Region-Based Feature Extraction Using TRUS Images

This paper introduces a new feature extraction method that assists in identifying cancerous regions in prostate Trans Rectal UltraSound (TRUS) images. The main aim of this paper is to elicit the radiologists´ medical knowledge by creating a set of fuzzy rules that are then brought to radiologists to fine tune. The proposed method uses a fuzzy inference system (FIS) to mimic the expert radiologists´ interpretation of the TRUS images. Nine elected features are fed into the proposed FIS to produce a new aggregated feature set. The membership functions and the fuzzy rules of the FIS are generated using the estimated probability density functions of the features. Experiments show that the new aggregated feature set is at least 13% better than each of the feature alone, when measured using mutual information (MI).