Table 3.

Diagnostic performance results of the first stage classification (RCC vs. AML) using different individual feature sets along with multilayer perceptron artificial neural network (MLP-ANN) classification models. The RC-CAD system diagnostic performance using the combined features outperformed the diagnostic abilities using individual feature sets. Sens: sensitivity, Spec: specificity, DSC: Dice coefficient of similarity, ${\mathrm{hl}}_n$: size of hidden layer n.

RCC vs. AML Classification Performance (Mean ± SD ≈)
Feature Set	Sens%	Spec%	DSC	MLP-ANN
Set 1	94.1 ± 1.5	97.9 ± 1.5	0.96 ± 0.01	hl${}_1$ = 10 nodes
Set 2	92.4 ± 2.9	95.1 ± 3.5	0.94 ± 0.02	hl${}_1$ = 10 nodes
Set 3	94.9 ± 2.2	95.3 ± 2.5	0.95 ± 0.02	hl${}_1$ = 10 nodes
Set 4	92.0 ± 2.4	96.6 ± 2.0	0.94 ± 0.02	hl${}_1$ = 10 nodes, hl${}_2$ = 5 nodes
Set 5	82.7 ± 4.1	91.7 ± 2.0	0.87 ± 0.02	hl${}_1$ = 10 nodes
RC-CAD	95.3 ± 2.0	99.9 ± 0.4	0.98 ± 0.01	hl${}_1$ = 50 nodes, hl${}_2$ = 25 nodes

Hyperparameters: MLP-ANN (optimization function: trainlm, max epochs = 500, goal = 0, max validation failure = 6, min gradient = ${10}^{-7}$, training gain ($\mu$): initial $\mu$ = 0.001, $\mu$ decrease factor = 0.1, $\mu$ increase factor = 10, max $\mu$ = 1e${}^{10}$).