Figure 2.

Overview of the steps involved in our implementation. The segmentation dataset is used to train and test the DL segmentation model. The segmentation model is used over the classification dataset to generate masks and extract hand-crafted features for level-1 of the Cascade generalization pipeline. The classification dataset is also used to train a DL classification model (level-0), which is used with the hand-crafted features to train conventional classifiers for the final diagnosis classification.