[Advances in low-dose cone-beam computed tomography image reconstruction methods based on deep learning].

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Cone-beam computed tomography (CBCT) is essential in dentistry, surgery, and radiotherapy.
• Repeated CBCT scans increase patient radiation dose and secondary malignant tumor risk.
• Low-dose CBCT reconstruction aims to mitigate these risks by improving image quality with reduced radiation.

Purpose of the Study:

• To systematically review deep learning-based methods for low-dose CBCT image reconstruction.
• To compare different deep learning network architectures for noise reduction, artifact removal, detail preservation, and efficiency.
• To explore emerging technologies like multimodal fusion and self-supervised learning in this domain.

Main Methods:

• Systematic review of deep learning approaches for low-dose CBCT.
• Comparison of image-domain, projection-domain, and dual-domain reconstruction techniques.
• Analysis of network architectures focusing on image quality and computational performance.

Main Results:

• Deep learning methods show promise in reducing noise and artifacts while preserving details in low-dose CBCT.
• Different network architectures offer varying trade-offs between image quality and computational efficiency.
• Emerging techniques like multimodal fusion and self-supervised learning may further enhance reconstruction performance.

Conclusions:

• Deep learning-based low-dose CBCT reconstruction is a rapidly advancing field with significant potential.
• Understanding the strengths and weaknesses of current methods is crucial for algorithm optimization.
• Further research and validation are needed to support widespread clinical adoption of these advanced techniques.