US 12,406,510 B2
Method for the detection, segmentation and morphological mapping on neural cell images of the whole brain
Tsang-Wei Tu, Rockville, MD (US); Yi-Yu Hsu, Rockville, MD (US); Chao-Hsiung Hsu, College Park, MD (US); Artur Agaronyan, College Park, MD (US); and Paul C. Wang, Rockville, MD (US)
Assigned to Howard University, Washington, DC (US)
Filed by HOWARD UNIVERSITY, Washington, DC (US)
Filed on Mar. 28, 2022, as Appl. No. 17/706,252.
Claims priority of provisional application 63/166,953, filed on Mar. 26, 2021.
Prior Publication US 2022/0309810 A1, Sep. 29, 2022
Int. Cl. G06T 5/92 (2024.01); G06T 7/00 (2017.01); G06T 7/11 (2017.01); G06T 7/149 (2017.01); G06T 7/194 (2017.01); G06V 20/69 (2022.01)
CPC G06V 20/695 (2022.01) [G06T 5/92 (2024.01); G06T 7/0012 (2013.01); G06T 7/149 (2017.01); G06T 7/194 (2017.01); G06T 2207/20084 (2013.01); G06T 2207/30016 (2013.01)] 16 Claims
OG exemplary drawing
 
1. A method for morphological analysis of neural cells, the method executed by one or more processors, the method comprising:
receiving an immunohistochemistry (IHC) image of one or more neural cells;
detecting the one or more neural cells in the IHC image using a first neural network, wherein the detecting includes identifying boundaries of the one or more neural cells, wherein the detecting the one or more neural cells comprises:
identifying one or more signature features of a neural cell from a plurality of patches; and
based on identifying the one or more signature features, generating a bounding box enclosing the neural cell;
generating a segmentation cell mask using a second neural network based on the identified boundaries of the one or more neural cells, wherein the generating the segmentation cell mask comprises:
receiving the bounding box enclosing the neural cell:
extending the bounding box enclosing the neural cell to cover a cell body of the neural cell and to cover extending processes of the neural cell;
receiving a cell location for the neural cell, the cell location including a unique identification associated with a patch from the plurality of patches where the extended bounding box is located and corresponding coordinates of the patch; and
generating the segmentation cell mask based on the extended bounding box and the received cell location; and
classifying the one or more neural cells based on the generated segmentation cell mask using a trained classification model.