Image analysis – Applications – Biomedical applications
Patent
1994-07-18
1997-07-08
Couso, Jose L.
Image analysis
Applications
Biomedical applications
382268, 356301, 356349, 428219, 2504921, 250398, G06K 900
Patent
active
056470185
DESCRIPTION:
BRIEF SUMMARY
FIELD OF THE INVENTION
This invention relates to three-dimensional image generation using a variety of types of interrogating radiation, including electromagnetic, particle and acoustic radiation. In particular, it finds application in the processing of images produced with X-rays.
BACKGROUND OF THE INVENTION
In conventional X-ray tomography the apparatus produces images of a single slice of an object by moving an X-ray camera and the image plate in a complementary manner so that everything is blurred except for a predetermined plane. Commercial machines produce images of a number of separate slices, moving the X-ray machine to produce a planar picture and then moving the object under test in the Z direction to produce a planar picture at different slices. That needs more total pictures than the views spread over a solid angle and it produces a cruder representation of a three-dimensional image because volume cells intermediate between slices have to be derived by interpolation.
Conventional tomography thus produces images of discrete slices normal to the axis of objects of short axial length. It is more common, however, to require images to discrete slices normal to the axis of a long object. The established form of computer tomography (CT) is designed to do this. It does it by taking a number line images, at different viewing angles in the plane of each such slice (typically 360-1440 views, taken at 1.degree. to 1/4.degree. intervals). These are then used to compute the corresponding slice image. However, the invention--and even the established form of computer tomography--is better understood by first postulating an extension of computer tomography which, unlike existing systems, would produce a true three-dimensional mapping of the X-ray densities (i.e. opacities) of a complete set of symmetrical, equally-spaced volume cells (voxels), in place of merely producing such information for each of a limited number of parallel slices through the relevant object. This hypothetical version of computer tomography might map a cubical volume L.sup.3 into (L/c).sup.3 voxels of size c.sup.3. In its simplest form, this postulated ideal form of computer tomography would involve L/c distinct X-ray projections, of (L/c).sup.2 pixels. Each pixel represents the aggregate density of a column of L/c voxels. The individual densities of all voxels are then computed by (L/c).sup.3 simultaneous equations. No assumptions about the features of the specimen are involved. The resulting picture is fully three-dimensional, amenable to three-axis rotation, slicing etc.
In practice, the voxels are not independent: signal
oise considerations require a significant degree of smoothing and integration (by the human eye or brain and/or by a computer algorithm), which causes the effective image to be composed of reasonably smooth multi-voxel elements.
We have found that it is possible to make use of this assumption to simplify data extraction, storage and processing.
The processing may be simplified by treating an object under test as a set of component objects, each defined by one or more closed surfaces, where the precision of the surface shape of these component objects should ideally match the resolution of the sensor system. Indeed, for smoothly-varying surfaces, a least mean-squares or similar fit can produce a precision superior to that of the individual constituent measurements. Furthermore these component objects are generally of individually uniform specific density, and of reasonably "well-behaved" shape. (Complex shapes may be broken down into multiple constituent objects of simpler shape.)
These component objects can be cross-identified, between different views, by projection on to the axis common to each pair of such views. Each such projection also defines the three-dimensional position of two outermost surface points of the object (as projected on to this common axis), with a precision matching that of the source pictures. Thus n views generate n(n-1) such points on each object surface, and these points cluster most clos
REFERENCES:
patent: 4407008 (1983-09-01), Schmidt et al.
patent: 4498770 (1985-02-01), Corwin et al.
patent: 4621911 (1986-11-01), Lanni et al.
patent: 5421330 (1995-06-01), Thirion et al.
Giblin et al: "Reconstruction of surgaces from profiles", First International Conference on Computer Vision Jun. 8, 1987, pp. 136-144.
Van Hove et al: "A silhouette-slice theorem for opaque 3D objects", ICASSP85.International Conference on Acoustic, Speech, and Signal Processing, vol. 2, Mar. 26, 1985, pp. 933-936.
Couso Jose L.
Do Anh Hong
Imperial College of Science Technology and Medicine
LandOfFree
Method and apparatus for generating images does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for generating images, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for generating images will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2413408