Increasing the sensitivity of detection of targeted MRI contrast agents using image analysis

Magnetic resonance imaging (MRI) typically images water protons and gives excellent images of soft tissues since the signal intensity depends on both the water distribution and the MR properties of the water protons, which can vary significantly between and within tissues. Furthermore, these MR properties can be a dynamic function of tissue physiology and thus image contrast can report not only on tissue structure but also on aspects of tissue function. This capability of MRI to image not only tissue architecture but also tissue function has been expanded considerably in recent years by the advent of 'molecular imaging'. Molecular imaging typically involves the introduction of a paramagnetically labelled probe molecule or contrast agent that reports in the MR image of tissue morphology, on some aspect of underlying tissue biochemistry or physiology. Molecular imaging in MRI now encompasses a number of different applications, including pH mapping, cell surface receptor mapping, gene reporter constructs which permit mapping of gene activity and methods for labelling cells with paramagnetic labels that allow their tissue trafficking to be imaged in real time in vivo. However, a fundamental limitation of MR-based molecular imaging methods is their relative insensitivity when compared to radiochemical (PET, SPECT) or optical (bioluminescence, near infra-red fluorescence) molecular imaging modalities. A solution to this problem, which we are pursuing currently, is to increase the concentration of the paramagnetic label and/or the effect that it has on signal intensity in the MR image. However, the problem with these approaches are that they almost invariably involve an increase in the size of the label, which can compromise the function of the ligand or cell, to which the label is attached. We are proposing here a parallel approach to increasing the sensitivity of label detection in MR imaging by using sophisticated image analysis methods, of a type used by the astrophysics community to de-noise images and enhance feature recognition in images of star clusters. The capability to perform reliable automated segmentation of MR images does not currently exist within the MR imaging community. Some simple approaches have been tried, but none are capable of performing the task to the required level of accuracy and reliability. We will use a novel Bayesian algorithm that uses a generative model for the imaging process, combined with efficient methods for obtaining the statistically optimal result. We will start by performing conventional texture analyses of MR images of well-defined in vitro model systems that mimic the punctate nature of cell labelling in vivo and at the same time model different labelled cell densities, at both macroscopic (mm) and microscopic levels (micron). The basic texture analysis methods employed previously will be extended and will provide a baseline for comparison with the results obtained subsequently using the novel Bayesian approach. All analyses will be directly validated against the known geometry (cell density, label content, cell cluster spacing) of the gelatin phantoms. These experiments will determine whether the sensitivity of labelled cell detection can be improved by image analysis. The best analysis methods will be further validated using pre-existing in vivo data of the type already published by us and for which we have good histological data from corresponding tissue sections obtained post-mortem. The expected outcomes of this project are image analysis methods that improve the sensitivity of labelled cell and molecule detection using MRI. By improving the sensitivity of detection, these methods may also permit lower label concentrations to be used, with attendant improvements in cell and molecular function.

Molecular imaging is revolutionising the application of magnetic resonance imaging (MRI) in biology and medicine. Molecular imaging in MRI includes the use of contrast media targeted at specific molecular entities, such as cell surface receptors, and labelling cells with paramagnetic agents, which allows them to be tracked non-invasively in vivo. However, a fundamental limitation of molecular imaging using MR, is its relative lack of sensitivity as compared to other imaging modalities. One solution to this problem is to increase the relaxivity of the paramagnetic tag. However, this almost invariably involves an increase in its size, which can compromise the function of the ligand or cell, to which the label is attached. Another approach is to use sophisticated image analysis methods, of a type used by the astrophysics community to de-noise images and enhance feature recognition, to improve the sensitivity of label detection. This approach, which can be pursued in parallel with increasing the relaxivity of the paramagnetic tag, is the subject of this proposal. The capability to perform reliable automated segmentation of MR images does not currently exist within the MR imaging community. Some simple approaches have been tried, but none are capable of performing the task to the required level of accuracy and reliability. We will use a novel Bayesian segmentation algorithm that uses a generative model for the imaging process, combined with efficient methods for obtaining the statistically optimal result. The technique is based on Markov-chain Monte Carlo sampling from the posterior distribution of the parameters in the model of the object. We have shown that this corresponds to the theoretically optimal method for parameterised object detection and characterisation. The expected outcomes of this project are image analysis methods that improve the sensitivity of labelled cell and molecule detection using MRI.