing lung Stereotactic Body Radiation Therapy (SBRT) as part of a prospective clinical trial. For each patient, a cine-MR sequence consisting of 450 frames was acquired. These images were processed and analysed to track tumour motion in the sagittal plane using three methods: SIFT, RR and DR. The SIFT method identified and tracked tumour motion using extracted keypoints, while RR and DR methods aligned images through rigid and deformable transformations, respectively. Tumour motion was quantified and visualised using probability-of-presence (POP) maps. To evaluate the accuracy of image alignment, similarity metrics were used: mean squared difference (MS), cross-correlation (CC), and mutual information (MI). These metrics measured the similarity between the transformed reference frame and each subsequent frame within the region of interest containing the tumour. Results: The highest similarity metric scores were achieved by DR, with median CC values of 0.88 and 0.95 for the two patients, compared to 0.79 and 0.85 for SIFT, and 0.86 and 0.91 for RR, respectively. DR captured the largest tumour displacements (17.9 mm and 13.8 mm for the two patients) while RR and SIFT provided slightly lower displacement estimates but successfully captured the overall motion patterns. POP maps further illustrated that DR provided more comprehensive coverage of the tumour's motion range, particularly during the extremes of the breathing cycle, seen for patient 2 in Figure 1. The displacements can be seen for both patients in Figure 2.