System and method for single channel whole cell segmentation

1. A computer-implemented method for transforming a sample image of a biological sample stained with only one or more non-nuclear cell marker stains which do not specifically stain cell nuclei into a segmented image comprising one or more cells having delineated nuclei and cytoplasm regions, the method comprising:

(a) providing a computer system having:

a processor configured to execute instructions; and

a memory or a storage device operatively coupled to the processor, wherein at least one of the memory and the storage device is configured to store:

a model generated from training data comprising a plurality of training images of biological samples, the training images comprising regions identified as at least one of nuclei, cells and background;

the sample image of the biological sample stained with the one or more non-nuclear cell marker stains; and

processor-executable instructions,

(b) accessing, in the memory or the storage device, the model and the sample image of the biological sample stained with the one or more non-nuclear cell marker stains;

(c) generating, by applying the model to the sample image of the biological sample stained with the one or more non-nuclear cell marker stains, a nuclei probability map comprising predicted nuclei regions; and a cell probability map comprising predicted cell regions;

(d) extracting, by the processor, a binary nuclear mask from the nuclei probability map;

(e) extracting, by the processor, a nuclei seeds map from the binary nuclear mask, the nuclei seeds map comprising extracted individual nuclei seeds separated by delineated nuclei regions;

(f) applying, by the processor, the extracted nuclei seeds to the sample image; and

(g) transforming the sample image into the segmented image comprising one or more cells having delineated nuclei and cytoplasm regions.

2. A method according to claim 1 , wherein the sample image comprises a plurality of pixels and step (c) further comprises determining a probability scale of each pixel of the plurality of pixels being a background pixel, a cell pixel or a nucleus pixel.

3. A method according to claim 2 , wherein the probability scale is a continuous scale.

4. A method according to claim 1 , wherein step (d) further comprises performing at least one of a blob detection and a multi-level thresholding.

5. A method according to claim 4 , further comprising:

identifying regions of nuclei with different sizes on the nuclei probability map;

applying a multi-level thresholding on the identified regions of nuclei;

assigning binary value to each pixel of a plurality of pixels in the nuclei probability map; and

extracting the binary nuclear mask from the nuclei probability map.

6. A method according to claim 1 , wherein step (e) further comprises performing a shape-based watershed segmentation.

7. A method according to claim 6 , further comprising:

determining, by the processor, an inverse distance transform of the binary nuclear mask;

applying, by the processor, an extended h-minima transform to the determined inverse distance transform;

extracting, by the processor, regional minima from the binary nuclear mask; and

applying, by the processor, using the regional minima as seeds, a seeded watershed transform to the determined inverse distance transform.

8. A method according to claim 7 , wherein applying the extended h-minima transform comprises applying an H-minima transform at a depth h to suppress regional minima having a depth less than h, and extracting the regional minima from a resulting image with the suppressed regional minima.

9. A method according to claim 8 , wherein the depth h is a user-defined value.

10. A method according to claim 8 , wherein the depth h has a default value of 3 μm.

11. A method according to claim 1 , wherein the sample image is preprocessed for intensity adjustment, the preprocessing comprises performing, by the processor, at least one of the steps of:

denoising of the sample image;

background illumination correction of the sample image; and

rescaling of the sample image.

12. A method according to claim 1 , wherein the sample image is an enhanced sample image generated by applying a pixel-level weighting to the sample image.

13. A method according to claim 1 , wherein step (g) further comprises determining the image background and assigning background labels based on multi-level thresholding.

14. A method according to claim 13 , wherein the multi-level thresholding is performed by:

identifying, by the processor, a number of nuclei regions in the nuclei probability map and an expected area of cell regions in the cell probability map, and

selecting, by the processor, a threshold value which results in an area estimated to be closest to the expected area of the cell regions.

15. A method according to claim 12 , wherein step (g) further comprises performing a seeded watershed segmentation using the assigned background labels and extracted nuclei seeds comprising individual nuclei seeds separated by delineated nuclei regions.

16. A computer system for transforming a sample image of a biological sample stained with only one or more non-nuclear cell marker stains which do not specifically stain cell nuclei into a segmented image comprising one or more cells having delineated nuclei and cytoplasm regions, the computer system comprising:

a processor configured to execute instructions;

a memory or a storage device operatively coupled to the processor, one or both of the memory and the storage device configured to store:

a model generated from training data comprising a plurality of training images of biological samples, the training images comprising regions identified as at least one of nuclei, cells and background;

the sample image of the biological sample stained with the one or more non-nuclear cell marker stains; and

processor-executable instructions that, when executed by the processor, cause acts to be performed comprising:

(a) accessing, in the memory or the storage device, the model and the sample image of the biological sample stained with the one or more non-nuclear cell marker stains;

(b) generating, by applying the model to the sample image of the biological sample stained with the one or more non-nuclear cell marker stains, a nuclei probability map comprising predicted nuclei regions; and a cell probability map comprising predicted cell regions;

(c) extracting, by the processor, a binary nuclear mask from the nuclei probability map;

(d) extracting, by the processor, a nuclei seeds map from the binary nuclear mask, the nuclei seeds map comprising extracted individual nuclei seeds separated by delineated nuclei regions;

(f) applying, by the processor, the extracted nuclei seeds to the sample image; and

(g) transforming the sample image into the segmented image comprising one or more cells having delineated nuclei and cytoplasm regions.