TABLE 2.

Summary of studies using artificial intelligence in IBD diagnosis.

References	Published year	Aim of study	Type of AI	Number of subjects	Outcomes
Maeda et al., 2019	2019	Prediction of persistent histologic inflammation associated with UC	SVM	Training set: 12,900 images from 87 patients. Test set: 9935 images from 100 patients	Sensitivity: 74%, specificity: 97%, accuracy: 91%
Mahapatra et al., 2016	2016	Segmentation of CD from abdominal MRI	Active learning framework combined with semi-supervised learning	70 patients (fivefold cross validation)	Dice Metric: 92.4%, Hausdorff distance: 7.0 mm
Hubenthal et al., 2015	2015	Diagnostics of IBD	SVM	114 patients	AUROC: 0.75–1.00
Mossotto et al., 2017	2017	Classification of Pediatric Inflammatory Bowel Disease	SVM	239 patients	Accuracy: 82.7% (model utilizing combined endoscopic/histological data achieved the best performance)
Douglas et al., 2018	2018	Classification of disease state and treatment outcome in pediatric Crohn’s disease	RF	Intestinal biopsies of 20 treatment-naïve CD and 20 control pediatric patients	Accuracy: 84.2% (model utilizing 16S taxonomic datasets achieved the best performance)

IBD, inflammatory bowel disease; UC, ulcerative colitis; SVM, support vector machine; CD, Crohn’s disease; MRI, magnetic resonance images; AUROC, area under receiver operating characteristic; RF, random forest.