Table 5.

Comparative performance of three different CNNs in the identification of normal versus abnormal frames in VCE footage. The Residual Network (ResNet152) model incorporates only the frame in question into its prediction. The m0110 multi-frame model relies on the frame in question plus the frame immediately preceding it. The m1110 multi-frame model incorporates the frame in question and the two preceding frames.

Model	Accuracy	Precision	Sensitivity	F1 score	Video Prediction Time (minutes)
ResNet152	96.6%	90.3%	91.1%	90.7%	2.0
m0110	97.1%	91.0%	93.5%	92.3%	1.7
m1110	97.5%	91.5%	94.8%	93.1%	2.2