A Normal Mammogram Classifier using Binary Decision Trees with Multiresolution Features

摘要

The majority of screening mammograms are normal, and it may be beneficial if a normal mammogram detection system were designed to help radiologists identify normal regions, allowing them to focus on suspicious areas. In this paper, we present a binary decision classifier based on global features extracted from different levels of 2-D Quincunx Wavelet Decomposition of pre-processed regional mammograms. This is fundamentally different than other approaches, which identify normal mammograms by detecting the type of lesion. Our approach presented here is independent of the types of abnormalities present. The experiment clearly shows that one can classify most normal mammogram regions, keeping the clinically critical misclassification rate of abnormal regions small.