Cerebral Microbleed Detection Via Fourier Descriptor with Dual Domain Distribution Modeling

Abstract

In this study we propose a novel cerebral microbleed (CMB) detection technique which simultaneously utilizes distribution information in dual domains and shape information obtained by a Fourier descriptor, and does not rely on a large set of training data. Specifically, the dual domain distribution modeling aims to simultaneously examine the image content in both gradient domain and voxel domain, while the Fourier descriptor further characterize the shape of the candidate region. A set of labeled data is used to form the dualdomain distribution as well as the distribution of Fourier coefficients. Then the probability of a region containing a CMB is estimated by combining the two types of distributions. Experimental results show that the proposed approach is efficient and desirable for scenarios where the number of samples is limited.