Table 4.

Comparative performance of three different CNNs, each with the addition of few-shot learning layers, in their ability to identify four known categories of pathology in VCE footage (Whipples, ulcer, bleeding, and angioectasia) as well as a fifth unknown category. Abbreviations: Visual Geometry Group (VGG), Residual Network (ResNet).

Model	Accuracy	Precision	Sensitivity	F-score
VGGNet + Few Shot	89.3%	89.5%	88.8%	89.0%
ResNet + Few Shot	85.4%	82.3%	82.4%	82.4%
AlexNet + Few Shot	90.8%	91.4%	90.9%	91.0%