Automated selection of image regions

A method for automated selection of image regions (“tiles”) in an overall image, the method comprising:

• a) generating a computed image from at least one of a co-occurrence matrix of the overall image and an entropy image of the overall image, the computed image distinguishing between tiles on the basis of whether they are relatively more or relatively less likely to exhibit a property of interest;
• b) selecting a prior subset of the tiles which are relatively more likely to exhibit the property of interest;
• c) for tiles in the prior subset, deriving a feature measure associated with at least one of colour and texture;
• and
• d) selecting from the prior subset a further subset of tiles at least partly on the basis of tile feature measures wherein the step of generating a computed image uses a combination of the co-occurrence matrix, the entropy image and a vignette mask arranged to exclude from the computed image outer regions of the overall image having low illumination relative to inner regions.

A method for cancer grading comprising the automated selection of image regions (“tiles”) in an overall image, the method comprising:

• a) generating a computed image from at least one of a co-occurrence matrix of the overall image and an entropy image of the overall image, the computed image distinguishing between tiles on the basis of whether they are relatively more or relatively less likely to exhibit a property of interest;
• b) selecting a prior subset of the tiles which are relatively more likely to exhibit the property of interest;
• c) for tiles in the prior subset, deriving a feature measure associated with at least one of colour and texture;
• and
• d) selecting from the prior subset a further subset of tiles at least partly on the basis of tile feature measures;
• the method further including the step of applying a Hue/Saturation/Value transformation to image data, thresholding Hue and Saturation components to obtain areas of ‘light pink’, ‘dark pink’ and ‘white’ tissue and computing a feature measure for each tile based on colour and texture.

A method for steroid/protein expression analysis comprising the automated selection of image regions (“tiles”) in an overall image, the method comprising:

• a) generating a computed image from at least one of a co-occurrence matrix of the overall image and an entropy image of the overall image, the computed image distinguishing between tiles on the basis of whether they are relatively more or relatively less likely to exhibit a property of interest;
• b) selecting a prior subset of the tiles which are relatively more likely to exhibit the property of interest;
• c) for tiles in the prior subset, deriving a feature measure associated with at least one of colour and texture;
• and
• d) selecting from the prior subset a further subset of tiles at least partly on the basis of tile feature measures, and further including the step of transforming image data to provide cyan image data and computing a feature measure for each tile based on a measure of colour and texture of the cyan image data and such data's Sobel filtered equivalent.

Apparatus for automated selection of tiles in an overall image, the apparatus comprising:

• a) computer apparatus for generating a computed image from at least one of a co-occurrence matrix of the overall image and an entropy image of the overall image, the computed image distinguishing between tiles on the basis of whether they are relatively more or relatively less likely to exhibit a property of interest
• b) computer apparatus for selecting a prior subset of the tiles which are relatively more likely to exhibit the property of interest
• c) computer apparatus for deriving a feature measure associated with at least one of colour and texture for tiles in the prior subset and
• d) computer apparatus for selecting from the prior subset a further subset of tiles at least partly on the basis of tile feature measures wherein the means for selecting from the prior subset a further subset of tiles is arranged to avoid excessive tile overlap.

Apparatus for automated selection of tiles in an overall image, the apparatus comprising:

• a) computer apparatus for generating a computed image from at least one of a co-occurrence matrix of the overall image and an entropy image of the overall image, the computed image distinguishing between tiles on the basis of whether they are relatively more or relatively less likely to exhibit a property of interest
• b) computer apparatus for selecting a prior subset of the tiles which are relatively more likely to exhibit the property of interest
• c) computer apparatus for deriving a feature measure associated with at least one of colour and texture for tiles in the prior subset and
• d) computer apparatus for selecting from the prior subset a further subset of tiles at least partly on the basis of tile feature measures wherein the means for generating a computed image uses a combination of the co-occurrence matrix, the entropy image and a vignette mask arranged to exclude from the computed image outer regions of the overall image having low illumination relative to inner regions.

A computer software product comprising a non-transitory computer readable medium containing computer readable instructions for controlling operation of computer apparatus to implement automated selection of tiles in an overall image, wherein the computer readable instructions provide a means for controlling the computer apparatus to carry out the steps of:

• a) generating a computed image from at least one of a co-occurrence matrix of the overall image and an entropy image of the overall image, the computed image distinguishing between tiles on the basis of whether they are relatively more or relatively less likely to exhibit a property of interest;
• b) selecting a prior subset of the tiles which are relatively more likely to exhibit the property of interest;
• c) for tiles in the prior subset, deriving a feature measure associated with at least one of colour and texture;
• and
• d) selecting from the prior subset a further subset of tiles at least partly on the basis of tile feature measures and further having instructions for implementing the step of selecting from the prior subset a further subset of tiles having regard to avoidance of excessive tile overlap.

A computer software product for cancer grading comprising a non-transitory computer readable medium containing computer readable instructions for controlling operation of computer apparatus to implement automated selection of tiles in an overall image, wherein the computer readable instructions provide a means for controlling the computer apparatus to carry out the steps of:

• a) generating a computed image from at least one of a co-occurrence matrix of the overall image and an entropy image of the overall image, the computed image distinguishing between tiles on the basis of whether they are relatively more or relatively less likely to exhibit a property of interest;
• b) selecting a prior subset of the tiles which are relatively more likely to exhibit the property of interest;
• c) for tiles in the prior subset, deriving a feature measure associated with at least one of colour and texture;
• and
• d) selecting from the prior subset a further subset of tiles at least partly on the basis of tile feature measures and further having instructions for implementing the step of applying a Hue/Saturation/Value transformation to image data, thresholding Hue and Saturation components to obtain areas of ‘light pink’, ‘dark pink’ and ‘white’ tissue and computing a feature measure for each tile based on colour and texture.

A computer software product for steroid/protein expression analysis comprising a non-transitory computer readable medium containing computer readable instructions for controlling operation of computer apparatus to implement automated selection of tiles in an overall image, wherein the computer readable instructions provide a means for controlling the computer apparatus to carry out the steps of:

• a) generating a computed image from at least one of a co-occurrence matrix of the overall image and an entropy image of the overall image, the computed image distinguishing between tiles on the basis of whether they are relatively more or relatively less likely to exhibit a property of interest;
• b) selecting a prior subset of the tiles which are relatively more likely to exhibit the property of interest;
• c) for tiles in the prior subset, deriving a feature measure associated with at least one of colour and texture;
• and
• d) selecting from the prior subset a further subset of tiles at least partly on the basis of tile feature measures and further having instructions for implementing the step of transforming image data to provide cyan image data and computing a feature measure for each tile based on a measure of colour and texture of the cyan image data and such data's Sobel filtered equivalent.