ColoRectalCADx: Expeditious Recognition of Colorectal Cancer with Integrated Convolutional Neural Networks and Visual Explanations Using Mixed Dataset Evidence

Table 7

Comparison of accuracies and AUC for various CNN architectures.

Individual CNN	CNN training accuracy in %	CNN testing accuracy in %	AUC in %
AlexNet	80.86	72.3	96.62
DarkNet-19	77.55	61.95	95.62
ResNet-50v2	73.41	75.78	96.37
DenseNet-201	79.59	76.31	96.76
EfficientNet-B7	62.48	70.67	95.7
VGG-16	67.51	66.78	95.64
VGG-19	63.38	64.54	94.59
NasNetLarge	79.67	71.14	96.46
InceptionResNetV2	79.71	74.44	96.57
Integrated CNN	CNN training accuracy in %	CNN testing accuracy in %	AUC in %
ADaRDEV^2I-22	76.32	72.2	96.16
ADaRDEV^2-22	72.11	69.85	95.46
RDEV^2-22	73.57	70.75	94.89
ADaDR-22	77.42	72.2	95.09
ADaR-22	67.55	63.6	94.4
DaRD-22	72.59	72.13	95.28
DEV-22	61.71	61.13	94.66
ADa-22	59	54.61	93.04
RD-22	68.26	67.53	94.46
RV-22	34.27	35.69	84.8
AD-22	71.96	64.13	94.44
DaR-22	70.15	66.59	91.96