Five-hundred thirty-five (535) digitalized biopsies were graded against three standard indexes. A neural network was trained to distinguish remission from activity. The model predicted endoscopic-verified remission or activity with approximately 80% accuracy.