The Neural Network Diagnostic Model
The neural network diagnostic model for distinguishing malignant breast cancer and benign disease for breast biopsy outcome predictions is shown in Figure 3. This diagnostic model contained an input vector processor, two-stage neural network classification unit, and two-lever hard-limit classifier. The input vector processor first normalized each input of the 5 attributes (including BIRADS, age, shape, margin, and density) by dividing each by its maximum value and then calculated two additional products of combined attributes, “Age*BIRADS” and “Shape*BIRADS.” Thus, for each clinical instance, there was a total of 7 attributes resulting from the outputs of the input vector processor. The 7 attributes were simultaneously fed into the two-stage neural network classification unit. This unit had the same structure as shown in Figure 2, with the final weights loaded from the training model, where R = 7 attributes and S1 = 40 hidden neurons in the tan-sigmoid transfer functions in the first neuron layer