FBFormer: interventional ultra-sparse CT reconstruction with image prior using feature back-projection and...

Area of Science:

• Medical Imaging
• Computer Vision
• Radiology

Background:

• CT-guided interventions are crucial but involve high radiation doses.
• Prolonged scan times and numerous images contribute to patient radiation exposure.

Purpose of the Study:

• To develop an alternative to intraoperative CT guidance using X-ray imaging.
• To reconstruct interventional CT by integrating preoperative images with intraoperative X-ray data.
• To reduce radiation exposure while maintaining reconstruction quality.

Main Methods:

• Feature back-projection Transformer (FBFormer) framework for tomographic data reconstruction.
• Utilizes ultra-sparse intraoperative X-ray images (less than 10 views).
• Employs a mixed UNet and PVT encoder for feature extraction and a feature back-projection module for 2D-3D conversion.

Main Results:

• FBFormer demonstrates effectiveness on simulated 4DCT and clinical validation datasets.
• Achieves superior reconstruction performance metrics compared to other methods.
• Outperforms existing compelling reconstruction techniques.

Conclusions:

• The FBFormer framework can significantly reduce radiation exposure in interventional procedures.
• Enhances clinical reliability of interventional CT reconstruction from sparse X-ray projections.
• Integrates image priors and a geometrically consistent feature back-projection module for high-quality results.