Optimised region of interest selection
View Patent ↗A statistical framework for identifying regions of discrimination between two groups of imaged objects involves determining a weight map defining the regions of discrimination by minimizing the sample size required to discriminate the two groups in a numerical optimization scheme.
1. A method for identifying a region of interest (ROI) in an object from a set of similar images of different instances of said object, said method comprising:
inputting the set of similar images into a computer suitably programmed with an algorithm for calculating optimised weights;
calculating an initial weight map of a feature of said images,
applying said algorithm to the initial weight map to calculate optimised weights of said feature at locations in said map to minimise the sample size needed to distinguish a first group of said instances from a second group of said instances, thereby producing a weighted map, and
identifying said ROI as consisting of a set of locations in said weighted map at which said feature has the highest optimised weights.
2. A method as claimed in claim 1 , comprising the preliminary steps of segmenting each image and spatially aligning the segmented images.
3. A method as claimed in claim 1 , wherein the optimised weight of said feature at each location in the map is restricted to positive values.
4. A method as claimed in claim 1 , wherein the optimised weight of said feature at each location in the map is the absolute value of a positive or negative weight.
5. A method as claimed in claim 1 , said calculation to minimise sample size is conducted by minimising for each location in said map a function indicative of sample size incorporating a penalty term which is a measure of the smoothness of the weighted map at said location.
6. A method as claimed in claim 1 , said calculation to minimise sample size is conducted by minimising for each location in said map a function indicative of sample size incorporating a penalisation term which provides sparsity in the weighting of said feature at locations in said map.
7. A method as claimed in claim 6 , wherein said penalisation term is substantially equivalent in its effect to a least absolute shrinkage and selection operator.
8. A method as claimed in claim 4 , wherein said calculation to minimise sample size is carried out by computation of a functional to produce results substantially equivalent to computing the functional:
argmin
w
1
∑
j
∈
{
1
,
2
}
∑
k
∈
G
j
(
F
~
j
k
[
W
]
-
μ
~
j
)
2
n
j
+
λ
∇
W
1
+
η
W
1
(
7
)
in which λ and η are optimised parameters, subject to the constraints, W≧0, ΣW=1.
9. A method as claimed in claim 1 , wherein said images are of biological objects.
10. A method as claimed in claim 9 , wherein said images are of joint cartilages, bone, or brain, or are mammographic images.
11. A method as claimed in claim 9 , wherein said images are of joint cartilages and said weighted map is a map weighting locations in said cartilages at which thickness or volume of the cartilage distinguishes cartilages impaired by arthritis from non-imparied cartilages.
12. A computer programmed for identifying a region of interest (ROI) in an object via instructions to:
receive an input comprising a set of similar images of different instances of said object, calculate an initial weight map of a feature of said images,
apply an algorithm to the initial weight map to calculate optimised weights of said feature at locations in said map to produce a weighted map of said images in which the weights of said feature at the locations in said map are calculated to minimise the sample size needed to distinguish a first group of said instances from a second group of said instances, and
identify said ROI as consisting of a set of locations in said weighted map at which said feature has the highest optimised weights.
13. An instruction set on a non-transitory computer readable storage medium for programming a computer to:
receive an input comprising a set of similar images of different instances of an object,
calculate an initial weight map of a feature of said images,
apply an algorithm to said initial weight map to calculate optimised weights of said feature at locations in said map to produce a weighted map of said images in which the weights of said feature at the locations in said map are calculated to minimise the sample size needed to distinguish a first group of said instances from a second group of said instances, and
identify said ROI as consisting of a set of locations in said weighted map at which said feature has the highest optimised weights.