Predicting Crohn's disease severity in the colon using mixed cell nucleus density from pseudo labels

Area of Science:

• Digital pathology
• Computational biology
• Gastroenterology

Background:

• Crohn's disease (CD) is a chronic inflammatory bowel disease lacking a cure.
• Whole slide images (WSIs) of H&E stained colon biopsies offer potential for discovering cellular features linked to disease severity.
• Previous studies utilized cell nuclei features for slide-level predictions, but not specifically for classifying normal vs. active CD in WSIs.

Purpose of the Study:

• To computationally analyze H&E stained colon biopsy WSIs from CD patients.
• To classify normal tissue versus active CD using nucleus density histograms.
• To assess the predictive performance of individual and combined cell nucleus types.

Main Methods:

• Utilized 413 WSIs from CD patient biopsies.
• Calculated normalized nucleus density histograms for six cell types (neutrophils, eosinophils, epithelial cells, lymphocytes, plasma cells, connective cells).
• Employed a support vector machine with truncated SVD for classification and four-fold cross-validation.

Main Results:

• Neutrophil nucleus density was the most predictive individual feature, yielding an AUC of 0.92 ± 0.0003.
• Adding other cell nucleus types improved performance in early cross-validation rounds and on the test set.
• Performance gains diminished significantly when features beyond neutrophils were included.

Conclusions:

• Neutrophil nucleus density is a strong indicator for differentiating active Crohn's disease from normal colon tissue in WSIs.
• While combining cell features offers marginal improvements, neutrophils alone provide substantial predictive power.
• This computational approach aids in understanding CD pathology through digital image analysis.