Hybrid-Supervised Deep Learning for Domain Transfer 3D Protoacoustic Image Reconstruction

TL;DR

This paper introduces a hybrid-supervised deep learning approach with a two-stage strategy to improve 3D protoacoustic image reconstruction for proton therapy dose verification, significantly reducing artifacts and processing time.

Contribution

It proposes a novel transformer-based reconstruction network combined with a 3D U-net enhancement, trained with hybrid supervision and transfer learning, to address limited view artifacts in protoacoustic imaging.

Findings

Achieved high image quality with RMSE 0.0292 and SSIM 0.9618.

Reduced processing time to 6 seconds for online dose verification.

Demonstrated high dose verification accuracy with RMSE 0.018 and SSIM 0.9891.

Abstract

Protoacoustic imaging showed great promise in providing real-time 3D dose verification of proton therapy. However, the limited acquisition angle in protoacoustic imaging induces severe artifacts, which significantly impairs its accuracy for dose verification. In this study, we developed a deep learning method with a Recon- Enhance two-stage strategy for protoacoustic imaging to address the limited view issue. Specifically, in the Recon-stage, a transformer-based network was developed to reconstruct initial pressure maps from radiofrequency signals. The network is trained in a hybrid-supervised approach, where it is first trained using supervision by the iteratively reconstructed pressure map and then fine-tuned using transfer learning and self-supervision based on the data fidelity constraint. In the Enhance-stage, a 3D U-net is applied to further enhance the image quality with supervision from the ground truth pressure map. The final protoacoustic images are then converted to dose for proton verification. The results evaluated on a dataset of 126 prostate cancer patients achieved an average RMSE of 0.0292, and an average SSIM of 0.9618, significantly out-performing related start-of-the-art methods. Qualitative results also demonstrated that our approach addressed the limit-view issue with more details reconstructed. Dose verification achieved an average RMSE of 0.018, and an average SSIM of 0.9891. Gamma index evaluation demonstrated a high agreement (94.7% and 95.7% for 1%/3 mm and 1%/5 mm) between the predicted and the ground truth dose maps. Notably, the processing time was reduced to 6 seconds, demonstrating its feasibility for online 3D dose verification for prostate proton therapy.