Author: Jadon Buller, Zhuoran Jiang, Yankun Lang, Lei Ren, Leshan Sun, Liangzhong Xiang, Yifei Xu 👨🔬
Affiliation: University of Maryland School of Medicine, University of California, Irvine, University of California, Stanford University 🌍
Purpose: Electroacoustic tomography (EAT) and Protoacoustic (PA) imaging are novel modalities for treatment verification of electroporation and proton therapy. However, the limited acquisition angle in both EAT and PA imaging induces severe image distortion, limiting their verification accuracy. This study aims to develop SAM-Med3D, a 3D segmentation large model, to enhance PA and EAT images to correct image distortion to achieve precision treatment verification.
Methods: The SAM-Med3D model was enhanced by modifying its encoder-decoder architecture. A global-local feature fusion strategy in the encoder preserved structural details, while a lightweight decoder incrementally improved image resolution. The model was trained to refine distorted acoustic images to match ground truth. The EAT dataset consisted of 30 scans of water and tissue phantom (chicken and pork) using various electrode settings (2 electrodes, various distances, and angles ) and voltages (600–1000 Volts/cm). 120 views equally distributed over 360° were acquired using a matrix array in each scan. The PA imaging dataset consisted of 126 prostate proton therapy patients . The protoacoustic signals were simulated based on CT and planned dose for an ultrasound matrix array (64 64 size, 500kHz central frequency) placed under the perineum. Data splitting (training and testing) for EAT and PA images is 1:1 and 3:1, respectively.
Results: Image enhancement was evaluated on 15 EAT and 40 prostate PA images. Metrics included PSNR/SSIM (38.24/0.98 for EAT) and RMSE/SSIM (36.01/0.95 for PA). The enhanced pressure map demonstrated high accuracy, with a total processing time of ~1s.
Conclusion: The application of SAM-Med3D in PA and EAT 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.