Immunohistochemistry-Free Enhanced Histopathology of the Rat Spleen Using Deep Learning

Area of Science:

• Immunotoxicology
• Computational Pathology

Background:

• Assessing lymphocyte populations in lymphoid organs during toxicology studies is challenging due to sampling variability and limited cytologic detail in H&E staining.
• While immunohistochemistry provides definitive T- and B-cell characterization, routine toxicologic evaluations rely solely on H&E slides.

Purpose of the Study:

• To develop and validate a deep learning model for precise, compartment-specific quantification of splenic lymphoid tissues using H&E stained slides.
• To establish a quantitative baseline for normal splenic lymphoid compartment area and cellularity.

Main Methods:

• A deep learning model was trained on H&E stained rat spleen slides, using co-registered images from destained and dual-labeled (CD3, CD79a) slides as ground truth.
• The model was validated for its accuracy in identifying splenic compartments: periarteriolar lymphoid sheaths, follicles, germinal centers, and marginal zones.

Main Results:

• The deep learning model achieved high accuracy (97.8% Dice similarity coefficient) in quantifying splenic compartments directly from H&E stained slides.
• The model was successfully applied to determine the normal range of splenic lymphoid compartment area and cellularity.

Conclusions:

• Deep learning models can accurately quantify splenic lymphoid compartments from routine H&E slides, overcoming limitations of traditional histopathology.
• This approach offers potential for improved accuracy, precision, and time efficiency in enhanced immune system histopathology evaluations.
• Expansion to other lymphoid tissues and integration into routine toxicologic pathology workflows is recommended.