Segmentation of placental tissue using immunofluorescent staining and an artificial intelligence-based analysis workflow.

Abstract

The placenta continuously remodels in response to maternal and fetal signals, with proteins dynamically regulated across placental regions. However, methods to evaluate proteins within these distinct regions are limited. To address this, we developed a placental tissue classifier to segment regions corresponding to the villous core, villous trophoblast (VT) and the intervillous space (IVS) using HALO AI imaging analysis. Tissue sections of biopsies from human term placentas embedded in OCT or paraffin (FFPE) were stained by immunofluorescence with antibodies to placental alkaline phosphatase (PLAP), syndecan-1 (SDC-1), vimentin or E-cadherin. The placental tissue classifier was trained using PLAP and DAPI staining with distinct image-feature patterns to distinguish VT from the villous core and the IVS on stained sections. The accuracy of HALO to measure area and intensity of staining in classified regions was demonstrated by staining for SDC-1 on VT and vimentin on stromal cells. As expected, SDC-1 staining was low in the villous core, and higher on VT than in the IVS, whereas vimentin staining was only detected in the villous core. By applying this analysis method to images collected by whole slide scanning microscopy, the area of staining was increased 245 times compared to a single field of view, which increases the probability of detecting pathological changes in different regions of the placenta. The advantages of using this validated classifier are the speed and accuracy of analysis across large tissue areas, the flexibility to detect other cell types, and to expand to single villi analysis.