Table 7. Evaluation results for the proposed E-CNN, its individuals (modified TL models) when number of epochs = 30, and the standard TL models on Kather’s colon histopathlogical images dataset (Kather et al., 2016) based on the average accuracy, sensitivity, specificity, and average standard deviation (STD) in 10 runs, best results in bold.

Pretrained models	Accuracy	Sensitivity	Specificity
CNN architecture in Rachapudi & Lavanya Devi (2021)	77.0	–	–
ResNet152 feature extraction in Rachapudi & Lavanya Devi (2021)	80.004 ± 1.307	–	–
NASNetMobile feature extraction in Ohata et al. (2021)	89.263 ± 1.704	–	–
Modified DenseNet121	89.4. ± 0.56	78.32 ± 0.49	99.0 ± 0.2
Modified MobileNetV2	87.27 ± 0.57	76.4 ± 0.5	98.7 ± 0.43
Modified InceptionV3	89.04 ± 0.36	78.0 ± 0.32	99.4 ± 0.64
Modified VGG16	83.3 ± 1.38	72.9 ± 1.26	99.1 ± 0.0
Proposed E-CNN (product)	91.28 ± 3.4	79.97 ± 3.0	99.1 ± 0.0
Proposed E-CNN (Majority voting)	90.63 ± 4.03	79.4 ± 4.02	99.1 ± 0.0