Real-time low-dose region-of-interest fluoroscopy with spatially different bilateral filter

Abstract

A region-of-interest (ROI) fluoroscopy is effective low-dose technique that uses a filtration outside the ROI. This method reduces the dose exposure to both patients and operators while preserving the image quality of the ROI. But beam filtration by ROI filter increases noise and degrades image quality of the periphery of fluoroscopic images. To solve this problem, we developed an image processing algorithm for noise reduction and brightness equalization of ROI image. The noise reduction is performed by spatially different bilateral filtering to appropriately address different noise levels between the inner ROI and the periphery of the ROI. And brightness was equalized by weight mask multiplication using the precalculated weight mask before image processing. On the other hand, it would be desirable to achieve the image processing fast enough to accommodate a real-time fluoroscopic imaging. So we accelerated our image processing using GPU parallel computing. As a result, the noise level of the periphery of the ROI remarkably reduced and the dose exposure reduced more than 80% by ROI filter used in this study. And the elapsed time for our image processing was reduced from 3.43 second (CPU) to 0.111 second (GPU). By applying this GPU-accelerated image processing including spatially different bilateral filter to the ROI fluoroscopy, one can preserve or enhance the quality of low-noise ROI image in real-time.