Lymphocytes subtyping on H&E slides with automatic labelling through same-tissue stained ImmunoFluorescence images

Etienne Pochet, Luis Cano Ayestas, Qi Tang, Roger Trullo

Keywords: Deep learning, Immunofluorescent (IF) staining, H&E staining, Lymphocyte sub-typing, T-cell

TL;DR: This work combines H&E and IF stained images to automatically label immune cell types in H&E samples. The labels are used to train deep learning models that identify lymphocytes subtypes on H&E images

Abstract:

Accurate identification and classification of immune cells within tissue samples are critical for understanding disease mechanisms and predicting treatment responses as a cornerstone for personalized medicine. Traditional histopathology relies on hematoxylin and eosin (H&E) staining, which provides structural context but lacks specificity for immune cell sub-types, preventing pathologists from more precise identification. In contrast, immunofluorescent (IF) staining enables precise targeting of specific markers, but this recently developed technology is very costly and not widely applied in clinical practice yet.
In this work, we propose a method to leverage registered pairs of H&E and IF stained images from the same tissue to automatically generate cell type labels for H&E from IF marker expression, allowing for precise identification. In particular, we demonstrate the feasibility of lymphocyte sub-typing from H\&E images by training cell-level classifiers to accurately distinguish T-cells subtypes (CD45 / CD3e / CD4 / CD8a).

Submission Number: 4