MIKAIA mIF Cell Segmentation AI Gallery

Cell Segmentation

MIKAIA‘s FL Cell Analysis App includes a flexible mIF cell segmentation AI with 3 modes:

1. Segment nuclei and dilate: User selects the nuclei marker (e.g., DAPI). The AI then segments the nuclei and applies a user-defined dilation radius.  This allows for independent measurement of the expression of all selected markers in both the nucleus and the cytoplasm/membrane compartment of each cell.
2. Segment membrane and nucleus: User selects (1) the nuclei marker (e.g., DAPI) and (2) one or multiple (!) cytoplasmic or membranous markers. The app then delineates both the outer contour of the cells and nucleus.
3. Segment cell contours: User selects a single nuclear, cytoplasmic, or membrane marker. The app then delineates the outer contours of the cells, making this mode ideal for scenarios where no nuclei marker is available.

Cell Typing

Once cells have been detected and marker expressions have been measured, the app offers various cell typing modes:

• No cell typing: Users can create multiple annotations for each cell, with one annotation for each expressed marker.
• Cell typing based on coexpression
  • Custom cell lineage map: User can provide cell lineage map, where they define various cell types and specify which markers are required to be positive (or negative) for each type.
  • Default cell typing: If no lineage map is provided, the app will treat each permutation of markers as a distinct cell type by default.
  • Cell typing via unsupervised clustering: This mode employs unsupervised clustering techniques to categorize cells based on their marker expression profiles.
    

---

---

---

Available downstream spatial analyses in MIKAIA^®

The mIF Cell segmentation AI is only the first step. After cells have been detected and typed, a broad range of downstream analyses is available:

• The Cell-Cell Connections App (or cell-cell contacts) yields statistics on direct cell neighborhoods.
• The Proximity Analysis App finds and visualizes the distance between a specified cell type and a target object (e.g., to a tumor interface) in relation to another cell type. This functionality allows users to yield the mean distance between cell types.
• The Cellular Neighborhood App iterates over each cell and then quantifies the cell composition within a defined neighborhood. This neighborhood can be specified by a radius (e.g., 200 µm) and/or by utilizing k-nearest neighbors. The app provides a detailed breakdown of the average neighborhood composition for each cell type, based on varying distances, e.g., in steps of 10 or 25 µm.
• The Grid Analysis App provides statistics on spatial heterogeneity by overlaying a virtual grid onto the analyzed area. For each tile within the grid, the app computes a user-selected metric, such as cell abundance, the ratio of different cell types, and other relevant measures.
• The Spatial Clustering App groups nearby cells of the same type into clusters and then yields statistics in the form of a histogram on whether few large or many small clusters exist.
• The Density Heatmap functionality allows overlaying a heatmap that visualizes the local abundance of selected annotation objects (e.g., one or multiple cell types). Users can interactively set up the heatmap’s sensitivity, bin size, opacity, and other attributes.

An overview of this portfolio of spatial analyses is demonstrated with short videos in our Overview of MIKAIA’s Spatial Analysis Apps for Multiplexed Immunofluorescence Slides.

---

Source: Picard, D. (2022). Human breast tissue microarray stained for PAICS [Data set]. Zenodo. https://doi.org/10.5281/zenodo.7149674

---

Source: Picard, D. (2022). Human breast tissue microarray stained for PAICS [Data set]. Zenodo. https://doi.org/10.5281/zenodo.7149674

---

---

Source: Fraunhofer ITEM-R, PEDRA