Bayesian and non-Bayesian probabilistic models for medical image analysis

Imaging Science and Biomedical Engineering Division, Medical School, University of Manchester, Stopford Building, Oxford Road, Manchester M13 9PT, UK
Image and Vision Computing (Impact Factor: 1.96). 01/2003; DOI: 10.1016/S0262-8856(03)00072-6
Source: DBLP

ABSTRACT Bayesian approaches to data analysis are popular in machine vision, and yet the main advantage of Bayes theory, the ability to incorporate prior knowledge in the form of the prior probabilities, may lead to problems in some quantitative tasks. In this paper we demonstrate examples of Bayesian and non-Bayesian techniques from the area of magnetic resonance image (MRI) analysis. Issues raised by these examples are used to illustrate difficulties in Bayesian methods and to motivate an approach based on frequentist methods. We believe this approach to be more suited to quantitative data analysis, and provide a general theory for the use of these methods in learning (Bayes risk) systems and for data fusion. Proofs are given for the more novel aspects of the theory. We conclude with a discussion of the strengths and weaknesses, and the fundamental suitability, of Bayesian and non-Bayesian approaches for MRI analysis in particular, and for machine vision systems in general.

  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We report a general method of Bayesian estimation that uses prior measurements to improve the signal-to-noise ratio of parametric images computed from dynamic PET scanning. In our method, the ordinary weighted least squares cost function is augmented by a penalty term to yield Phi(K,S)=minK{(C-f(K))(T)Omega(C)(-1)(C-f(K))+SPhi(K,S=0)(K-K;)(T)Omega(K)(-1)(K-K;)}, where C is a PET concentration history and Omega(C) is its variance, f is the model of the concentration history, K=[k(1),k(2),...,k(m)](T) is the parameter vector, K; is the vector of population means for the model parameters, Omega(K) is its covariance, Phi(K)(K,S=0) is the conventional weighted sum of squares. S>0 is chosen to control the balance between the prior and new data. Data from a prior population of subjects are analyzed with standard methods to provide maps of the mean parameter values and their variances. As an example of this approach we used the dynamic image data of 10 normal subjects who had previously been studied with (11)C-raclopride to estimate the prior distribution. The dynamic data were transformed to stereotactic coordinates and analyzed by standard methods. The resulting parametric maps were used to compute the voxel-wise sample statistics. Then the cohort of priors was analyzed as a function of S, using nonlinear least squares estimation and the cost function shown above. As S is increased the standard error in estimating BP in single subjects was substantially reduced allowing measurement in BP in thalamus, cortex, brain stem, etc. Additional studies demonstrate that a range of S values exist for which the bias is not excessive, even when parameter values differ markedly from the sample mean. This method can be used with any kinetic model so long as it is possible to compute a map of a priori mean parameters and their variances.
    NeuroImage 05/2009; 45(4):1183-9. · 6.25 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Present trends in biotechnological manufacturing are mostly based on multidisciplinary approach. The fast development of computers enabled the planning and designing of biomedical implants through the 3D view model of bone defects in orofacial region. Regarding this fact, the detailed model analysis is of special importance since it will enable the creation of realistic skull models in the future through a new generation of 3D printers. A wide range of new biocompatible materials, including polymethyl methacrylate and titanium alloys are available for the creation of biomedical implants.
    Medicina (; Vol.45 No.2. 01/2009;
  • [Show abstract] [Hide abstract]
    ABSTRACT: We present Bayesian methodologies and apply Markov chain sampling techniques for exploring normal mixture models with an unknown number of components in the context of magnetic resonance imaging (MRI) segmentation. The experiments show that by estimating the number of components using sample-based approaches based on variable dimension models the discriminating power of the estimated components is improved. Two different MCMC methods are compared to perform the segmentation of simulated magnetic resonance brain scans, the reversible jump MCMC model and the Dirichlet process (DP) mixture model. The preference given to the Dirichlet process mixture model is discussed.
    Computer methods and programs in biomedicine 12/2008; 94(1):1-14. · 1.56 Impact Factor

Full-text (3 Sources)

Available from