Quantification of nuclei and their architecture: Model performance
Nuclei in placental chorion are arranged closely to each other, and have varying intensities (i.e.
knots are brighter, and deeper in tissue signal to noise ratio decreases). This makes intensity-based segmentation very challenging. The machine-based learning approach Stardist 33was employed to segment nuclei. An overview, of the developed imaging methodology and quantification, is given in Figure S4.
Representative images of placental nuclei of term control and EO-PE are illustrated in Figure 2 and Figure S5. Knots appeared bright and very dense (2D images Figures 2A, 2E). Corresponding 2D images (Figures 2B, 2F) and 3D images (Figures 2C, 2G) segmented with Stardistdemonstrate the labels of all individual nuclei identified. These images were used to calculate nuclear density. The classification of knots performed with KnotMiner is presented in Figures 2D and 2H.
Although non-significant, EO-PE placentas tended towards higher nuclear density compared to preterm control placenta (Figure 2I). Knot fraction (the villous volume occupied by knots) tended to be higher in EO-PE compared to preterm control placenta (Figure 2J). While significance of LO-PE could not be tested (single value), IUGR knot fraction differed non-significantly from term control. LO-PE and IUGR placenta both tended towards increased knot fraction compared to term control placenta. Notably, preterm placenta did not have a lower knot fraction than term control placenta (Figure 2J).
Knot shape was on average slightly (non-significantly) flatter and (significantly) less elongated in term control- and preterm placenta (Figure 2K) compared to IUGR and EO-PE placenta, respectively. Individual knot shape quantification is represented in Figure S6.