Premium
SU‐GG‐I‐83: Temporal Filtering of Noise in Low‐Dose X‐Ray Fluoroscopy
Author(s) -
Wang J,
Zhu L,
Chai A,
Xing L
Publication year - 2008
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.2961481
Subject(s) - fluoroscopy , imaging phantom , noise reduction , mathematics , noise (video) , shot noise , gaussian noise , regularization (linguistics) , image noise , image quality , algorithm , computer science , artificial intelligence , physics , optics , image (mathematics) , detector , nuclear physics
Purpose: To improve quality of low‐dose X‐ray fluoroscopic images using statistics‐based restoration algorithm so that the patient fluoroscopy can be performed with much reduced radiation dose. Method and Materials: We first studied the noise properties of low‐dose X‐ray fluoroscopic images through repeated measurements using a Varian Acuity simulator. Noise in X‐ray fluoroscopic images can be modeled as Poisson noise plus background electronic Gaussian noise and the mean‐variance of the noise can be described by an analytical formula. Based on the noise model in fluoroscopic images, a penalized weighted least‐squares (PWLS) objective function was constructed to restore fluoroscopic images acquired with low mAs protocol. Furthermore, the Karhunen‐Loève (KL) transform was utilized to consider correlations among neighboring frames of fluoroscopy. The KL transform manipulates a sequence of correlated measurements into an uncorrelated, ordered principle component series and, therefore, provides a unique means for de‐correlation, feature extraction and noise reduction. After the KL transform, the regularization parameter in PWLS criterion will be inverse proportional to its corresponding eigenvalue. Such a choice is favorable because the regularization parameter varies adaptively according to the signal‐to‐noise ratio of that component. A smaller KL eigenvalue is associated with a component having a lower SNR; therefore, a larger regularization value should be used to penalize this noisier data. Results: We tested the proposed K‐L domain PWLS noise reduction algorithm using an anthropomorphic chest phantom. Low‐dose fluoroscopic images were acquired with X‐ray tube current of 10 mA and duration of x‐ray pulse 2 ms. In the image restored by the proposed KL domain PWLS algorithm, noise is greatly suppressed while fine structures are well preserved. Conclusion: Experiment studies show that image quality of low‐dose X‐ray fluoroscopic image can be dramatically improved by using statistics‐based temporal filtration. The proposed noise reduction technique shows potential for dose reduction of X‐ray fluoroscopy.