Real-time algorithm for Poissonian noise reduction in low-dose fluoroscopy: performance evaluation

Abstract

Background: Quantum noise intrinsically limits the quality of fluoroscopic images. The lower is the X-ray dose the higher is the noise. Fluoroscopy video processing can enhance image quality and allows further patient's dose lowering. This study aims to assess the performances achieved by a Noise Variance Conditioned Average (NVCA) spatio-temporal filter for real-time denoising of fluoroscopic sequences. The filter is specifically designed for quantum noise suppression and edge preservation. It is an average filter that excludes neighborhood pixel values exceeding noise statistic limits, by means of a threshold which depends on the local noise standard deviation, to preserve the image spatial resolution. The performances were evaluated in terms of contrast-to-noise-ratio (CNR) increment, image blurring (full width of the half maximum of the line spread function) and computational time. The NVCA filter performances were compared to those achieved by simple moving average filters and the state-of-the-art video denoising block matching-4D (VBM4D) algorithm. The influence of the NVCA filter size and threshold on the final image quality was evaluated too.

Results: For NVCA filter mask size of 5 × 5 × 5 pixels (the third dimension represents the temporal extent of the filter) and a threshold level equal to 2 times the local noise standard deviation, the NVCA filter achieved a 10% increase of the CNR with respect to the unfiltered sequence, while the VBM4D achieved a 14% increase. In the case of NVCA, the edge blurring did not depend on the speed of the moving objects; on the other hand, the spatial resolution worsened of about 2.2 times by doubling the objects speed with VBM4D. The NVCA mask size and the local noise-threshold level are critical for final image quality. The computational time of the NVCA filter was found to be just few percentages of that required for the VBM4D filter.

Conclusions: The NVCA filter obtained a better image quality compared to simple moving average filters, and a lower but comparable quality when compared with the VBM4D filter. The NVCA filter showed to preserve edge sharpness, in particular in the case of moving objects (performing even better than VBM4D). The simplicity of the NVCA filter and its low computational burden make this filter suitable for real-time video processing and its hardware implementation is ready to be included in future fluoroscopy devices, offering further lowering of patient's X-ray dose.

Keywords: Fluoroscopy; NVCA; Quantum noise; Real-time processing; VBM4D; Video denoising; X-ray dose reduction.

Fig. 1

Average autocorrelation function evaluated in space (black crosses) and over time (white squares). The fluoroscopy sequence of the aluminum phantom was used

Fig. 2

a A and b B coefficients estimated via the algorithm provided in ref [33]. Since A and B are estimated for each frame, the image frame number is reported on the x-axis

Fig. 3

Expected value–variance relationship as estimated in space according to Ref. [33] and in time using the fluoroscopy sequence of the aluminum phantom

Fig. 4

a Raw image; b VBM4D with algebraic inversion; c NVCA, 3 × 3 × 3 mask T = 2σ; d moving average 3 × 3 × 3 mask; e NVCA, 5 × 5 × 5 mask T = 2σ; f moving average 5 × 5 × 5 mask. The yellow vertical line in a indicates the profile selected for the spatial resolution evaluation

Fig. 5

Profiles across the edge outlined in Fig. 4a for the raw image, the image filtered via NVCA algorithm (7 × 7 × 7 mask T = 2σ) and that filtered with a 7 × 7 × 7 average filter. Continuous lines represent the fitting curves adopted. The fitting parameter R² is 0.9782 for the raw data, 0.9970 in the case of the average filter and 0.9977 in the case of the NVCA filter

Fig. 6

FWHM values for the evaluated filters and the raw data. The W/O label corresponds to the moving average filters

Fig. 7

CNR values for the NVCA filters, VBM4D and raw data. The W/O label corresponds to the moving average filters

Fig. 8

FSIM indices obtained using the NVCA, VBM4D and the average filter (labeled W/O) denoisers. It was evaluated on the static image of the aluminum step phantom; the reference image was obtained by averaging over time all the raw fluoroscopic frames

Fig. 9

Edge profiles across the edge of the moving object in the sequence of the digital phantom for 5 × 5 × 5 average filter, VBM4D and 5 × 5 × 5 NVCA filter (T = 2σ). Insert speed = 1 pixel/frame

Fig. 10

FWHM evaluated across a moving edge in the digital phantom for 5 × 5 × 5 average filter, VBM4D and 5 × 5 × 5 and 7 × 7 × 7 NVCA filter (T = 2σ)

Fig. 11

The enlargement of the real fluoroscopic image including the radiopaque needle: a raw image, the yellow vertical line was manually placed across the needle to evaluate blur, b image filtered with VBM4D and algebraic inversion of the Anscombe transform; image filtered with NVCA (T = 1.5σ) with mask size of b 5 × 5 × 1 pixels, d 5 × 5 × 3 pixels, e 5 × 5 × 5 pixels and f 5 × 5 × 7 pixels

Fig. 12

Profile across the needle outlined in Fig. 11a

Fig. 13

Schematic of the home-made test object used for the noise characterization

Fig. 14

A frame of the aluminum step phantom fluoroscopic sequence. The manually selected regions of interest are shown in yellow: the ROI1 was chosen for the FWHM estimation and the ROIA and ROIB were chosen for CNR estimation. The two dark circles in the middle correspond to the two metallic screws and bolts that hold the aluminum sheets together

Fig. 15

Digital phantom with a moving rectangular insert. The red arrows indicate the moving direction of the insert over the consecutive frames. The yellow ROI outlines the region for the evaluation of the FWHM on the insert edge

Fig. 16

A fame of the surgical fluoroscopy sequence (left) and an enlargement of its central region (right). The radiopaque needle appears at the top of the image enlargement