Automated tracking of migrating cells in phase-contrast video microscopy sequences using image registration.

Analysis of in vitro cell motility is a useful tool for assessing cellular response to a range of factors. However, the majority of cell-tracking systems available are designed primarily for use with fluorescently labelled images. In this paper, five commonly used tracking systems are examined for their performance compared with the use of a novel in-house cell-tracking system based on the principles of image registration and optical flow. Image registration is a tool commonly used in medical imaging to correct for the effects of patient motion during imaging procedures and works well on low-contrast images, such as those found in bright-field and phase-contrast microscopy. The five cell-tracking systems examined were Retrac, a manual tracking system used as the gold standard; CellTrack, a recently released freely downloadable software system that uses a combination of tracking methods; ImageJ, which is a freely available piece of software with a plug-in for automated tracking (MTrack2) and Imaris and Volocity, both commercially available automated tracking systems. All systems were used to track migration of human epithelial cells over ten frames of a phase-contrast time-lapse microscopy sequence. This showed that the in-house image-registration system was the most effective of those tested when tracking non-dividing epithelial cells in low-contrast images, with a successful tracking rate of 95%. The performance of the tracking systems was also evaluated by tracking fluorescently labelled epithelial cells imaged with both phase-contrast and confocal microscopy techniques. The results showed that using fluorescence microscopy instead of phase contrast does improve the tracking efficiency for each of the tested systems. For the in-house software, this improvement was relatively small (<5% difference in tracking success rate), whereas much greater improvements in performance were seen when using fluorescence microscopy with Volocity and ImageJ.

Evaluating an optical-flow-based registration algorithm for contrast-enhanced magnetic resonance imaging of the breast.

Dynamic contrast-enhanced magnetic resonance imaging studies of the breast are frequently degraded by patient motion. In order to correct for this, any registration algorithm must overcome two major challenges: the highly deformable nature of the breast itself and the need to remove changes in signal intensity due to patient motion whilst leaving potentially significant changes in signal intensity due to changes in contrast agent concentration unchanged. In this paper, we evaluate the use of a non-rigid registration method that uses optical flow equations to drive the displacement of a grid of control points. With conventional optical flow techniques it is assumed that changes in image intensity are solely due to motion, making it unsuitable for use with contrast-enhanced studies. The registration algorithm evaluated in this paper overcomes this problem by including an additional term to account for changes in image intensity. Studies simulating physiologically plausible deformations of the breast together with realistic changes in contrast-enhancement derived from patient studies demonstrate that the algorithm is capable of registering images to sub-voxel accuracy within minutes. This technique has now been successfully incorporated into a breast cancer screening protocol allowing registered images to be provided routinely to the radiologist immediately after the scanning session.

Characterisation of the haemodynamics of the superior mesenteric artery.

In contrast to its prevalence in the surrounding vasculature, occurrence of primary atherosclerotic disease in the superior mesenteric artery (SMA) is rare (Glagov et al., 1988. Hemodynamics and atherosclerosis, Insights and perspectives gained from studies of human arteries. Archives of Pathology and Laboratory Medicine 112(10), 1018-1031; Hansen et al., 2004. Mesenteric artery disease in the elderly. Journal of Vascular Surgery 40(1), 45-52). We hypothesise that this sparing might be attributed to more favourable haemodynamic characteristics in the SMA than in other vessels locally. Dynamic magnetic resonance imaging (MRI) images established that the SMA is highly mobile (Jeays, 2006. Investigation of blood flow in the superior mesenteric artery and its potential influence on atheroma and gut ischaemia. Ph.D. Thesis, University of Sheffield), and thus that an analysis based on rigid geometry might be inappropriate. This paper describes an efficient methodology for the construction of a patient-specific, time-dependent model of an arterial segment and reports the results of a haemodynamic characterisation of the SMA for one individual. A transient computational fluid dynamic (CFD) model was constructed by morphing a parametric mesh constructed from simple geometric primitives. This process has the merit that it is easy to control the element size distribution mapped onto the original geometric primitives. It is robust in operation, and is ideally suited to the generation of dynamic CFD meshes of arterial systems that are free from major pathology. Flow boundary conditions were determined based on phase contrast MRI velocity measurements. Comparative studies with rigid walls and with moving walls, based on the transient data, indicated that, despite the significant motion of the SMA (radial dilation of the order of 10% and translation of the order of the radius), the maximum (spatially and temporally-resolved) wall shear stresses changed by no more than 21.6% of a global norm, and the average change was less than 2.1%.

Efficient computational fluid dynamics mesh generation by image registration.

Most implementations of computational fluid dynamics (CFD) solutions require a discretisation or meshing of the solution domain. The production from a medical image of a computationally efficient mesh representing the structures of interest can be time consuming and labour-intensive, and remains a major bottleneck in the clinical application of CFD. This paper presents a method for deriving a patient-specific mesh from a medical image. The method uses volumetric registration of a pseudo-image, produced from an idealised template mesh, with the medical image. The registration algorithm used is robust and computationally efficient. The accuracy of the new algorithm is measured in terms of the distance between a registered surface and a known surface, for image data derived from casts of the lumen of two different vessels. The true surface is identified by laser profiling. The average distance between the surface points measured by the laser profiler and the surface of the mapped mesh is better than 0.2 mm. For the images analysed, the new algorithm is shown to be 2-3 times more accurate than a standard published algorithm based on maximising normalised mutual information. Computation times are approximately 18 times faster for the new algorithm than the standard algorithm. Examples of the use of the algorithm on two clinical examples are also given. The registration methodology lends itself immediately to the construction of dynamic mesh models in which vessel wall motion is obtained directly using registration.

Photosensitizing activities of picket fence porphyrins in vitro and in vivo.

A set of structurally distinct o-substituted tetraphenylporphyrins, the picket fence porphyrins, were evaluated for their ability to photosensitize tumor mitochondria in vitro, in vivo-in vitro, and tumor implants in situ. Differential photosensitized inactivation efficiencies toward mitochondrial enzymes in vitro are reported for the 12 compounds studied as a function of side chain length and isomer structure. Fluorescence studies in aqueous solution coupled with mitochondrial uptake studies indicate that the observed range of inactivation efficiencies are due to different inherent solubilization properties for the picket fence porphyrins. Studies with the most soluble compound, 3,1-meso-tetrakis(o-propionamidophenyl)porphyrin, using an in vivo-in vitro protocol indicate that a more effective photosensitization can be obtained by using an interval of 4 h between photosensitizer administration and irradiation as compared to 24 h for Photofrin II. Irradiation of tumors in vivo 4 h following administration of 3,1-meso-tetrakis(o-propionamidophenyl)porphyrin, resulted in a mean tumor doubling time more than eight times longer than that observed for untreated tumors. 31P NMR spectroscopy in situ indicated that photodynamic therapy using 3,1-meso-tetrakis(o-propionamidophenyl)porphyrin induced a rapid and significant reduction in high energy phosphate metabolites.

Picket-fence porphyrins as potential phototherapeutic agents.

The synthetic "picket fence" porphyrin, tetra(o-acetamidophenyl)porphine (TAc), as a biological photosensitizer has been evaluated both in vitro and in vivo in mitochondria from the R3230AC mammary tumor. Studies in vitro, consisting of incubation of mitochondria with TAc at a concentration of 4.0 micrograms/ml followed by photolysis, result in the inhibition of cytochrome c oxidase, proton translocating ATPase, succinate dehydrogenase, and malate dehydrogenase. The diminution in activity of the first three enzymes is approximately 2-fold greater than that seen with Photofrin II under the same conditions. Although TAc exists as four isolable atropisomers, no differences among these different forms were observed in their photosensitized inhibition of mitochondrial enzymes. Administration to tumor-bearing rats of TAc i.p. at a dose of 25 mg/kg did result in accumulation of porphyrin within the mitochondria of the R3230AC tumor as determined by subsequent irradiation of isolated mitochondria. The potential utility of TAc and related porphyrins in cancer phototherapy is discussed.

Automatic segmentation of medical images using image registration: diagnostic and simulation applications.

Automatic identification of the boundaries of significant structure (segmentation) within a medical image is an are of ongoing research. Various approaches have been proposed but only two methods have achieved widespread use: manual delineation of boundaries and segmentation using intensity values. In this paper we describe an approach based on image registration. A reference image is prepared and segmented, by hand or otherwise. A patient image is registered to the reference image and the mapping then applied to ther reference segmentation to map it back to the patient image. In general a high-resolution nonlinear mapping is required to achieve accurate segmentation. This paper describes an algorithm that can efficiently generate such mappings, and outlines the uses of this tool in two relevant applications. An important feature of the approach described in this paper is that the algorithm is independent of the segmentation problem being addresses. All knowledge about the problem at hand is contained in files of reference data. A secondary benefit is that the continuous three-dimensional mapping generated is well suited to the generation of patient-specific numerical models (e.g. finite element meshes) from the library models. Smoothness constraints in the morphing algorithm tend to maintain the geometric quality of the reference mesh.

A feasibility study of a rotary planar electrode array for electrical impedance mammography using a digital breast phantom.

A feasibility study of an electrical impedance mammography (EIM) system with a rotary planar electrode array, named RPEIM, is presented. The RPEIM system is an evolution of the Sussex MK4 system, which is a prototype instrument for breast cancer detection. Comparing it with the other planar electrode EIM systems, the rotation feature enables a dramatic increase in the number of independent measurements. To assist impedance evaluation exploiting electrode array rotation, a synchronous mesh method is proposed. Using the synchronous mesh method, the RPEIM system is shown to have superior performance in image accuracy, spatial resolution and noise tolerance over the MK4 system. To validate the study, we report simulations based on a close-to-realistic 3D digital breast phantom, which comprises of: skin, nipple, ducts, acinus, fat and tumor. A digital breast phantom of a real patient is constructed, whose tumor was detected using the MK4 system. The reconstructed conductivity image of the breast phantom indicates that the breast phantom is a close replica of the patient's real breast as assessed by the MK4 system in a clinical trial. A comparison between the RPEIM system and the MK4 system is made based on this phantom to assess the advantages of the RPEIM system.

Automatic image registration of diagnostic and radiotherapy treatment planning CT head images.

PURPOSE: Accurate image registration is an essential step to integrate information from diagnostic and radiotherapy treatment planning (RTP) CT images. In this paper a fully automatic algorithm is presented to quickly register such diagnostic and RTP CT head scans. METHODS AND MATERIALS: The registration algorithm, which regards one image as the moved version of the other, was applied to seven clinically obtained diagnostic and RTP CT data set pairs. During the RTP scan patients were in treatment position and wearing mold masks. Hence, patient position differed strongly in both image sets. Registrations were inspected visually and compared with results obtained minimizing the sum-of-square difference. RESULTS: Registrations were accurate upon visual inspection. Differences between the two algorithms were at subvoxel level. All cases were successfully registered, using several different starting points. Registration calculations took 1-2 minutes. Minimization of the sum-of-square difference took 1-1.5 hours. CONCLUSIONS: The results show that a fast and accurate image registration is achieved without prior segmentation or feature extraction and that the algorithm is robust, which makes it clinically applicable.

Magnetic impedance tomography.

Tissue can be characterized by its electrical impedance, especially if measurement can be extended over a range of frequencies. Recently, there has been a great deal of interest in imaging the distribution of electrical impedance through the technique of electrical impedance tomography (EIT). However, EIT has a number of practical problems relating to the placement of electrodes on the body. Such contacts are not required to collect magnetic field data around an object through which current is flowing and thus this approach may be more practical than EIT in the clinical environment. This paper describes the technique of magnetic impedance tomography (MIT), which allows reconstruction of the current distribution from magnetic field measurements. The reconstruction techniques used to generate the images and the prototype data collection system are described. Images produced using data collected from discrete and distributed current phantoms and the thorax during human respiration are presented.

A review of image reconstruction techniques for electrical impedance tomography.

There has recently been an increasing interest in the possibility of producing images of electrical impedance within the human body. When an electric current is applied to the body of a voltage distribution is developed across the body surface. This distribution is in part dependent on the internal impedance distribution within the body and it is possible to estimate this distribution from a suitable set of voltage measurements. Because of the nonlinear relationship between the impedance distribution and the voltage distribution at the surface of the body, the reconstruction problem is much more difficult than for other tomographic imaging techniques, but a significant amount of progress has been made, and it is now possible to produce tomographic images of in vivo distributions of impedance, albeit with low spatial resolution. Future developments should improve image quality.

A method of generalized projections (MGP) ghost correction algorithm for interleaved EPI.

Investigations into the method of generalized projections (MGP) as a ghost correction method for interleaved EPI are described. The technique is image-based and does not require additional reference scans. The algorithm was found to be more effective if a priori knowledge was incorporated to reduce the degrees of freedom, by modeling the ghosting as arising from a small number of phase offsets. In simulations with phase variation between consecutive shots for n-interleaved echo planar imaging (EPI), ghost reduction was achieved for n = 2 only. With no phase variation between shots, ghost reduction was obtained with n up to 16. Incorporating a relaxation parameter was found to improve convergence. Dependence of convergence on the region of support was also investigated. A fully automatic version of the method was developed, using results from the simulations. When tested on in vivo 2-, 16-, and 32-interleaved spin-echo EPI data, the method achieved deghosting and image restoration close to that obtained by both reference scan and odd/even filter correction, although some residual artifacts remained.

Registration of low resolution medical images.

There are several situations in which the registration of two medical images is desirable. One example is the registration of two images of the same organ taken using different radionuclide tracers, for example the ventilation and perfusion components of a V/Q lung scan, where the aim is to compare the regional uptake of the two tracers. Another example is the registration of images of an organ belonging to a single patient but taken at different times, where the aim is to follow changes in tracer uptake. Such techniques require a reliable method of registering images. One image is usually brought into registration with another image using a coordinate transfer function (CTF) and the central problem in image registration is the determination of the appropriate CTF. A new semi-automatic approach to the problem of finding the CTF for similar images is described which is especially applicable to low resolution images.

Digital computer processing of brain scans using principal components.

Most of the scan processing methods used to date are unable to distinguish between normal and abnormal features. This paper describes a method of constructing a particular orthogonal transform from a set of normal images, the method of principal components, which may be used selectively to filter out normal features, leaving behind abnormal features. The method has been applied to brain scans and has been implemented on a small digital computer system with magnetic disc.

The use of principal components in the quantitative analysis of gamma camera dynamic studies.

The reduction of the enormous quantity of data in a radionuclide dynamic study to a few diagnostic parameters presents a problem. Conventional methods of data reduction using regions-of-interest or functional images have several defects which potentially limit their usefulness. Using a principal components analysis of the elemental curves representing the change of activity with time in each pixel, followed by a further factor analysis, it is possible to extract the fundamental functional changes of activity which underly the observed variation of activity. An example of this analysis on a dynamic brain scan suggests that the three fundamental phases of activity represent activity in the arterial system, the venous system and diffusion of tracer into the tissues.

A completely automatic method of processing 131I-labelled Rose Bengal dynamic liver studies.

A completely automatic method of measuring Rose Bengal uptake by the liver, expressed in terms of the half-time T1/2, is described. There is no requirement to construct time-activity curves with blood background activity correction as in the conventional regions-of-interest method. All the dixels in the image of a study are used in the actual data analysis. The method is therefore independent of operator influence. The intersection method is offered as an alternative to the manual method. The intersection method uses principal components analysis as a first step in the computation of the intersection of a theory space and a study space. A simple exponential function is used to generate the liver theory space. Sixty 131I Rose Bengal liver function studies were processed by the intersection method. The first forty minutes of patient data were used in the analysis. To validate the new method, computed T1/2 values were compared with those obtained by the manual method. A standard statistical test showed no significant difference between the two methods. Regression analysis gave a value for the coefficient of correlation of 0.89. The intersection method is currently in routine use for the automatic analysis of Rose Bengal liver studies and is faster than the manual method.

Factor analysis of dynamic function studies using a priori physiological information.

The computation of physiological factors and factor images by factor analysis in dynamic structures using the constraints of positive factors and spatial distribution of these factors (FADS), currently used by a number of research workers, is investigated. While the positivity constraints used may be quite acceptable physically, they cannot be strictly said to have direct correlations with the underlying physiological mechanisms in a dynamic study. In principle, FADS estimates the underlying model in the absence of a priori physiological information, and therefore, it is possible that in some situations an incorrect model is extracted. A procedure called IBFADS (information-based factor analysis in dynamic structures) is described which incorporates the IM (intersection method) technique previously developed into FADS, in order to reduce the error in the estimation of the correct model. IM uses a constraint based on physiology of one of the dynamic structures in the model. A computer simulated dynamic phantom study is used to demonstrate IBFADS.

Use of forward projection to correct patient motion during SPECT imaging.

We have developed an iterative method to correct axial and tangential patient motion occurring during tomographic acquisition. The method uses axial images reconstructed from the uncorrected projection images, which are then forward projected to form a basis for registering the original planar images and, in the process, directly seeks to establish a consistent data set. Our method can be applied to all SPECT scans including myocardial and brain SPECT. We demonstrate that the method is capable of detecting and quantitatively correcting for complex motion in both axial and tangential directions. Results from phantom experiments show excellent resolution and contrast recovery after simulated movement in both the axial and tangential directions and initial results with clinical data sets are encouraging.

Towards automatic analysis of dynamic radionuclide studies using principal-components factor analysis.

A method is proposed for automatic analysis of dynamic radionuclide studies using the mathematical technique of principal-components factor analysis. This method is considered as a possible alternative to the conventional manual regions-of-interest method widely used. The method emphasises the importance of introducing a priori information into the analysis about the physiology of at least one of the functional structures in a study. Information is added by using suitable mathematical models to describe the underlying physiological processes. A single physiological factor is extracted representing the particular dynamic structure of interest. Two spaces "study space, S' and "theory space, T' are defined in the formation of the concept of intersection of spaces. A one-dimensional intersection space is computed. An example from a dynamic 99Tcm DTPA kidney study is used to demonstrate the principle inherent in the method proposed. The method requires no correction for the blood background activity, necessary when processing by the manual method. The careful isolation of the kidney by means of region of interest is not required. The method is therefore less prone to operator influence and can be automated.

Automatic registration of SPECT images as an alternative to immobilization in neuroactivation studies.

Subtraction of the two components (baseline and stimulation) of a neuroactivation study using 99Tcm HMPAO SPECT requires accurate registration of the two images. Immobilization of the subject during and between the two components of the study can prove difficult and degrades signal to noise ratio. The use of an automated image registration technique for registering the two components of the test can, even in the case where the subject is removed from the scanner, produce significantly better registration than immobilization.