Deep-learning systems and methods for joint cell and region classification in biological images
Automated systems and methods for training a multilayer neural network to classify cells and regions from a set of training images are presented. Automated systems and methods for using a trained multilayer neural network to classify cells within an unlabeled image are also presented.
1. A method for classifying cells, the method comprising:
receiving an image comprising a plurality of pixels depicting a plurality of cells and one or more tissue regions;
detecting a cell nucleus for each cell of the plurality of cells;
computing a foreground segmentation mask for the image based on the detected cell nuclei;
identifying each cell of the plurality of cells by filtering the image with the computed foreground segmentation mask;
generating a predictive label for each pixel of the plurality of pixels with a multilayer neural network trained with training images depicting cells and tissue regions, wherein the predictive label indicates a cell type of a plurality of cell types for the respective pixel; and
assigning a cell label to each identified cell, the cell label corresponding to one cell type of the plurality of cell types, wherein the cell label is assigned based on the generated predicative labels for pixels corresponding to the respective identified cell indicating the one cell type.
2. The method of claim 1 , further comprising quantifying differently labeled individual cells and computing an expression score.
3. The method of claim 1 , further comprising quantifying a number of lymphocytes in a tumor region or a stromal region.
4. The method of claim 1 , wherein the assigning of the cell label to each identified cell comprises (i) quantifying a number of pixels bearing each predictive label within the identified cell; and (ii) assigning as the cell label the predictive label having a greatest quantity.
5. A non-transitory computer-readable storage medium encoded with instructions executable by one or more processors of a computing system to cause the computing system to perform one or more operations comprising:
receiving an image comprising a plurality of pixels depicting a plurality of cells and one or more tissue regions;
detecting a cell nucleus for each cell of the plurality of cells;
computing a foreground segmentation mask for the image based on the detected cell nuclei;
identifying each cell of the plurality of cells by filtering the image with the computed foreground segmentation mask;
generating a predictive label for each pixel of the plurality of pixels with a multilayer neural network trained with training images depicting cells and tissue regions, wherein the predictive label indicates a cell type of a plurality of cell types for the respective pixel; and
assigning a cell label to each identified cell, the cell label corresponding to one cell type of the plurality of cell types, wherein the cell label is assigned based on the generated predicative labels for pixels corresponding to the respective identified cell indicating the one cell type.
6. The non-transitory computer-readable storage medium of claim 5 , wherein the one or more operations further comprise quantifying differently labeled individual cells and computing an expression score.
7. The non-transitory computer-readable storage medium of claim 5 , wherein the one or more operations further comprise quantifying a number of lymphocytes in a tumor region or a stromal region.
8. The non-transitory computer-readable storage medium of claim 5 , wherein the assigning of the cell label to each identified cell comprises (i) quantifying a number of pixels bearing each predictive label within the identified cell; and (ii) assigning as the cell label the predictive label having a greatest quantity.
9. A system comprising:
one or more processors; and
a non-transitory computer-readable memory storing instructions which, when executed by the one or more processors, cause the one or more processors to perform one or more operations comprising:
receiving an image comprising a plurality of pixels depicting a plurality of cells and one or more tissue regions;
detecting a cell nucleus for each cell of the plurality of cells in the image;
computing a foreground segmentation mask for the image based on the detected cell nuclei;
identifying each cell of the plurality of cells by filtering the image with the computed foreground segmentation mask;
generating a predictive label for each pixel of the plurality of pixels with a multilayer neural network trained with training images depicting cells and tissue regions, wherein the predictive label indicates a cell type of a plurality of cell types for the respective pixel; and
assigning a cell label to each identified cell, the cell label corresponding to one cell type of the plurality of cell types, wherein the cell label is assigned based on the generated predicative labels for pixels corresponding to the respective identified cell indicating the one cell type.
10. The system of claim 9 , wherein the one or more operations further comprise quantifying differently labeled individual cells and computing an expression score.
11. The system of claim 9 , wherein the one or more operations further comprise quantifying a number of lymphocytes in a tumor region or a stromal region.
12. The system of claim 9 , wherein the assigning of the cell label to each identified cell comprises (i) quantifying a number of pixels bearing each predictive label within the identified cell; and (ii) assigning as the cell label the predictive label having a greatest quantity.