MxIF staining, whole tissue imaging, and AI-driven image analysis of human BM core biopsy tissues at single-cell resolution. (A) Experimental strategy is shown. Archival normal BM specimens from 29 individuals were collected. Paraffin-embedded BM core biopsy tissues were sectioned on to positively charged glass microscope slides. Automated iterative MxIF staining was performed with the following panel of antibodies plus 4′,6-diamidino-2-phenylindole: CD34, CD117, CD38, CD71, CD61, and CD15. Whole slide images (WSIs) were captured with the Vectra Polaris Automated Quantitative Pathology Imaging System and WSI tiles were spectrally unmixed in InForm (Akoya) and subsequently stitched in HALO (Indica Labs). Spectrally unmixed WSIs were parsed through BostonGene-developed AI/machine learning (ML)–based custom Python pipelines. (B) Computational pipeline involving CNN based cell-type identification (n = 11) using marker mean fluorescence intensities (top); in parallel, machine learning–based masks for bone trabeculae, fat, and CD34+ vasculature were generated (bottom). Morphologic features of identified cell types (n = 11) were assessed. Distances between different cell types and to various structures of interest were determined. (C) Validation of detected cell types based on MxIF staining (left); cell types are specified vertically and antibody channels listed horizontally (scale bars = 10 microns). A total of 11 different cell types were identified: HSPCs (median = 1041 per case, total = 30463 cells, density = 130 cells per mm2), myeloblasts (median = 190 per case, total = 9102 per case, and density = 39 cells per mm2), promyelocytes (median = 246 per case, total = 10 928 per case, and density = 50 cells per mm2), proerythroblasts (median = 199 per case, total = 10 074 per case, and density = 40 cells per mm2), maturing granulocytes and monocytes (MMCs, median = 16 500.0 per case, total = 597 387 cells, density = 2500 cells per mm2), erythroid normoblasts (median = 11 248 per case, total = 339 276 per case, density = 1450 cells per mm2), megakaryocytes (median = 372 per case, total = 12 739 cells, density = 60 cells per mm2), likely B-cell precursors [median = 1979 per case, total = 71 396 per case, density = 290 cells per mm2], and likely plasma cells (median = 86.0 per case, total = 3520 per case, density = 10 cells per mm2). Nucleated cells lacking expression of any of the interrogated antigens were classified as NHEs (median = 12 682 per case, total = 392 506 per case, density = 1740 cells per mm2). Mast cells and CD34+ endothelial cells were also detected. MxIF WSI of a representative healthy bone marrow biopsy specimen is shown (right). y.o., years old.

MxIF staining, whole tissue imaging, and AI-driven image analysis of human BM core biopsy tissues at single-cell resolution. (A) Experimental strategy is shown. Archival normal BM specimens from 29 individuals were collected. Paraffin-embedded BM core biopsy tissues were sectioned on to positively charged glass microscope slides. Automated iterative MxIF staining was performed with the following panel of antibodies plus 4′,6-diamidino-2-phenylindole: CD34, CD117, CD38, CD71, CD61, and CD15. Whole slide images (WSIs) were captured with the Vectra Polaris Automated Quantitative Pathology Imaging System and WSI tiles were spectrally unmixed in InForm (Akoya) and subsequently stitched in HALO (Indica Labs). Spectrally unmixed WSIs were parsed through BostonGene-developed AI/machine learning (ML)–based custom Python pipelines. (B) Computational pipeline involving CNN based cell-type identification (n = 11) using marker mean fluorescence intensities (top); in parallel, machine learning–based masks for bone trabeculae, fat, and CD34+ vasculature were generated (bottom). Morphologic features of identified cell types (n = 11) were assessed. Distances between different cell types and to various structures of interest were determined. (C) Validation of detected cell types based on MxIF staining (left); cell types are specified vertically and antibody channels listed horizontally (scale bars = 10 microns). A total of 11 different cell types were identified: HSPCs (median = 1041 per case, total = 30463 cells, density = 130 cells per mm2), myeloblasts (median = 190 per case, total = 9102 per case, and density = 39 cells per mm2), promyelocytes (median = 246 per case, total = 10 928 per case, and density = 50 cells per mm2), proerythroblasts (median = 199 per case, total = 10 074 per case, and density = 40 cells per mm2), maturing granulocytes and monocytes (MMCs, median = 16 500.0 per case, total = 597 387 cells, density = 2500 cells per mm2), erythroid normoblasts (median = 11 248 per case, total = 339 276 per case, density = 1450 cells per mm2), megakaryocytes (median = 372 per case, total = 12 739 cells, density = 60 cells per mm2), likely B-cell precursors [median = 1979 per case, total = 71 396 per case, density = 290 cells per mm2], and likely plasma cells (median = 86.0 per case, total = 3520 per case, density = 10 cells per mm2). Nucleated cells lacking expression of any of the interrogated antigens were classified as NHEs (median = 12 682 per case, total = 392 506 per case, density = 1740 cells per mm2). Mast cells and CD34+ endothelial cells were also detected. MxIF WSI of a representative healthy bone marrow biopsy specimen is shown (right). y.o., years old.

Close Modal

or Create an Account

Close Modal
Close Modal