Figure 3.

The architecture of the proposed tumor classification approach. Texture analysis is performed by dividing the tumor into a group of nonoverlapping ROIs and extracting texture features from each ROI. A selected set of texture features are used to classify each individual ROI and compute its posterior tumor class likelihood. The posterior tumor class likelihoods of the individual ROIs are combined. Morphological analysis is performed by extracting morphological features from the outlined tumor and employing a selected set of the features to predict the posterior tumor class likelihood. Decision fusion is then used to combine the posterior tumor class likelihoods obtained using the texture and morphological analyses and determine the tumor class.