Cell Type Prediction for Intestine Tissue Samples from Brightfield Histology via Deep Learning
Cell types present in a biopsy provide information on disease processes and organ health and are useful in a research setting. Knowledge of cell types, their distributions, and their spatial locations relative to other cells can assist a clinician or researcher in their diagnosis or their investigations. The multiplex imaging technology like co-detection by indexing (CODEX) can provide spatial context for protein expression and detect cell types on a whole slide basis. The CODEX workflow also allows for hematoxylin and eosin (H&E) staining on the same sections used in molecular imaging. In this work, we develop a deep learning pipeline to automatically detect cell types in histologically stained intestine tissue sections by leveraging the ground truth annotations obtained from the corresponding CODEX images. The data consists of 32 paired CODEX and histology samples obtained from four donors where cell composition is determined by multiplexed imaging and single-nucleus RNA and open chromatin assays. We then train a semantic segmentation model on 24 annotated images and holdout 8 samples for testing. In this analysis, we achieved a notable ~4.5X improvement with balanced accuracy over the random baseline for 12 classes. Our work has potential to be used in clinical and basic research to study cell type and cell distributions with functional tissue units to obtain a holistic overview of intestine tissue biology, and the study can be expanded to other organ types as well.
