Sunday 02 February 2025
A new study has shed light on how combining preoperative CT scans with intraoperative CBCT scans can improve medical imaging tasks such as semantic segmentation. The researchers used a dataset of synthetic CBCT volumes generated by simulating Digital Reconstructed Radiographs from high-quality CT scans.
The team found that enriching the CBCT scans with roughly aligned, preoperative CT images significantly improved segmentation performance in 18 out of 20 setups investigated. This suggests that adding high-quality CT scans can partly mitigate the decreasing image quality of intraoperative CBCT scans.
The study also discovered that the worse the CBCT quality, the more can be gained by adding high-quality CT scans. This implies that there may be some implicit registration taking place, where the computer is able to automatically align the two images to improve the results.
One potential application of this research is in liver tumor segmentation, a crucial task in medical imaging. By combining CBCT and CT scans, doctors may be able to more accurately identify tumors and develop more effective treatment plans.
The researchers used a 3D U-Net architecture to segment the images, which involved training the model on a dataset of paired CBCT and CT scans. They found that the multimodal approach outperformed the baseline results in most cases, with the only exception being liver tumor segmentation with a high-quality CBCT scan.
The study’s findings have implications for other medical imaging applications beyond liver tumor segmentation. The researchers suggest that this form of multimodal learning could be theoretically applicable to other architectures and tasks.
The project was funded by several organizations, including the Austrian Research Promotion Agency and the county of Salzburg. The results were published in a recent issue of Medical Image Analysis.
In this study, the authors have demonstrated the potential benefits of combining preoperative CT scans with intraoperative CBCT scans for improving medical imaging tasks such as semantic segmentation. By leveraging the strengths of both modalities, doctors may be able to develop more accurate and effective treatment plans for patients.
Cite this article: “Enhancing Medical Imaging with Multimodal Learning: Combining CT and CBCT Scans”, The Science Archive, 2025.
Medical Imaging, Ct Scans, Cbct Scans, Semantic Segmentation, Multimodal Learning, Liver Tumor Segmentation, 3D U-Net Architecture, Image Analysis, Preoperative Imaging, Intraoperative Imaging
