Medical Image Reconstruction Research Articles

PC-based 3-D reconstruction of medical images

Three-dimensional surface reconstruction from serial slices, such as CT or MRI scans, may be performed using a personal computer. Using the C language on a networked IBM PC-clone personal computer, a complete set of software to perform 3-D surface reconstruction from serial sections has been developed. Surface detection is based on level slicing or thresholding. Both depth coding and gradient surface shading models are included. This system is efficient in terms of computational and memory requirements.

Medicube: A 3D medical imaging architecture

A 3D medical imaging system, called the MediCube system, is described. This system supports the reconstruction, manipulation, analysis, and display of 3D volumetric medical images. The MediCube system is based upon the general-purpose voxel-based Cube architecture, which employs parallel memory and parallel processing to support real-time manipulation and display of voxel imagery. Primary applications for the MediCube system include medical diagnosis, surgical planning, radiation therapy, reconstructive surgery, medical education and research. Examples from an MRI study of the brain are presented.

Error bounds for the method of alternating projections

Given a collection of closed subspaces of a Hilbert space, the method of alternating projections produces a sequence which converges to the orthogonal projection onto the intersection of the subspaces. A large class of problems in medical and geophysical image reconstruction can be solved using this method. A sharp error bound will enable the userto estimate accurately the number of iterations necessary to achieve a desired relative error. We obtain the sharpest possible upper bound for the case of two subspaces, and the sharpest known upper bound for more than two subspaces.

Dynamic elastic interpolation for 3D medical image reconstruction from serial cross sections

An interpolation method is proposed for generating the intermediate contours between a start contour and a goal contour. Coupled with the display method for voxel-based objects, it provides a powerful tool for reconstructing the 3D object from serial cross sections. The method tries to fill in the lost information between two slices, assuming that there is smooth change between them. This is a reasonable assumption provided that the sampling is at least twice the Nyquist rate, in which case the result of the interpolation is expected to be very close to reality. One of the major advantages of this approach is its ability to handle the branching problem. Another major advantage is that after each intermediate contour is generated and sent to display device, there is no need to keep it in the memory unless the solid model will be used for further processing. Thus, the space complexity of this algorithm is relatively low.

Reconstruction of a three-dimensional object from a limited range of views

We define the Radon transform R and the back projection R ∗ (adjoint of R ) on the space of C ∞ rapidly decreasing functions J and on the space of distributions of compact support E ′. We show that R ∗ R is a convolution operator. The result is true in the case of full views as well as in the case of a limited range of views. An application to the medical image reconstruction is given.

Laser light diffraction, spatial filtering, and reconstruction of medical radiographic images: a pilot study.

Using a laser light source, an optical bench, transform lenses, and spatial filters in a system called LaserScan, <R xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</R> we are studying laser light diffraction, spatial filtering, and reconstruction of medical radiographic images, with the help of GTS Corporation, whose laboratories house the laser computer. Medical radiographs are minified to a 35 mm second-generation transparency, then transilluminated by monochromatic, coherent, collimated light from a helium-neon gas laser. A first transform lens produces a Fourier Transform of the minified image in the form of a diffraction pattern that is photographed on Polaroid film, and/or viewed with a CCTV system. Spatial filtering is accomplished in the plane of the diffraction pattern by using various opaque elements of appropriate geometric form to filter undesired spatial frequencies. Spatial filters take the form of ``wires,'' ``sector-wedges,'' and circular apertures, as well as other empirically determined shapes. A second lens performs an inverse Fourier transform on the diffraction pattern producing a reconstructed, filtered, real image of the minified film. This reconstructed image is also photographed on Polaroid film or viewed on the CCTV screen. One of our goals is to remove certain unnecessary pattern detail from particular radiographic images so that characteristic patterns of diagnostic picture detail might be more readily identified and diagnosed. Experimental results, including photographs of diffraction patterns and some filtered reconstructed images, are presented.