Table 7.

The complexity analysis of the three classification stages of RADIC.

Model	Input Data Dimesnion to the Model	Number of the model hyperparameters parameters (N)	Number of Layers	Training Complexity per Layer (O)
Stage I of Classification
DenseNet-201	Images of Size 224x224x3	16.5 M	201	$O\left(k\bullet n\bullet d^2\right)$ [96] k: kernel size n: sequence length (number of input data) d: representation dimension
DarkNet-53	Images of Size 256x256x3	41.0 M	53
MobileNet	Images of Size 224x224x3	3.5 M	28
Stage II of Classification
Deep Features Extraction from Radiomics-based Generated Images and Reduction with FWHT and SVM classifiers	500 Features	SVM C: Regularization parameter Gamma: width of the Kernel $\mathit{L}\mathit{D}\mathit{A}$ $N=\left(C-1\right)\left(p+1\right)$ C: number of class labels. p: number of features	–	SVM $O(m^{2}p+m^3$) [97] p: number of features m: number of input samples LDA $O(\frac{3}{2}m\bullet p\bullet t+\frac{s}{2}{t}^3$) [98] p: number of features m: number of input samples s: the average number of non-zero features of one sample t: min (m,p)
Stage III of Classification
Incorporated Deep Features of the three CNNs using DCT and SVM classifiers	500 Features	SVM C: Regularization parameter Gamma:width of the Kernel $\mathit{L}\mathit{D}\mathit{A}$ $N=\left(C-1\right)\left(p+1\right)$ C: number of class labels. p: number of features	–	SVM $O(m^{2}p+m^3$) p: number of features m: number of input samples LDA $O(\frac{3}{2}m\bullet p\bullet t+\frac{s}{2}{t}^3$) p: number of features m: number of input samples s: the average number of non-zero features of one sample t: min (m,p)