Adaptive kernel learning for detection of clustered microcalcifications in mammograms

Abstract

Adaptive kernel learning is a Bayesian learning technique developed recently, which can be viewed as a variant of the well known relevance vector machine (RVM). The purpose of adaptive kernel learning is to automatically optimize the parameters associated with the kernel basis functions in a predictive model. In this paper, we explore the use of adaptive kernel learning for detection of clustered microcalcifications in mammograms, which is formulated as a two-class classification problem. The proposed approach is tested using a set of clinical mammograms, and compared with an RVM classifier developed previously. It is demonstrated that the adaptive kernel learning classifier can achieve better detection performance than the RVM classifier; it also yields a much sparser model with lower computational complexity.