2131 - 3D Segment Anything Model (SAM) for 3D Protoacoustic Imaging Enhancement and <em>In Vivo</em> Dose Verification in Proton Therapy

02:30pm - 04:00pm PT

Hall F

Screen: 26

POSTER

Presenter(s)

Lei Ren, PhD - University of Maryland Cancer Center, Baltimore, MD

Y. Lang¹, Z. Jiang², L. Sun³, Y. Xu³, J. Buller³, P. T. Tran⁴, L. Xiang³, and L. Ren⁴; ¹Department of Radiation Oncology, University of Maryland School of Medicine, Baltimore, MD, ²Duke University, Durham, NC, ³University of California, Irvine, CA, ⁴University of Maryland School of Medicine, Baltimore, MD

Purpose/Objective(s):

Proton therapy is an advanced form of radiation treatment known for its potential benefit in targeting tumors while minimizing damage to surrounding healthy tissue. However, the accuracy of proton dose delivery can be compromised by uncertainties in patient positioning and anatomical changes, necessitating real-time dose verification methods. Protoacoustic (PA) imaging has emerged as a promising technique for this purpose, capturing acoustic signals generated during proton interactions to reconstruct dose distributions. This study aims to explore the application of large-scale model, in PA image reconstruction for improved accuracy in proton dose verification. The primary goal is to address the limited-angle issue inherent in PA imaging, thereby enhancing the precision of dose verification during proton therapy.

Materials/Methods:

The SAM-Med3D model was adapted for PA imaging by modifying its encoder-decoder architecture. A global-local feature fusion strategy was implemented in the encoder to preserve structural details, while a lightweight decoder replaced the original prompt-based decoder to improve image resolution incrementally. The proposed method was evaluated on a dataset derived from prostate cancer patients, utilizing both CT images and clinical treatment plans.

Results:

Our proposed model achieved an average root mean square error (RMSE) of 0.024 (p<0.05), and an average structural similarity index measure (SSIM) of 0.938 (p<0.05). Qualitative results also demonstrated that our approach acquired better imaging quality with more fine details reconstructed when comparing with the baseline models. Dose verification achieved an average RMSE of 0.017 (p<0.05), and an average SSIM of 0.954 (p<0.05). Gamma index evaluation demonstrated a high agreement 95.4% (p<0.05) and 96.2% (p<0.05) for 1%/3 mm and 1%/5 mm, respectively, between the predicted and the ground truth dose maps. Our approach approximately took 1 second to complete the enhancement for each patient, demonstrating its feasibility for online 3D dose verification for proton therapy.

Conclusion:

The application of SAM-Med3D in PA imaging offers a viable solution to the limited-angle problem, facilitating high-precision 3D dose verification in proton therapy. This study underscores the potential of large-scale models like SAM-Med3D for advancing image enhancement techniques in medical imaging, particularly in the context of real-time radiation therapy monitoring.