Clinical Application of a Low-Dose CBCT AI Model
Cone Beam CT (CBCT) is an imaging modality used in interventional digital subtraction angiography (DSA). It produces three-dimensional images via cone-beam X-ray scanning and computer reconstruction. Clinically, CBCT guides puncture for pulmonary and hepatic lesions and evaluates post-intervention outcomes in liver cancer and cerebrovascular diseases. However, CBCT-guided interventions carry high patient radiation exposure; dose reduction often degrades image quality and impairs procedural results. Studies report that each 100 mGy radiation increment elevates cancer risk by 1.96-fold. Artificial intelligence enables low-dose CBCT. Our prior DeepPriorCBCT model embedded anatomical priors using neural discrete representation learning, reducing thoracic CBCT dose to one-sixth of routine protocols while preserving image quality. We further developed DeepPriorCBCT-V2 using 55 000 pre-reconstruction CBCT datasets covering brain, thorax and abdomen. This multi-organ model maintains image quality at one-sixth standard radiation dose. Nevertheless, its real-world clinical performance remains unvalidated. We therefore designed this prospective multicenter randomized controlled trial to evaluate the clinical applicability of DeepPriorCBCT-V2.
Age ≥18 years. [+2]
History of high-dose radiation exams or treatments. [+4]