Table 2.

Automated quantification of RCM features is feasible

TiME feature	Test/validation data	Validation result
Inflammation (U-Net model)	9.7%	Dice coefficient: 0.72
Vessel segmentation	5.2%	Dice coefficient: 0.29–0.78
Trafficking – optimization	10.9%	Spearman r: 0.79–0.82
Trafficking – final validation	2.5%	Spearman r: 0.74–0.89

Summary of results for validation for each automated quantification approach using data from n = 92 distinct lesions. Quantification of immune cells (leukocyte-like, dendritic cells and macrophages) using a 3-class UNet model resulted in Dice coefficient of 0.72. Segmentation of blood vessels for quantifying area and diameter of vessels showed a wide range of accuracy. Automated leukocyte trafficking counts were correlated with manual counts as ground truth during optimization and validation, demonstrate Spearman r between 0.74–0.89 depending on track length.