MAGNETIC RESONANCE (MR) IMAGE ANALYSIS — A STATISTICAL APPROACH

Abstract

Following the rapid advance in applications of Magnetic Resonance (MR) imaging, there is an upsurge in contemporary image analysis. Image analysis techniques are implemented in different ways in the various frameworks of image processing. This paper proposes a new statistical approach. It first analyzes statistical properties of MR data (bulk magnetizations, MR signals, and k-space samples); then based on these statistics and through MR image reconstruction, derives statistical properties of MR image (a single pixel, any two pixels, and a group of pixels — i.e. an image region); finally by using these image statistics establishes a new stochastic image model (the correlated Finite Normal Mixture (cFNM)) and develops a new statistical image analysis method (the extended Expectation-Maximization (eEM) algorithm). The results obtained by using this cFNM-eEM framework demonstrate promise. The novel feature of this approach is that it eliminates heuristic assumptions and ad-hoc conditions in currently employed image analysis techniques and therefore leads to more accurate image analysis results. Although this approach is developed for the conventional MRI, but, following the same strategy and with some modifications, it can be extended to more advanced MRI protocols and other medical imaging modalities.