Intelligent Dermatologist Tool for Classifying Multiple Skin Cancer Subtypes by Incorporating Manifold Radiomics Features Categories

Table 2

Classification accuracy of incorporating manifold feature levels.

Incorporated manifold feature sets	Linear	Quadratic	Cubic
Single high-level features incorporated with low-level features
ResNet-50 + DWT-A	95.7	96.1	96.2
ResNet-50 + LBP	96.1	96.4	96.4
ResNet-50 + LBP + DWT-A	96.5	96.9	96.8
DarkNet-53 + DWT-A	94.6	94.9	95.1
DarkNet-53 + LBP	95.8	96	96.1
DarkNet-53 + DWT-A + LBP	95.7	95.9	96
DenseNet-201 + DWT-A	96.5	97.1	97.2
DenseNet-201 + LBP	97.1	97.5	97.4
DenseNet-201 + DWT-A + LBP	97.5	97.9	97.9
Two high-level feature sets incorporated
ResNet-50 + DarkNet-53	95.9	96.3	96.6
ResNet-50 + DenseNet-201	97.7	98	98.1
DenseNet-201 + DarkNet-53	97.9	98.1	98
Two high-level feature sets incorporated with low-level feature sets
ResNet-50 + DenseNet-201 + DWT-A	98.2	98.5	98.5
ResNet-50 + DenseNet-201 + LBP	97.8	98.1	98.1
ResNet-50 + DenseNet-201 + DWT-A + LBP	98.2	98.6	98.5
ResNet-50 + DarkNet-53 + DWT-A	96.4	96.9	96.9
ResNet-50 + DarkNet-53 + LBP	96.7	96.9	97
DenseNet-201 + DarkNet-53 + DWT-A	98.2	98.4	98.5
DenseNet-201 + DarkNet-53 + LBP	97.9	98.4	98.4
DenseNet-201 + DarkNet-53 + DWT-A + LBP	98.3	98.6	98.6
Three high-level feature sets incorporated
ResNet-50 + DenseNet-201 + DarkNet-53	98.2	98.5	98.5
Three high-level feature sets incorporated with low-level feature sets
ResNet-50 + DenseNet-201 + DarkNet-53 + DWT-A	98.6	98.8	98.8
ResNet-50 + DenseNet-201 + DarkNet-53 + LBP	98.5	98.8	98.7
ResNet-50 + DenseNet-201 + DarkNet-53 + DWT-A + LBP	98.7	99	99