Author: Michael Baine, Charles Enke, Yang Lei, Yu Lei, Ruirui Liu, Su-Min Zhou 👨🔬
Affiliation: Icahn School of Medicine at Mount Sinai, University of Nebraska Medical Center, Department of Radiation Oncology, University of Nebraska Medical Center 🌍
Purpose: This study presents a framework for generating synthetic CT images using a Cycle Diffusion model, which can be utilized to enhance needle conspicuity in ultrasound-guided prostate HDR brachytherapy.
Methods:
The proposed framework involves generating synthetic CT (sCT) from real-time ultrasound images acquired during HDR procedures using a Cycle Diffusion model. sCTs are generated from ultrasound images both with and without needle catheters to aid in needle insertion and reconstruction. A preliminary investigation was conducted using a dataset consisting of ultrasound (needle-free), CT, and MRI images from ten patients undergoing HDR prostate brachytherapy, along with a set of ultrasound and CT images acquired from a prostate training phantom. The Cycle Diffusion model was trained on this dataset, and the performance of the framework was subsequently validated.
Given the substantial contrast differences between the patient CT and ultrasound images, initial registration of the ultrasound images to the MRI images was performed, followed by registration of the ultrasound images to the CT images using the established CT-MRI registration chain. The generated sCT images were visually evaluated by comparing the registration between the sCT and the real CT images of both the patients and the phantom.
Results:
The registration of the sCT and CT images of the phantom appears excellent. The sampled line profiles of HU numbers in both the sCT and CT match very closely.
Conclusion:
This study presents a deep learning-driven approach for generating synthetic CT from ultrasound images during HDR brachytherapy, showing promising potential in enhancing needle conspicuity. Future efforts will focus on refining the proposed deep learning methodology by calibrating sCT with needle data.