Automated stereology for determining tissue characteristics

1. A method of performing computerized stereology, the method comprising:

capturing, by an imager, a Z-stack of images of a three-dimensional (3D) object, the Z-stack of images being a sequence of images of the 3D object captured in increments having a step size along a Z-axis of the 3D object; and

determining, by a processor in operable communication with a non-transitory computer-readable medium, a stereology parameter of the Z-stack of images using a deep learning structured model,

wherein the non-transitory computer-readable medium has instructions stored thereon that, when executed, cause the processor to use the deep learning structured model to determine the stereology parameter of the Z-stack of images.

2. The method according to claim 1 , wherein the deep learning structured model comprises a convolutional neural network (CNN).

3. The method according to claim 2 , wherein the deep learning structured model further comprises an adaptive segmentation algorithm (ASA) to segment stained cells from images created from the Z-stack of images.

4. The method according to claim 3 , wherein the images created from the Z-stack of images, from which the ASA segments stained cells, are extended depth of field (EDF) images.

5. The method according to claim 2 , wherein the CNN comprises a convolutional layer, a Rectified Linear Unit (ReLU) layer, a pooling layer, and a fully connected (FC) layer.

6. The method according to claim 5 , wherein the convolutional layer comprises a plurality of filters configured to detect features of the Z-stack of images.

7. The method according to claim 6 , wherein each filter of the plurality of filters has the same biases and weights, and analyzes the same number of input neurons, as every other filter of the plurality of filters.

8. The method according to claim 7 , wherein each filter convolves across dimensions of an image of the Z-stack of images and computes a dot product of the respective filter and an image subset from among the Z-stack of images to generate a matrix or a feature map, and

wherein each filter preserves a spatial relationship between pixels of the image of the Z-stack of images as it convolves across dimensions of the image.

9. The method according to claim 8 , wherein the ReLU layer applies an activation function to the matrix or the feature map to introduce a non-linear element to the matrix or the feature map, and

wherein the pooling layer reduces dimensions of the matrix or the feature map, generating an output matrix or an output image for the FC layer.

10. A system for performing computerized stereology, the system comprising:

an imager configured to capture a Z-stack of images of a three-dimensional (3D) object, the Z-stack of images being a sequence of images of the 3D object, and the imager being configured to capture the sequence of images of the 3D object in increments having a step size along a Z-axis of the 3D object;

a processor in operable communication with the imager; and

a non-transitory computer-readable medium in operable communication with the processor and having instructions stored thereon that, when executed, cause the processor to determine a stereology parameter of the Z-stack of images using a deep learning structured model,

wherein the deep learning structured model comprises a convolutional neural network (CNN).

11. The system according to claim 10 , wherein the deep learning structured model further comprises an adaptive segmentation algorithm (ASA) to segment stained cells from images created from the Z-stack of images.

12. The system according to claim 11 , wherein the images created from the Z-stack of images, from which the ASA segments stained cells, are extended depth of field (EDF) images.

13. The system according to claim 10 , wherein the CNN comprises a convolutional layer, a Rectified Linear Unit (ReLU) layer, a pooling layer, and a fully connected (FC) layer.

14. The system according to claim 13 , wherein the convolutional layer comprises a plurality of filters configured to detect features of the Z-stack of images.

15. The system according to claim 14 , wherein each filter of the plurality of filters has the same biases and weights, and analyzes the same number of input neurons, as every other filter of the plurality of filters.

16. The system according to claim 15 , wherein each filter is configured to convolve across dimensions of an image of the Z-stack of images and compute a dot product of the respective filter and an image subset from among the Z-stack of images to generate a matrix or a feature map, and

wherein each filter is configured to preserve a spatial relationship between pixels of the image of the Z-stack of images as it convolves across dimensions of the image.

17. The system according to claim 16 , wherein the ReLU layer is configured to apply an activation function to the matrix or the feature map to introduce a non-linear element to the matrix or the feature map, and

wherein the pooling layer is configured to reduce dimensions of the matrix or the feature map, generating an output matrix or an output image for the FC layer.

18. A method for computerized stereology, the method comprising:

capturing, by an imager, a Z-stack of images of a three-dimensional (3D) object, the Z-stack of images being a sequence of images of the 3D object captured in increments having a first step size along a Z-axis of the 3D object;

determining, by a processor in operable communication with a non-transitory computer-readable medium, a stereology parameter of the Z-stack of images using a deep learning structured model,

wherein the non-transitory computer-readable medium has instructions stored thereon that, when executed, cause the processor to use the deep learning structured model to determine the stereology parameter of the Z-stack of images, and

wherein the deep learning structured model comprises the following steps:

constructing images from the Z-stack of images;

performing clump segmentation on the constructed images by binarizing the constructed images using a threshold determined by estimating a Gaussian Mixture Model to pixel intensities;

preprocessing the constructed images by converting the constructed images into grayscale and opening by reconstruction followed by closing by reconstruction;

performing watershed segmentation on the constructed images, wherein regional minima are extracted as foreground markers and boundaries between regions are used as background markers, and the watershed segmentation is applied using the background and foreground makers that overlap with clumps;

constructing Voronoi diagrams and smoothing, including constructing a Voronoi map using centers of foreground regions and refining region boundaries using a Savitzy-Golay filter; and

determining the stereology parameter of the Z-stack of images.

19. The method according to claim 18 , wherein the constructed images are extended depth of field (EDF) images.

20. The method according to claim 19 , wherein the deep learning structured model comprises a convolutional neural network (CNN), and

wherein the CNN comprises a convolutional layer, a Rectified Linear Unit (ReLU) layer, a pooling layer, and a fully connected (FC) layer.

21. The method according to claim 19 , wherein the deep learning structured model further comprises the following steps:

generating a 3D computer simulation of the 3D object;

generating an X-stack of sections being a sequence of sections of the 3D computer simulation captured in increments having a second step size along an X-axis of the 3D computer simulation;

determining a first number of cells contained in the 3D computer simulation from an X-direction;

generating a Y-stack of sections being a sequence of sections of the 3D computer simulation captured in increments having a third step size along a Y-axis of the 3D computer simulation; and

determining a second number of cells contained in the 3D computer simulation from a Y-direction.