Search

Segmentation of the lung outline from gamma camera transmission images of the thorax is useful in attenuation correction and quantitative image analysis. This paper describes and compares two threshold-based methods of segmentation. Simulated gamma camera transmission images of test objects were used to produce a knowledge base of the variation of threshold defining the lung outline with image resolution and chest wall thickness. Two segmentation techniques based on global (GT) and context-sensitive (CST) thresholds were developed and evaluated in simulated transmission images of realistic thoraces. The segmented lung volumes were compared to the true values used in the simulation. The mean distances between segmented and true lung surface were calculated. The techniques were also applied to three real human subject transmission images. The lung volumes were estimated and the segmentations were compared visually. The CST segmentation produced significantly superior segmentations than the GT technique in the simulated data. In human subjects, the GT technique underestimated volumes by 13% compared to the CST technique. It missed areas that clearly belonged to the lungs. In conclusion, both techniques segmented the lungs with reasonable accuracy and precision. The CST approach was superior, particularly in real human subject images.

Simulation is useful in the validation of functional image analysis methods, particularly when considering the number of analysis techniques currently available lacking thorough validation. Problems exist with current simulation methods due to long run times or unrealistic results making it problematic to generate complete datasets. A method is presented for simulating known abnormalities within normal brain SPECT images using a measured point spread function (PSF), and incorporating a stereotactic atlas of the brain for anatomical positioning. This allows for the simulation of realistic images through the use of prior information regarding disease progression. SPECT images of cerebral perfusion have been generated consisting of a control database and a group of simulated abnormal subjects that are to be used in a UK audit of analysis methods. The abnormality is defined in the stereotactic space, then transformed to the individual subject space, convolved with a measured PSF and removed from the normal subject image. The dataset was analysed using SPM99 (Wellcome Department of Imaging Neuroscience, University College, London) and the MarsBaR volume of interest (VOI) analysis toolbox. The results were evaluated by comparison with the known ground truth. The analysis showed improvement when using a smoothing kernel equal to system resolution over the slightly larger kernel used routinely. Significant correlation was found between effective volume of a simulated abnormality and the detected size using SPM99. Improvements in VOI analysis sensitivity were found when using the region median over the region mean. The method and dataset provide an efficient methodology for use in the comparison and cross validation of semi-quantitative analysis methods in brain SPECT, and allow the optimization of analysis parameters.

Quantitative indices of radionuclide uptake in an object of interest provide a useful adjunct to qualitative interpretation in the diagnostic application of radionuclide imaging. This note describes a new measure of total uptake of an organ, the specific uptake size index (SUSI). It can either be related in absolute terms to the total activity injected or to the specific activity in a reference region. As it depends on the total activity in the object, the value obtained will not depend on the resolution of the imaging process, as is the case with some other similar quantitative indices. This has been demonstrated in an experiment using simulated images. The application of the index to quantification of dopamine receptor SPECT imaging and parathyroid-thyroid subtraction planar scintigraphy is described. The index is considered to be of potential value in reducing variation in quantitative assessment of uptake in objects with applications in all areas of radionuclide imaging.

Detailed information on biological branching networks (optical nerves, airways or blood vessels) is often required to improve the analysis of 3D medical imaging data. A semi-automated algorithm has been developed to obtain the full 3D topology and dimensions (direction cosine, length, diameter, branching and gravity angles) of branching networks using their CT images. It has been tested using CT images of a simple Perspex branching network and applied to the CT images of a human cast of the airway tree. The morphology and topology of the computer derived network were compared with the manually measured dimensions. Good agreement was found. The airways dimensions also compared well with previous values quoted in literature. This algorithm can provide complete data set analysis much more quickly than manual measurements. Its use is limited by the CT resolution which means that very small branches are not visible. New data are presented on the branching angles of the airway tree.

A technique is described for manually outlining a volume of interest (VOI) in a three-dimensional SPECT dataset. Regions of interest (ROIs) are drawn on three orthogonal maximum intensity projections. Image masks based on these ROIs are backprojected through the image volume and the resultant 3D dataset is segmented to produce the VOI. The technique has been successfully applied in the exclusion of unwanted areas of activity adjacent to the brain when segmenting the organ in SPECT imaging using 99mTc HMPAO. An example of its use for segmentation in tumour imaging is also presented. The technique is of value for applications involving semi-automatic VOI definition in SPECT.

Robust techniques for automatic or semi-automatic segmentation of objects in single photon emission computed tomography (SPECT) are still the subject of development. This paper describes a threshold based method which uses empirical rules derived from analysis of computer simulated images of a large number of objects. The use of simulation allowed the factors affecting the threshold which correctly segmented objects to be investigated systematically. Rules could then be derived from these data to define the threshold in any particular context. The technique operated iteratively and calculated local context sensitive thresholds along radial profiles from the centre of gravity of the object. It was evaluated in a further series of simulated objects and in human studies, and compared to the use of a global fixed threshold. The method was capable of improving accuracy of segmentation and volume assessment compared to the global threshold technique. The improvements were greater for small volumes, shapes with large surface area to volume ratio, variable surrounding activity and non-uniform distributions. The method was applied successfully to simulated objects and human studies and is considered to be a significant advance on global fixed threshold techniques.

Planar images of known, theoretical distributions of radioaerosol in the lung have been simulated using lung models derived from magnetic resonance studies on human subjects. Total lung activity was evaluated from the simulated images together with the absolute penetration index (PI) and a relative value expressed as a fraction of that in a simulated ventilation image. The accuracy and precision of these measurements were calculated by comparison with the true values used in the simulation. Total activity was assessed with systematic errors within 5% and precision within 6.5%. Measured PIs varied only slowly with true PI and inter-model variation masked changes between measurements on the different distributions. The relative PI reduced inter-model variation and provided significant differences between all the distributions. PI was significantly affected by misalignment of the lung region of interest. The conducting airways deposition fraction (CADF) used in the simulation correlated linearly with the fractional activity in a central lung region, allowing CADF to be estimated with a precision of 21%.

Simulation of the gamma camera imaging process has applications in radionuclide imaging in image formation, interpretation and evaluation. Current algorithms are hindered by the length of time taken to perform accurate simulation. In this study an algorithm is developed which uses a number of simplifying approximations to improve speed of operation: restriction of the extent of the primary and scatter point spread functions, coarse sampling of the PSFs in the direction perpendicular to the camera face and use of a circularly symmetric scatter function. The effect of these approximations is assessed qualitatively and quantitatively by comparing simulations with corresponding real measurements. The application of the technique to clinical images using magnetic resonance data is demonstrated. The results show that improvements in speed of about a factor of ten may be achieved at the expense of minimal loss of accuracy. The algorithm should prove useful in speeding up applications of simulation.

Planar scintigrams are deconvolved with a point spread function using the maximum entropy method with the aim of improving image quality. The technique requires the specification of several parameters. These are related to the level of noise present in the data and our a priori knowledge of the object imaged. The performance of the technique is tested for a wide range of these parameters using images of a Williams phantom in scattering material and a figure of merit, derived from the detectability of the smallest cold spot, is calculated. For close to optimal values of the parameters a factor of two improvement in the figure is found. A processed bone image shows improved contrast and resolution. Maximum entropy processing could be used to increase image quality or allow comparable image quality with reduced imaging time or patient dose.

A technique for simulating the point spread function (PSF) of gamma-camera images which treats primary photons and detected scattered photons separately is described. The primary photon PSF was measured at different distances in air from the camera. The scatter PSF was also assessed with different thicknesses of water between the point and the camera, and fitted to a circularly symmetric bi-exponential function. An analysis of the variation of detected scattered radiation based on the results is presented. Simulation of the PSF at a given distance from the camera and thickness of attenuating medium was then possible from the sum of the appropriate primary photon PSF corrected for attenuation and the appropriate scatter PSF. The accuracy of the technique has been studied with variation in depth, density and uniformity of thickness of the attenuating medium.

The renogram, defined as the time-activity curve obtained from measurements with a gamma detector over the kidneys after a prior injection of a radioactive tracer, can be quantified using the deconvolution method. Essentially all the inherent information in the renogram, the estimated relative renal uptake function and transit time spectrum through the kidney, can be derived from the computed renal retention function. This study shows how statistical and physiological noise and different backgrounds affect the accuracy of the derived parameters. Confidence intervals for the estimated relative renal function and mean transit time (MTT) are presented. The principal source of error in relative renal function was due to extrarenal background. It was found that the error in mean transit time due to statistical noise was proportional to MTT, that the presence of extrarenal background strongly affected the accuracy of the MTT, whereas the vascular background in the renogram was of minor importance. Physiological noise, interpreted as periodic changes in transit times does, strictly speaking, invalidate the deconvolution principle, but it was possible to calculate a valid mean value of the different actual transit times. The transit time spectrum, measured by differentiation of the computed retention function, was found to be of no practical value with use of the unconstrained matrix method. The signal to noise ratio and consequently the need for smoothing can be estimated from the sum of the squared second derivatives of the renogram itself. The plateau levels in the renal retention function provide a more reliable estimate for the relative function ratio than the relative amplitude of a renogram in which extrarenal background only has been subtracted.