World Congress on Medical and Biomedical Engineering, San Antonio, Texas, 6-13 August. Physics in Medicine and Biology, 33(Supplement I):151, 1988.
Eigenimage filtering is a linear filter which produces a composite image from a sequence of images in which a desired feature is enhanced while an interfering feature is suppressed. A potential source of error in the interpretation of filtered images are projection artifacts. Projection artifacts are the result that any pixel vector may be represented as the linear combination of the desired and interfering signature vectors, and a specific vector which is orthogonal to the subspace R(d,u) containing the desired and interfering signature vectors. That component of the pixel vector parallel to the desired feature will be projected into the filtered image, and represents real information or is an artifact. A method which minimizes artifacts has been developed. This method determines a vector which is orthogonal to the subspace R(d,u). This vector is used to form a filtered image that contains only those pixels whose pixel vectors have a component orthogonal to the subspace R(d,u). This image is used to form a mask image. Any pixels contained in the mask image are not retained in the filtered image since they may represent an artifact. This technique is demonstrated for simulated and Magnetic Resonance image sequences. Abstract