Multi-pass adaptive voting for nuclei detection in histopathological images

An apparatus for detecting cellular nuclei in a histopathology image, comprising:

• a processor;
• a memory;
• an input/output interface;
• a set of circuits;
• and
• an interface to connect the processor, the memory, the input/output interface and the set of circuits, where the set of circuits includes:
• an image acquisition circuit that acquires a histopathology image, where the histopathology image includes a plurality of pixels;
• an edge detection circuit that generates an nuclear edge map based, at least in part, on the histopathology image, where the nuclear edge map includes a subset of the plurality of pixels;
• a gradient field generation circuit that generates a gradient field based, at least in part, on the nuclear edge map, where the gradient field generation circuit computes an inverse gradient for a member of the subset of the plurality of pixels;
• a gradient field refinement circuit that generates a refined gradient field based, at least in part, on the gradient field;
• and
• a multi-pass adaptive voting (MPAV) circuit that iteratively generates a nuclear voting map based on the refined gradient field, and that identifies a cellular nucleus in the histopathology image based on the nuclear voting map and a global threshold, where the nuclear voting map includes a pixel.

A method for detecting cellular nuclei in a medical image, the method comprising:

• accessing a digital image of a stained tissue slide, where the digital image includes a set of pixels;
• generating a gradient field based on the digital image by producing a smoothed digital image and by detecting a nuclear edge in the smoothed digital image, where the nuclear edge defines a convex hull, where the nuclear edge includes a pixel, where the gradient field includes an inverse gradient associated with the pixel in the nuclear edge, where the convex hull has a centroid;
• generating a refined gradient field by computing an angular difference between the inverse gradient and a vector derived from the convex hull centroid and the pixel;
• upon determining that the angular difference is greater than a threshold angular difference:
• reversing the direction of the inverse gradient;
• generating a voting image based on the refined gradient field;
• adaptively iteratively updating the voting image;
• computing a global threshold;
• identifying a cellular nuclei based on the global threshold and the voting image;
• and
• controlling an automated tissue grading system to count cellular nuclei identified in the digital image.