PhenoScout: foundation AI models for end-to-end tissue analysis of multiplex IF slides

Digital pathology has enabled reliable and robust routine workflows for high-throughput, remote analysis of H&E and

multiplex immunofluorescence (mIF), no such routines are available as of now. Key challenges include heterogeneous
antibody panels and difficulty generalizing algorithms across different chemistries. To address this, we present a novel
end-to end workflow combining ZEISS Axioscan 7 spatial biology, SlideStream scan automation software, and
Mindpeak’s PhenoScout AI system. At the core is PhenoScout, a library of foundation AI models designed for tissue
analytics across wide variety of mIF and Immunohistochemistry (IHC) stainings, requiring no prior image analysis

An antibody panel, representative for interrogation of tumor microenvironment phenotypes (FOXP3, CD3, Ki67, CD8,
CD68, PanCK) was identified. Protocols for mIF staining of all markers and separate IHC stainings were developed and
used to stain a cohort of tumor tissue sections. Slides were digitized using the ZEISS Axioscan 7 spatial biology system
with SlideStream workflow automation software (Figure1), ensuring consistent, automated, high resolution image
capture. Images were processed through PhenoScout’s foundation models, which have NOT been trained on this
image set. PhenoScout includes pre-trained AI modules for epithelial region detection using cytokeratin markers, cell
center detection using nuclear staining and marker positivity classification for stainings with a nuclear (e.g. FoxP3, ER,
PR) or membrane (e.g. CD3, Her2, PD-L1) morphology. The results were reviewed qualitatively for accuracy of positive

Three-Step Image Analysis Process on Mindpeak mIF platform with correlating example of a representative IO-assay at Concept Life Sciences