Histogram-driven multi-dimensional adaptive filtering (HD-MAF)

Image noise is one of the main restrictions in computed tomography (CT) with respect to the trade-off between resolution and dose. An HD-MAF is proposed based on a local histogram function identifying the most representative CT value. The local histogram function is computed over an n-dimensional homogenous subvolume which is defined by a principal component analysis. Image filtering affects only predefined CT values which can be seen as image background. The filter method was applied to images of a low-contrast phantom, a low-dose head scan and a thorax scan. For the low-contrast phantom the contrast to noise ratio (CNR) was increased by a factor of 1.8. For the head images a noise reduction of 48% was achieved improving the overall image impression. The corresponding difference images showed no loss of structural information. The 4-dimensional HD- MAF provided images without motion artefacts in the pericardiac region and an overall reduced image noise. HD-MAF is a feasible method for low-contrast enhancement.