An Ad Hoc Random Initialization Deep Neural Network Architecture for Discriminating Malignant Breast Cancer Lesions in Mammographic Images

Figure 1

Workflow of our method. The original training set provided by CBIS-DDSM is further divided into a new “training set” and a “validation set.” The new training set is employed to fit the model parameters, and the validation set is employed to validate and compare the performance of each model on an unbiased set of images. The final model is chosen accordingly to its performance of the validation set and its performance quantified in an unbiased manner on the test set. Overall, the split was as follows: training set (1158 images), validation set (160 images), and test set (378 images).