Adaptive weighted median filtering for reduced blurring when fusing co-registered fast helical CT images

Abstract

Purpose. Fast-helical computed tomography (FHCT) images acquired for a motion model-based CT protocol can be deformably registered to a common geometry and combined through mean or median averaging to create a single low-noise fused image. However, while the fused image has improved noise characteristics relative to the individual FHCT images, registration errors cause the fused image to be relatively blurred. In this study, we present a method to combine multiple FHCT images to reduce imaging noise while retaining image sharpness. Methods. Ten lung cancer patients were imaged 25 times using a low dose, FHCT protocol. For each patient, the first scan was selected as a reference and deformably registered to the following 24 scans. The co-registered images were combined using three methods: mean averaging, median averaging and adaptive weighted median filtering (AWM). Image noise and sharpness were assessed for each technique. Results. Mean pixel-to-pixel image noise in the liver was 7.6 HU for mean averaging, 9.3 HU for median averaging and 9.8 HU for AWM. AWM images were found to be significantly sharper than both mean and median averaged images (p = 0.002) using the Wilcoxon signed rank test. Conclusion. AWM filtering was capable of producing sharper fused images with similar noise reduction to mean and median averaging. The proposed technique also allows for a user-adjustable balance between smoothing and sharpness preservation.