System, method and computer-accessible medium for quantification of blur in digital images

Patent metadata

Field	Value
Publication number	US-10878293-B2
Application number	US-201716328996-A
Country	US
Kind code	B2
Filing date	Aug 30, 2017
Priority date	Aug 30, 2016
Publication date	Dec 29, 2020
Grant date	Dec 29, 2020

Abstract

The present disclosure discusses systems and methods to detect blur in digital images. The solution can be incorporated into the quality control systems of pathology and other slide scanners or can be a stand-alone solution. The solution can identify scanned images that include blur and cause the scanner to automatically rescan the blurry image. The solution can also identify regions of the scanned image that include blur. The solution can generate blur maps for each of the scanned images that identify regions of the scanned image that include blur.

First claim

What is claimed is:

1. A system to prepare histological slides, comprising one or more processors to: a patch generator executed by the one or more processors to generate a plurality of patches from a digital image, each patch of the plurality of patches comprising a plurality of pixels; a feature extractor to calculate, for each patch of the plurality of patches, values according to one or more sharpness metrics; and a patch classifier to: determine a blur score for each patch of the plurality of patches, the blur score of the patch determined using the calculated values of the one or more sharpness metrics of the patch; and generate a blur map based on the blur score for each patch of the plurality of patches, the blur map including, for each patch of the plurality of patches, a parameter value between a first threshold value and a second threshold value that is based on the blur score of the patch;
further comprising: the patch classifier to determine the blur score for each of the plurality of patches with a neural network.

2. The system of claim 1 , further comprising:
the patch classifier to determine the blur score for each of the plurality of patches with one of a random forest regression algorithm or a logistic regression algorithm.

3. The system of claim 1 , further comprising:
the patch classifier to determine the blur score for each of the plurality of patches with a residual neural network.

4. The system of claim 1 , further comprising:
a background detector to discard a patch that comprises background data.

5. The system of claim 1 , wherein a portion of a first of the plurality of patches overlaps a portion of a second of the plurality of patches.

6. The system of claim 1 , further comprising:
the feature extractor to calculate a plurality of values for each patch in the group of the plurality of patches.

7. The system of claim 1 , wherein the one or more sharpness metrics comprise at least one pixel intensity-based feature, gradient-based feature, transform-based feature, and perceptual-based feature.

8. The system of claim 1 , wherein the one or more sharpness metrics comprise a variance metric, a range histogram metric, an entropy histogram metric, a Mason and Green's histogram metric, a Mendelsohn and Mayall's histogram metric, a gradient metric, a sum of modified laplacian metric, a Tenengrad metric, a blur metric in the frequency domain, a DCT blur metric, a Haar wavelet transform metric, a Marziliano metric, and a cumulative probability of blur detection metric.

9. The system of claim 1 , further comprising:
a background detector to convert the image to a grayscale image.

10. The system of claim 1 , further comprising the patch classifier to:
flag each patch in the group of the plurality of patches having the blur score above a predetermined threshold; and
generate the blur map based on the flagged patches.

11. A method for detecting a quantity of blur in images, comprising: generating, by a blur detector including one or more processors, a plurality of patches from a digital image, each patch of the plurality of patches comprising a plurality of pixels; calculating, by the blur detector, for each patch of the plurality of patches, values according to one or more sharpness metrics; determining, by the blur detector, a blur score for each patch of the plurality of patches, the blur score of the patch determined using the calculated values of the one or more sharpness metrics of the patch; and generating, by the blur detector, a blur map based on the blur score for each patch of the plurality of patches, the blur map including, for each patch of the plurality of patches, a parameter value between a first threshold value and a second threshold value that is based on the blur score of the patch;
further comprising: the patch classifier to determine the blur score for each of the plurality of patches with a neural network.

12. The method of claim 11 , further comprising:
determining, by the blur detector, the blur score for each of the plurality of patches with one of a random forest regression algorithm or a logistic regression algorithm.

13. The method of claim 11 , further comprising:
determining, by the blur detector, the blur score for each of the plurality of patches with a residual neural network.

14. The method of claim 11 , further comprising discarding, by the blur detector, a patch that comprises background data.

15. The method of claim 11 , wherein a portion of a first of the plurality of patches overlaps a portion of a second of the plurality of patches.

16. The method of claim 11 , further comprising:
calculating, by the blur detector, a plurality of values for each patch in the group of the plurality of patches.

17. The method of claim 11 , wherein the one or more sharpness metrics comprise at least one pixel intensity-based feature, gradient-based feature, transform-based feature, or perceptual-based feature.

18. The method of claim 11 , wherein the one or more sharpness metrics comprise a variance metric, a range histogram metric, an entropy histogram metric, a Mason and Green's histogram metric, a Mendelsohn and Mayall's histogram metric, a gradient metric, a sum of modified laplacian metric, a Tenengrad metric, a blur metric in the frequency domain, a DCT blur metric, a Haar wavelet transform metric, a Marziliano metric, or a cumulative probability of blur detection metric.

19. The method of claim 11 , further comprising converting the image to a grayscale image.

20. The method of claim 11 , further comprising:
flagging, by the blur detector, each patch in the group of the plurality of patches having the blur score above a predetermined threshold; and
generating, by the blur detector, the blur map based on the flagged patches.

Assignees

• Memorial Sloan Kettering Cancer Center

Inventors

• Campanella Gabriele
• Schüffler Peter J
• Fuchs Thomas

Classifications

• G06F18/2415

  based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate · CPC title

• G06V10/30 Primary

  Noise filtering · CPC title

• G06V20/695

  Preprocessing, e.g. image segmentation · CPC title

• G06K9/00

  Physics · mapped topic

• G06K9/0014

  Physics · mapped topic