A BAYESIAN-APPROACH TO SYNTHETIC MAGNETIC-RESONANCE-IMAGING

• 成果类型: Article
• 署名作者: GLAD, IK; SEBASTIANI, G
• 刊物名称: BIOMETRIKA
• ISSN/ISSBN: 0006-3444
• 发表日期: 1995
• 页码: 237250
• 关键词: statistical-analysis; images; restoration; relaxation; noise

摘要

Synthetic magnetic resonance imaging involves the estimation, based on a set of measured images with noise, of three basic physical quantities that are nonlinearly related to the observations. The methods currently available for this ill-conditoned inverse problem either do not provide sufficiently accurate estimates or require time-consuming data collection. We formulate this nonlinear problem in a Bayesian framework, taking into account knowledge about the physics of the magnetic resonance imaging experiment, statistical properties of the experimental noise, and prior information about the underlying physical quantities, modelled by a suitable Markov random held. A new multilayer Markov random field is proposed. Inference is drawn by means of Markov chain Monte Carlo methods or iterated conditional modes. Some examples are included to demonstrate how synthetic magnetic resonance imaging by this approach can be performed in an accurate and reliable way.