Method and system for automated detection of immunohistochemical (IHC) patterns
A method and system for automated detection of immunohistochemical (IHC) patterns. The method and system is able to automatically differentiate an epithelial cell part from a non-epithelial cell part of a digital image of a tissue sample to which an IHC compound has been applied. The method and system help to automatically correct errors made in a manual interpretation of a pathological or other manual analysis of tissue samples for diseases such as a human cancer.
1. An automated method for automated detection of immunohistochemical patterns, comprising: enhancing a digital image of a biological sample to which an immunohistochemical (IHC) compound has been applied; wherein the enhancing step includes: computing digital image statistics of pixel values for each of red, green and blue color planes independently in the digital image; determining whether a standard deviation from the computed digital image statistics in each of the red, green and blue color planes is less than a predetermined value, and if so, determining a maximum intensity value of an original range of intensity values using a distribution of pixel values in the red, green and blue color planes independently; and mapping pixel values in each color plane such that pixel values in an original range of intensity values are mapped to an enhanced range of intensity values to contrast enhance the digital image,
removing pre-determined types of unwanted cells in the enhanced digital image from consideration; wherein the step of removing pre-determined types of unwanted cells in the enhanced digital image from consideration includes: segmenting the enhanced digital image into a plurality of objects using a pre-determined Gaussian kernel; calculating a plurality of elongation ratios for the plurality of objects; and removing the pre-determined types of unwanted cells using the calculated plurality of elongation ratios,
identifying a plurality of cells of interest in the enhanced digital image; wherein the step of identifying a plurality of cells of interest in the enhanced digital image includes: segmenting the enhanced digital image into a plurality of objects using a pre-determined Gaussian kernel; calculating a plurality of curves of symmetry for the plurality of objects; and identifying the plurality of cells of interest using the calculated plurality of curves of symmetry,
identifying one or more areas of interest in the identified plurality of cells of interest in the enhanced digital image; and
removing cell artifacts from consideration in the one or more identified areas of interest, thereby creating and displaying one or more enhanced areas of interests used for creating a medical diagnosis or prognosis.
2. The method of claim 1 further comprising providing a computer readable medium having stored therein instructions for causing one or more processors to execute the steps of the method.
3. The method of claim 1 wherein the IHC compound includes a Haematoxylin and Eosin (WE) stain.
4. The method of claim 1 wherein the predetermined types of unwanted cells include stromal cells or lymph cells.
5. The method of claim 1 wherein the cells of interest include epithelial cells.
6. The method of claim 1 wherein the step of identifying a plurality of cells of interest includes identifying a plurality of cells of interest including estrogen receptors (ER) or progesterone receptors (PR) reacting to the IHC compound.
7. The method of claim 1 wherein the step of identifying a plurality of cells of interest includes identifying a plurality of cells of interest including Her2/neu receptors reacting to the IHC compound.
8. The method of claim 1 wherein the biological sample includes a plurality of human cells.
9. The method of claim 8 wherein the plurality of human cells potentially includes one or more human cancer cells.
10. The method of claim 9 wherein the one or more human cancer cells are breast cancer cells.
11. The method of claim 1 wherein the medical diagnosis or prognosis includes a diagnosis or prognosis for a human cancer.
12. The method of claim 1 wherein the step of removing pre-determined types of unwanted cells in the enhanced digital image from consideration includes: segmenting the enhanced digital image into a plurality of objects using a pre-determined Gaussian kernel; calculating a plurality of curves of symmetry for the plurality of objects; and removing the pre-determined types of unwanted cells using the calculated plurality of curves of symmetry.
13. The method of claim 1 wherein the unwanted cells include stromal cells.
14. The method of claim 1 wherein the step of removing pre-determined types of unwanted cells in the enhanced digital image from consideration includes: identifying a plurality of unstained cell pixels in the enhanced digital image; identifying a plurality of connected unstained cell pixels based on a pre-determined IHC pattern; and removing the identified plurality of connected unstained cell pixels from further consideration in the enhanced digital image.
15. The method of claim 14 wherein the unstained cell pixels include unstained non-epithelial cell pixels.
16. The method of claim 15 wherein the pre-determined IHC pattern includes an eight connectivity between unstained non-epithelial cell pixels.
17. The method of claim 1 wherein the step of identifying a plurality of cells of interest in the enhanced digital image includes: segmenting the enhanced digital image into a plurality of connected components of cells of interest; determining a plurality of convex hulls for the plurality of connected components; identifying the plurality of cells of interest using the determined plurality of convex hulls.
18. The method of claim 17 where the plurality of cells of interest include stained epithelial cells.
19. The method of claim 17 wherein the step of determining a plurality of convex hulls for the plurality of connected components includes: centering a plurality of neighborhood masks around a plurality of pixels on a boundary of a cell from a connected component; and determining whether the plurality of pixels on the boundary of the cell belong to a convex hull of the cell, and if so, calculating a convex hull factor.
20. The method of claim 17 wherein the neighborhood masks include three-by-three neighborhood masks.
21. The method of claim 1 wherein the step of identifying one or more areas of interest in the identified plurality of cells of interest in the enhanced digital image includes: segmenting the enhanced digital image into a plurality of connected components of cells of interest; calculating a number of pixels in a connected component; calculating a circularity of the connected component; and determining with the calculated number of pixels and calculated circularity whether the connected component includes an epithelial cell area.
22. The method of claim 1 wherein the step of identifying one or more areas of interest in the identified plurality of cells of interest in the enhanced digital image includes: segmenting the enhanced digital image into a plurality of connected components of cells of interest; scanning a plurality of rows in a connected component; scanning a plurality of columns of the connected component; and determining with a number of scanned rows or a number of scanned columns whether the connected component includes an epithelial cell area.
23. The method of claim 1 , wherein the step of removing cell artifacts from consideration in the one or more identified areas of interest includes: calculating a gradient within an identified area of interest; calculating a run length ratio within the identified area of interest; calculating a spike count within the identified area of interest; and removing artifacts from the identified area of interest using the calculated gradient, run length ratio and spike counts, thereby leaving an identified area of interest identified as an epithelial cell area without artifacts.