COMPUTERIZED SYSTEMS AND METHODS FOR ELECTRONIC IMAGE ANALYSIS FOR IDENTIFYING CELLS
Disclosed herein, inter alia, are computer-implemented methods for analyzing electronic images of a tissue sample.
1 . A computer-implemented method for analyzing electronic images of a tissue sample, comprising:
computationally grouping, using a machine learning model, cells based on a first signal signature to generate groups of cells of the tissue sample; and
computationally combining, using at least the machine learning model, a second signal signature within each group of cells to generate aggregates of signal signatures.
2 . The computer-implemented method of claim 1 , wherein the machine learning model is trained to categorize cells of the issue sample based on morphological features and related signal signatures.
3 . The computer-implemented method of claim 1 , wherein computationally grouping cells based on the first signal signature comprises the machine learning model using image analysis to quantify morphological features of the cells.
4 . The computer-implemented method of claim 1 , wherein a training dataset of the machine learning model comprises labeled cellular images with a plurality of morphological features.
5 . The computer-implemented method of claim 4 , wherein the morphological features comprise at least one of spatial proximity, geometric analysis, topological analysis, cluster density, connectivity within a defined radius, irregularities in cellular shape and/or size, membrane roughness, cytoplasmic texture, nucleus-to-cytoplasm ratio.
6 . The computer-implemented method of claim 1 , wherein the machine learning model comprises a graph-based aggregation model comprising at least one graph-based clustering algorithm, wherein the computationally grouping comprises the graph-based aggregation model computationally grouping cells into subgroups using graph-based aggregation.
7 . The computer-implemented method of claim 1 , wherein the machine learning model comprises a graph-based aggregation model, and wherein the computationally grouping comprises:
identifying similar groups of cells using at least unsupervised clustering on a data matrix; and representing distinct cell compositions and/or distinct cell states, using at least the unsupervised clustering with a fixed resolution.
8 . The computer-implemented method of claim 1 , wherein the machine learning model comprises a graph-based aggregation model, and wherein the computationally grouping comprises:
generating, using at least joint transcriptional and proteomic profiles of the graph-based aggregation model, a neighborhood graph; and
deriving, using at least the graph-based aggregation model, phenotypic similarity among the group of cells by applying at least unsupervised clustering on the neighborhood graph.
9 . The computer-implemented method of claim 1 , further comprising calculating, using at least the machine learning model and at least one clustering algorithm, a Euclidean distance and/or cosine similarities between pairs of expression vectors of a spatial dataset comprising locations of cells of the tissue sample.
10 . The computer-implemented method of claim 1 , further comprising:
computationally classifying, using at least the machine learning model and one or more unsupervised clustering algorithms, cells of the tissue sample into phenotypically and transcriptomically similar groups within a tissue section; and
mapping, using at least the machine learning model, locations of neurons alongside transcriptomically similar groups.
11 . The computer-implemented method of claim 1 , further comprising:
computationally classifying, using at least the machine learning model, a similarity score, and a segmentation algorithm, cells of the tissue sample into segmented phenotypically similar groups.
12 . The computer-implemented method of claim 1 , wherein the machine learning model computationally groups the cells based on the first signal signature using at least k-means clustering.
13 . The computer-implemented method of claim 1 , wherein the machine learning model computationally groups the cells based on the first signal signature using at least unsupervised hierarchical clustering.
14 . The computer-implemented method of claim 1 , wherein the machine learning model computationally groups the cells based on the first signal signature using at least unsupervised dimensionality reduction clustering.
15 . The computer-implemented method of claim 1 , wherein the machine learning model computationally groups the cells based on the first signal signature using at least machine learning clustering.
16 . A computer-implemented method for analyzing electronic images a tissue sample, comprising:
clustering, using at least a machine learning model trained to categorize cells of the tissue sample based on morphological features and related signal signatures, cells based on a first signal signature to generate groups of cells of the tissue sample; and
computationally combining, using at least the machine learning model, a second signal signature within each group of cells to generate aggregates of signal signatures.
17 . The computer-implemented method of claim 16 , further comprising:
calculating, using at least the machine learning model and at least one clustering algorithm, a Euclidean distance and/or cosine similarities between pairs of expression vectors of a spatial dataset comprising locations of cells of the tissue sample.
18 . The computer-implemented method of claim 16 , further comprising:
computationally classifying, using at least the machine learning model, a similarity score, and a segmentation algorithm, cells of the tissue sample into segmented phenotypically similar groups.
19 . The computer-implemented method of claim 16 , wherein the morphological features comprise at least one of spatial proximity, geometric analysis, topological analysis, cluster density, connectivity within a defined radius, irregularities in cellular shape and/or size, membrane roughness, cytoplasmic texture, nucleus-to-cytoplasm ratio.
20 . The computer-implemented method of claim 16 , wherein the computationally grouping cells based on the first signal signature comprises the machine learning model using image analysis to quantify morphological features of the cells.
21 . The computer-implemented method of claim 16 , wherein a training dataset of the machine learning model comprises labeled cellular images with a plurality of morphological features.
22 . A non-transitory computer-readable medium storing instructions that, when executed by a processor, perform a method for analyzing a tissue sample, the method comprising:
computationally grouping, using at least a machine learning model trained to categorize cells of the tissue sample based on morphological features and related signal signatures, cells based on a first signal signature to generate groups of cells of the tissue sample; and
computationally combining, using at least the machine learning model, a second signal signature within each group of cells to generate aggregates of signal signatures.
23 . The non-transitory computer-readable medium of claim 22 , wherein the morphological features comprise at least one of spatial proximity, geometric analysis, topological analysis, cluster density, connectivity within a defined radius, irregularities in cellular shape and/or size, membrane roughness, cytoplasmic texture, nucleus-to-cytoplasm ratio.
24 . The non-transitory computer-readable medium of claim 22 , wherein the computationally grouping cells based on the first signal signature comprises the machine learning model using image analysis to quantify morphological features of the cells.