Image sequence filtering in quantum-limited noise with applications to low-dose fluoroscopy

Area of Science:

• Medical Imaging
• Signal Processing
• Radiology

Background:

• Clinical angiography necessitates numerous X-ray images, posing risks to patients and medical personnel.
• Reducing radiation dosage compromises image quality, particularly due to signal-dependent, Poisson-distributed noise.
• Existing noise models for single images are insufficient for dynamic fluoroscopy sequences.

Purpose of the Study:

• To develop a novel noise model for signal-dependent, Poisson-distributed noise in low-dose X-ray imaging.
• To propose advanced stochastic temporal filtering techniques for enhancing fluoroscopy image sequences.
• To integrate displacement field estimation with temporal filtering to prevent motion-induced object blur.

Main Methods:

• A new signal-dependent, Poisson-distributed noise model was developed, generalizing existing single-image models.
• Stochastic temporal filtering algorithms were designed to leverage temporal correlations in fluoroscopy sequences.
• Displacement field estimation was incorporated into the filtering process to align image sequences along motion trajectories.

Main Results:

• The proposed noise model accurately represents quantum mottle in low-dose X-ray imaging.
• Stochastic temporal filtering significantly enhances the quality of clinical fluoroscopy sequences.
• Integrated displacement estimation effectively prevents object blur in dynamic imaging scenarios.

Conclusions:

• The developed noise model and temporal filtering techniques offer a viable solution for improving image quality in low-dose clinical angiography.
• These advancements can lead to reduced radiation exposure without sacrificing diagnostic accuracy.
• The integrated approach addresses key challenges in processing dynamic medical imaging sequences.