System and Method for Image Sequence Processing

Abstract

A method for processing a sequence of images. In an embodiment, one or more training datasets are analyzed having a sequence of images showing a first condition and a sequence of images showing a second condition. A multivariate regression model is used to determine a relationship between relative positions of one or more features in the sequence of images showing the first condition and relative positions of the one or more features in the sequence of images showing the second condition. In an embodiment, the determined relationship is used to predict positions of the one or more features in an inquire sequence of images showing the second condition given an inquire sequence of images showing the first condition. The predicted positions can then be refined using various methods. In an embodiment, sequences of images are aligned to a common time scale.

1 . A method of refining a feature associated with an image, comprising:
 receiving at least one time-aligned pair of images of a heart, wherein each at least one time-aligned pair of images comprises a rest image and a stressed image, wherein the rest image comprises an image of the heart at a particular point in time in the cardiac cycle of the heart when the heart is at rest and the stressed image comprises an image of the heart at the particular point in time in the cardiac cycle of the heart when the heart is under a stress stimulus, wherein each stressed image of the at least one time-aligned pair of images has a feature associated therewith, wherein the feature comprises at least one point wherein each at least one point has a position in the stressed image; and   refining the position of one or more points in the feature associated with the stressed image of one or more of the at least one time-aligned pair of images based on information derived from the rest image of one or more of the at least one time-aligned pair of images.   
     
     
         2 . The method according to  claim 1 , wherein the method further comprises identifying the at least one time-aligned pair of images from a plurality of images of the heart at rest and a plurality of images of the heart under the stress stimulus, wherein for each of the at least one time-aligned pair of images:
 the rest image is identified in the plurality of images of the heart at rest; and   the stressed image is identified in the plurality of images of the heart under the stress stimulus.   
     
     
         3 . The method according to  claim 1 , wherein the refining the position of each of the one or more points in the feature associated with the stressed image of each of the one or more of the at least one time-aligned pair of images, comprises:
 identifying a corresponding position in the rest image of the time-aligned pair of images; and   updating the position based on texture information near the corresponding position in the rest image.   
     
     
         4 . The method according to  claim 3 , wherein updating the position based on texture information near the corresponding position in the rest image, comprises:
 defining a template set of pixels in the rest image, wherein the template set comprises a pixel corresponding to the corresponding position and one or more additional template pixels;   defining a search window set of pixels in the stressed image, wherein the search window set comprises a pixel corresponding to the position and one or more additional search pixels, and wherein the search window set comprises a plurality of possible displacements of the template set;   computing a similarity measure for one or more of the plurality of possible displacements, wherein the similarity measure represents a degree of textural similarity between the template set and the displacement;   updating the position based on at least one of the plurality of possible displacements.   
     
     
         5 . The method according to  claim 4 , wherein the similarity measure is computed by one or more suitably programmed computers. 
     
     
         6 . The method according to  claim 4 , wherein the position is updated based on the displacement of the plurality of possible displacements with the greatest computed similarity measure. 
     
     
         7 . The method according to  claim 4 , wherein the position is updated based on two or more displacements of the plurality of possible displacements and the effect of each displacement of the two or more displacements on the position is weighted based on the computed similarity measure of the displacement. 
     
     
         8 . The method according to  claim 4 , wherein the similarity measure is based on Rayleigh distributed speckle in the template set and the displacement. 
     
     
         9 . The method according to  claim 6 , wherein:
 the template set is a block of pixels centered on the pixel corresponding to the corresponding position in the rest image;   the search window set is a block of pixels centered on the pixel corresponding to the position in the stressed image; and   the position is updated to correspond to a pixel at a center of the displacement with the greatest computed similarity measure.   
     
     
         10 . The method according to  claim 4 , wherein:
 the template set is a 3×3 block of pixels contiguous to the pixel corresponding to the corresponding position in the rest image;   the search window set is a 5×5 block of pixels centered on the pixel corresponding to the position in the stressed image; and   the plurality of possible displacements consists of nine overlapping displacements of the 3×3 block of pixels within the 5×5 block of pixels.   
     
     
         11 . The method according to  claim 4 , wherein the feature approximates an epicardial border and/or an endocardial border of the heart. 
     
     
         12 . A method of temporally aligning a plurality of images of a heart, comprising:
 identifying in a plurality of images of a heart at least three phase images corresponding to a corresponding at least three phases of a cardiac cycle of the heart, wherein the plurality of images comprises images of the heart at a corresponding plurality of timepoints within the cardiac cycle of the heart, wherein the plurality of images are sequenced in the order of their corresponding timepoints within the cardiac cycle of the heart; and   assigning a time value to each image of the plurality of images, wherein the at least three phase images are assigned time values based on an established timing of their corresponding phases in the cardiac cycle of the heart and the remaining images of the plurality of images are assigned time values based on at least two piecewise interpolation functions, wherein for each consecutive pair of phase images of the at least three phase images a suitable programmed computer applies one of the at least two piecewise interpolation functions to the images sequenced between the consecutive pair of phase images to determine the assigned time values of the images based on the time values assigned to the consecutive pair of phase images.   
     
     
         13 . The method according to  claim 12 , wherein one or more of the at least two piecewise interpolation functions is a linear function. 
     
     
         14 . One or more non-transitory computer-readable media having computer-useable instructions embodied thereon for performing a method of refining a feature associated with an image, the method comprising:
 receiving at least one time-aligned pair of images of a heart, wherein each at least one time-aligned pair of images comprises a rest image and a stressed image, wherein the rest image comprises an image of the heart at a particular point in time in the cardiac cycle of the heart when the heart is at rest and the stressed image comprises an image of the heart at the particular point in time in the cardiac cycle of the heart when the heart is under a stress stimulus, wherein each stressed image of the at least one time-aligned pair of images has a feature associated therewith, wherein the feature comprises at least one point wherein each at least one point has a position in the stressed image; and   refining the position of one or more points in the feature associated with the stressed image of one or more of the at least one time-aligned pair of images based on information derived from the rest image of one or more of the at least one time-aligned pair of images.   
     
     
         15 . The media according to  claim 14 , wherein the method further comprises identifying the at least one time-aligned pair of images from a plurality of images of the heart at rest and a plurality of images of the heart under the stress stimulus, wherein for each of the at least one time-aligned pair of images:
 the rest image is identified in the plurality of images of the heart at rest; and   the stressed image is identified in the plurality of images of the heart under the stress stimulus.   
     
     
         16 . The media according to  claim 14 , wherein the refining the position of each of the one or more points in the feature associated with the stressed image of each of the one or more of the at least one time-aligned pair of images, comprises:
 identifying a corresponding position in the rest image of the time-aligned pair of images; and   updating the position based on texture information near the corresponding position in the rest image.   
     
     
         17 . The media according to  claim 16 , wherein updating the position based on texture information near the corresponding position in the rest image, comprises:
 defining a template set of pixels in the rest image, wherein the template set comprises a pixel corresponding to the corresponding position and one or more additional template pixels;   defining a search window set of pixels in the stressed image, wherein the search window set comprises a pixel corresponding to the position and one or more additional search pixels, and wherein the search window set comprises a plurality of possible displacements of the template set;   computing a similarity measure for one or more of the plurality of possible displacements, wherein the similarity measure represents a degree of textural similarity between the template set and the displacement;   updating the position based on at least one of the plurality of possible displacements.   
     
     
         18 . The media according to  claim 17 , wherein the similarity measure is computed by one or more suitably programmed computers. 
     
     
         19 . The media according to  claim 17 , wherein the position is updated based on the displacement of the plurality of possible displacements with the greatest computed similarity measure. 
     
     
         20 . The media according to  claim 17 , wherein the position is updated based on two or more displacements of the plurality of possible displacements and the effect of each displacement of the two or more displacements on the position is weighted based on the computed similarity measure of the displacement. 
     
     
         21 . The media according to  claim 17 , wherein the similarity measure is based on Rayleigh distributed speckle in the template set and the displacement. 
     
     
         22 . The media according to  claim 19 , wherein:
 the template set is a block of pixels centered on the pixel corresponding to the corresponding position in the rest image;   the search window set is a block of pixels centered on the pixel corresponding to the position in the stressed image; and   the position is updated to correspond to a pixel at a center of the displacement with the greatest computed similarity measure.   
     
     
         23 . The media according to  claim 17 , wherein:
 the template set is a 3×3 block of pixels contiguous to the pixel corresponding to the corresponding position in the rest image;   the search window set is a 5×5 block of pixels centered on the pixel corresponding to the position in the stressed image; and   the plurality of possible displacements consists of nine overlapping displacements of the 3×3 block of pixels within the 5×5 block of pixels.   
     
     
         24 . The media according to  claim 17 , wherein the feature approximates an epicardial border and/or an endocardial border of the heart. 
     
     
         25 . One or more non-transitory computer-readable media having computer-useable instructions embodied thereon for performing a method of temporally aligning a plurality of images of a heart, the method comprising:
 identifying in a plurality of images of a heart at least three phase images corresponding to a corresponding at least three phases of a cardiac cycle of the heart, wherein the plurality of images comprises images of the heart at a corresponding plurality of timepoints within the cardiac cycle of the heart, wherein the plurality of images are sequenced in the order of their corresponding timepoints within the cardiac cycle of the heart; and   assigning a time value to each image of the plurality of images, wherein the at least three phase images are assigned time values based on an established timing of their corresponding phases in the cardiac cycle of the heart and the remaining images of the plurality of images are assigned time values based on at least two piecewise interpolation functions, wherein for each consecutive pair of phase images of the at least three phase images a suitable programmed computer applies one of the at least two piecewise interpolation functions to the images sequenced between the consecutive pair of phase images to determine the assigned time values of the images based on the time values assigned to the consecutive pair of phase images.   
     
     
         26 . The media according to  claim 25 , wherein one or more of the at least two piecewise interpolation functions is a linear function.