Automatic estimation and removal of emission image noise

The authors have developed a method of automatically estimating noise in 2and 3D emission images from the information intrinsic to them and then using this estimate to selectively remove the noise from the image. Unlike low cutoff filters, which reduce resolution and contrast, the morphing operation removes noise without adversely affecting these qualities. The noise estimation method uses contiguous volume analysis to obtain information about all of the hot spots in the image. If the image contains hot regions too small to contain information, the method analyzes their sizes and heights. This information is used to find a conservative estimate of noise for the image. The information about noise is used to perform an image morphing using a noise-specific structure. Erosion and dilation are morphing operations which remove image detail smaller than the specified structure, without distortion of the larger features. When applied to PET and SPECT images using the correct structure, these operations remove only noise, thereby improving image uniformity without degrading resolution or contrast.