Tumor detection and segmentation for DPI AI platform routes

1. A method of tumor detection and segmentation for Digital Pathology; the method comprising:

accepting as an input a first Whole Slide Image (WSI) characterized by a first resolution;

creating a second WSI by sampling the first WSI such that the second WSI is characterized by a second resolution lower than the first resolution;

applying an adaptive thresholding technique to the second WSI to create a mask distinguishing background from one or more patches, the mask being characterized by the second resolution;

applying the mask to the first WSI to extract the one or more patches, providing a third WSI with the extracted patches, characterized by a third resolution, greater than the second resolution and lower than the first resolution;

using a first machine learning system to operate on the third WSI to create a heat map at the third resolution, indicating a subset of the patches, each patch in the subset considered likely to include one or more clusters of tumor cells; and

using a second machine learning system to operate on the first WSI in conjunction with the heat map such that each patch of the subset is segmented in a corresponding output image at the first resolution, outlining one or more corresponding clusters;

wherein the adaptive thresholding technique comprises setting a threshold determined by a sharpness value characterizing a peak corresponding to the background in a distribution of pixel values over the second WSI.

2. The method of claim 1 ,

wherein the second WSI comprises image data in at least two channels; and

wherein the adaptive thresholding technique comprises removing pixels with values above a threshold value in a chosen one of the at least two channels.

3. The method of claim 1 ,

wherein the second WSI comprises image data in red, green and blue channels; and

wherein the adaptive thresholding technique comprises removing pixels with values above a threshold value in the green channel.

4. The method of claim 2 ,

wherein the threshold value is set for the second WSI by analyzing intensity statistics of image data in the chosen channel.

5. The method of claim 4 ,

wherein the analyzing comprises:

applying a Gaussian filter to a first histogram of image intensity data in the chosen channel to create a smoothed histogram;

calculating a gradient histogram corresponding to the smoothed histogram, from which a mean value of positive gradient is determined; and

setting the threshold value in relation to the mean value of positive gradient.

6. The method of claim 5 ,

wherein setting the threshold value in relation to the mean value of positive gradient comprises:

setting the threshold value equal to an intensity value corresponding to the mean value of positive gradient, such that all pixels of intensity value lower than or equal to the threshold value are assigned to be clear in the mask, and all pixels of intensity value higher than the threshold value are assigned to be opaque in the mask.

7. The method of claim 1 ,

wherein the first machine learning system is the second machine learning system.

8. An apparatus comprising:

one or more processors; and

logic encoded in one or more non-transitory media for execution by the one or more processors and when executed operable to:

accept as an input a first Whole Slide Image (WSI) characterized by a first resolution;

create a second WSI by sampling the first WSI such that the second WSI is characterized by a second resolution lower than the first resolution;

apply an adaptive thresholding technique to the second WSI to create a mask distinguishing background from one or more patches, where the mask is characterized by the second resolution;

apply the mask to the first WSI to extract the one or more patches, providing a third WSI with the extracted patches, characterized by a third resolution, greater than the second resolution and lower than the first resolution;

use a first machine learning system to operate on the third WSI to create a heat map at the third resolution, indicating a subset of the patches, each patch in the subset considered likely to include one or more clusters of tumor cells; and

use a second machine learning system to operate on the first WSI in conjunction with the heat map such that each patch of the subset is segmented in a corresponding output image at the first resolution, outlining one or more corresponding clusters;

wherein the adaptive thresholding technique comprises setting a threshold determined by a sharpness value characterizing a peak corresponding to the background in a distribution of pixel values over the second WSI.

9. The apparatus of claim 8 ,

wherein the second WSI comprises image data in at least two channels; and

wherein the adaptive thresholding technique comprises removing pixels with values above a threshold value in a chosen one of the at least two channels.

10. The apparatus of claim 8 ,

wherein the second WSI comprises image data in red, green and blue channels; and

wherein the adaptive thresholding technique comprises removing pixels with values above a threshold value in the green channel.

11. The apparatus of claim 9 ,

wherein the threshold value is set for the second WSI by analyzing intensity statistics of image data in the chosen channel.

12. The apparatus of claim 11 ,

wherein the analyzing comprises:

applying a Gaussian filter to a histogram of image intensity data in the chosen channel to create a smoothed histogram;

calculating a gradient histogram corresponding to the smoothed histogram, from which a mean value of positive gradient is determined; and

setting the threshold value in relation to the mean value of positive gradient.

13. The apparatus of claim 12 ,

wherein setting the threshold value in relation to the mean value of positive gradient comprises:

setting the threshold value equal to an intensity value corresponding to the mean value of positive gradient, such that all pixels of intensity value lower than or equal to the threshold value are assigned to be clear in the mask, and all pixels of intensity value higher than the threshold value are assigned to be opaque in the mask.

14. The apparatus of claim 8 ,

wherein the first machine learning system is the second machine learning system.

15. An apparatus comprising:

one or more processors; and

software encoded in one or more non-transitory computer-readable media for execution by the one or more processors and when executed operable to:

accept as an input a first Whole Slide Image (WSI) characterized by a first resolution;

create a second WSI by sampling the first WSI such that the second WSI is characterized by a second resolution lower than the first resolution;

apply an adaptive thresholding technique to the second WSI to create a mask distinguishing background from one or more patches, where the mask is characterized by the second resolution;

apply the mask to the first WSI to extract the one or more patches, providing a third WSI with the extracted patches, characterized by a third resolution, greater than the second resolution and lower than the first resolution;

use a first machine learning system to operate on the third WSI to create a heat map at the third resolution, indicating a subset of the patches, each patch in the subset considered likely to include one or more clusters of tumor cells; and

use a second machine learning system to operate on the first WSI in conjunction with the heat map such that each patch of the subset is segmented in a corresponding output image at the first resolution, outlining one or more corresponding clusters;

wherein the adaptive thresholding technique comprises setting a threshold determined by a sharpness value characterizing a peak corresponding to the background in a distribution of pixel values over the second WSI.

16. The apparatus of claim 15 ,

wherein the second WSI comprises image data in at least two channels; and

wherein the adaptive thresholding technique comprises removing pixels with values above a threshold value in a chosen one of the at least two channels.

17. The apparatus of claim 15 ,

wherein the second WSI comprises image data in red, green and blue channels; and

wherein the adaptive thresholding technique comprises removing pixels with values above a threshold value in the green channel.

18. The apparatus of claim 16 ,

wherein the threshold value is set for the second WSI by analyzing intensity statistics of image data in the chosen channel.

19. The apparatus of claim 18 ,

wherein the analyzing comprises:

applying a Gaussian filter to a first histogram of image intensity data in the chosen channel to create a smoothed histogram;

calculating a gradient histogram corresponding to the smoothed histogram, from which a mean value of positive gradient is determined; and

setting the threshold value in relation to the mean value of positive gradient.

20. The apparatus of claim 19 ,

wherein setting the threshold value in relation to the mean value of absolute gradient comprises:

selecting a group of values consisting of all values present in the first histogram that are equal to or lower than the mean value of absolute gradient; and

setting the threshold value as the highest value in the group.