Object Identifying System for Segmenting Unreconstructed Data in Image Tomography
AssigneeT.I.E.S., LLC
InventorUmar Riaz
Life of this patent
Estimating the term…
Abstract
Tomographic projection data is segmented spatially by forming segmentation masks over the full outlines of objects appearing in isolation within selected data frame images. Further processing locates the object outlines in other data frames. Predictable effects of attenuation are used to attribute appropriate amounts of intensity to the segmented object space. Separate image reconstructions can be made of the segmented object data and the remaining projection data to independently optimize the reconstructions. Composite images can be made of the separate reconstructions.
Claims37
A system for identifying objects among unreconstructed data in image tomography comprising:
- a detector for collecting raw data in the form of photon counts organized as a set of images from a plurality of angular positions around a rotational axis;
- a viewer for viewing the set of images as pixel arrays having a first dimension substantially traverse to the rotational axis and a second dimension substantially parallel to the rotational axis for referencing locations at which the photon counts are collected;
- a selector tool for selecting one or more pixels appearing within an object in at least one of the images;
- a filling tool for identifying pixels that are contiguous to the selected one or more pixels and that satisfy a condition for being within the object;
- an interpolation tool for inferring which among the pixels in other of the images are also within the object;
- and
- a data structure for representing the object based on the pixels appearing within the object through the set of images as a collection of traces circumscribing the rotational axis at different linear positions along one of the first and second dimensions of the pixel arrays.
A method of determining areas occupied by an object within sinograms of unreconstructed tomographic data, comprising the steps of:
- collecting photon counts along two spatial dimensions at each of a plurality of angular increments around a rotational axis, one of the spatial dimensions having at least a component transverse to the rotational axis and another of the spatial dimensions having at least a component along the rotational axis;
- arranging the photon counts in a set of data frames with each of the data frames having a pixel array referencing locations of the photon counts along the two spatial dimensions at one of the angular increments;
- selecting one or more pixels within an outline of an object from at least one of the data frames;
- identifying pixels that are contiguous to the selected one or more pixels in the at least one data frame and that satisfy a condition for being within the outline of the object;
- inferring pixels within other outlines of the same object appearing in other of the data frames;
- rearranging the pixels throughout the set of data frames into a set of sinograms having a first dimension referenced to one of the two spatial dimensions along which the photon counts are collected and a second dimension referenced to the angular increments around the rotational axis at which the photon counts are collected;
- and
- representing the pixels within the outlines of the object as a set of circumferential traces within the set of sinograms, such that each circumferential trace defines a prescribed area within one of the sinograms occupied by the object.
A method of spatially distinguishing a selected object from unreconstructed tomographic data comprising steps of:
- collecting photon counts at each of a plurality of angular increments around a rotational axis;
- spatially referencing the photon counts in data frames as angularly distinguished silhouettes of the target body;
- forming masks over areas bounded by outlines of a selected object within the data frames;
- and
- segmenting the photon counts within the masks for distinguishing the selected object from other features of the target body.
A method of producing a composite image of a target body cross section based on multiple reconstructions of different segments of tomographic data, comprising steps of:
- collecting collimated photon counts emitted by a target body at each of a plurality of angular increments around a rotational axis;
- spatially referencing the photon counts as statistically independent data corresponding to angular progressions about the target body;
- segmenting the statistically independent data corresponding to the angular progressions about the target body into at least first and second data segments containing different portions of the statistically independent data;
- reconstructing a first cross-sectional image of the target body from the first data segment;
- reconstructing a second cross-sectional image of the target body from the second data segment;
- and
- combining the first and second cross-sectional images of the target body to provide a composite image of the target body.
Dependencies are inferred from the claim text.