DeepSparse: A Foundation Model for Sparse-View CBCT Reconstruction | JoVE Visualize

Area of Science:

Medical Imaging and Radiation Oncology
Artificial Intelligence in Healthcare
Computational Imaging

Background:

Cone-beam computed tomography (CBCT) is essential for 3D medical imaging but involves high radiation exposure, posing risks, especially to vulnerable groups.
Sparse-view reconstruction aims to lower radiation dose by using fewer projections, but current methods struggle with computational cost and dataset adaptability.
Existing techniques for sparse-view CBCT reconstruction face limitations in efficiency and generalizability across diverse imaging datasets.

Purpose of the Study:

To introduce DeepSparse, the first foundation model specifically designed for sparse-view CBCT reconstruction.
To address the limitations of existing methods by developing a computationally efficient and generalizable solution.
To enable safer and more effective CBCT imaging by reducing radiation dose without compromising image quality.

Main Methods:

Developed DeepSparse, incorporating DiCE (Dual-Dimensional Cross-Scale Embedding) to integrate multi-view 2D and multi-scale 3D features.
Introduced the HyViP (Hybrid View Sampling Pretraining) framework for pretraining on extensive datasets with both sparse and dense projections.
Implemented a two-step finetuning strategy to adapt and refine the model for specific new datasets.

Main Results:

DeepSparse demonstrated superior reconstruction quality compared to current state-of-the-art methods in extensive experiments.
Ablation studies confirmed the effectiveness of the proposed DiCE and HyViP components.
The model achieved significant improvements in image quality for sparse-view CBCT reconstruction.

Conclusions:

DeepSparse represents a significant advancement in sparse-view CBCT reconstruction, offering a safer alternative to traditional high-dose imaging.
The developed foundation model and pretraining strategy enhance generalizability and computational efficiency.
This work paves the way for reduced radiation exposure in CBCT procedures, improving patient safety and diagnostic accuracy.