Addressing Image Variability While Learning Classifiers for Detecting Clusters of Micro-calcifications

In the context of computer aided mammography, many standard algorithms (e.g. SVM and neural networks) have been used for detecting lesions. However, these general purpose learning methods make implicit assumptions like sample independence that are commonly violated. A new ensemble algorithm is proposed to explicitly account for the small fraction of outlier images which tend to produce a large number of false positives. A bootstrapping procedure is used to ensure that the candidates from these outlier images do not skew the statistical properties of the training samples. We compared a standard state-of-the-art method (SVM) for detecting clusters of micro-calcifications, with our ensemble algorithm. This algorithm significantly improved the test set results, especially in the operating region of interest (around 0.2 FP per image).