UK audit of single photon emission computed tomography reconstruction software using software generated phantoms.

The purpose of this study was to undertake an audit of the quantitative characteristics of single photon emission computed tomography software using projection data from an analytically generated software phantom and a measured line source. The phantom consisted of three structures. A uniformly filled cylinder, a series of cylindrical rods of various diameters in a background activity with a rod to background ratio of 2:1 and lastly, a set of three concentric rings of activity in the ratio 1:0:1. The phantom contained no added statistical noise. No attenuation was imposed on the data. The phantom was generated analytically and projections were distributed at six different count densities. A single set of projections from a thin line source was also distributed. These data were distributed to centres throughout the UK. Centres were asked to reconstruct the data using a 'ramp only' reconstruction with no additional smoothing function applied. Data were requested for mean and standard deviation in the uniform cylinder, the maximum counts for each cylindrical rod and the mean counts in regions placed within the concentric rings. For the measured line source, centres were asked to measure the full width at half maximum and peak pixel counts for a profile through the reconstructed line. Results from 115 systems were obtained from 100 centres throughout the UK. These provided data from 12 software providers, 11 of these being commercial companies. Data were compared with the known input values and histograms of the distribution of results obtained. Significant differences in quantitative parameters were noted for the different input count densities as well as between suppliers and revisions of software from single suppliers.

UK audit and analysis of quantitative parameters obtained from gamma camera renography.

The purpose of this study was to perform an audit of quantitative values obtained from gamma camera renography in the UK. Ten patient image sequences representing normal and pathological renal function were obtained from archived studies and distributed to hospitals in the UK. Hospitals were asked to measure five parameters: relative function, renogram time-to-peak (left and right), and whole kidney mean transit time (left and right). Details of methodology, software used and operator experience were requested. This allowed the influence of operational factors on variations in reported values to be examined. A total of 180 responses from 81 hospitals were received. Values reported for the parameters, together with other details supplied, were entered into Excel and SPSS for statistical analysis. Histograms representing the distribution of values were produced for each parameter. The largest variations were found for mean transit time and occasionally for time-to-peak. The effect of factors was assessed using nonparametric statistical tests applied independently to each renogram. For all the parameters, the hospital, UK region, supplier, computer and software version influenced variations in the reported values. Algorithm and site of background region were influencing factors for relative function, the background subtraction method influenced time-to-peak, and curve smoothing influenced mean transit time.

National survey of imaging technologists in nuclear medicine 1998/99.

This survey investigated the nature of imaging staff and the distribution and structure of nuclear medicine departments in the UK. Total numbers of cameras and average per department were lower than in 1989/90 but higher than the 1992/93 survey. Total whole time equivalents, average per department and average per camera had increased. Excluding the Isle of Man, Wales had the lowest population per camera and Northern Ireland the highest. This survey identified 237 departments (90 independent, 116 integrated and 31 satellite), radiographers being the chief technologists in 64% and MTOs in 36%. Over half had one single-headed camera but one third had dual- or triple-headed cameras. Chief technologist grades ranged from Basic to Superintendent II and MTO3 to MTO5. The most common grades for other staff were Senior II or MTO3. Of the 786 technologists who provided details, 68% had initially trained as radiographers. Specific nuclear medicine qualifications were held by 67% of all technologists. In 52% of departments at least one member of staff rotated between nuclear medicine and another department or hospitals and 25% had no full-time staff. Survey returns revealed that, although 84% of imaging technologists often or occasionally attended conferences, 5% never attended.

A control systems approach for the simulation of renal dynamic software phantoms for nuclear medicine.

For renal dynamic studies, the COST B2 hybrid phantom is an example of an artificially created software phantom. Although this phantom is useful, it is not possible to implement the phantom in a self-consistent fashion to produce, for example, a collection of tracer in the bladder which is related to the flow from the kidneys. In this study control systems are used to provide a self-consistent model. A feed-forward control system was designed for the transport of DTPA in the human body using SIMULINK. The system is based on a three-compartment model described by a set of differential equations with flow rates which may be set by the operator. The differential uptake in the kidneys may also be specified, while the flow of tracer through the renal parenchyma and collecting system of each kidney is determined using two-parameter retention functions. Curves corresponding to normal or pathological conditions may be simulated for plasma, parenchyma, collecting system and bladder by appropriate selection of parameters. The system is user-friendly and can be used to simulate almost all conditions seen in patient studies. The next stage of using this information to design dynamic image simulations is in progress.

Intra- and interoperator variations in region-of-interest drawing and their effect on the measurement of glomerular filtration rates.

PURPOSE: Quantitative results are often obtained from images after drawing regions of interest (ROIs) about the organ or area being evaluated. The accuracy and reproducibility of ROIs is an important aspect of quality-control protocols. Attempts to increase ROI accuracy and precision by generating them automatically must be compared with manually processed images to evaluate the success of the automatic methods. Operators' abilities to reproduce ROIs and the effect this has on the reproducibility of estimating glomerular filtration rate from renograms were assessed. METHODS: Estimation of the glomerular filtration rate using Sampson's method requires a) exclusion ROIs around both kidneys for background subtraction, b) whole-kidney ROIs, and c) exclusion ROIs for the collecting system. Two nuclear medicine professionals were asked to produce glomerular filtration rate estimates for 20 patients with diverse renal function. This was repeated 1 month later. The intra- and interoperator variations were calculated for the glomerular filtration rate results and on a pixel basis for the ROIs. RESULTS: THE percentage of common pixels, on average, for a) intraoperator repeats and b) interoperator comparisons were a) 95%, 94%, 85%, and b) 94%, 93%, and 81% for region types 1, 2, and, 3, respectively. Analysis of variance for the glomerular filtration rate estimates showed significant variations in estimates for left kidneys (P < 0.025) but none (P > 0.1) for right kidneys. CONCLUSION: Spatial reproducibility in ROI drawing does not necessarily relate directly to the associated quantitative reproducibility.

Use of significance image to determine patterns of cortical blood flow abnormality in pathological and at-risk groups.

UNLABELLED: The purpose of this work was to determine whether certain pathological groups and other groups at risk for neurological damage exhibited distinctive patterns of regional cerebral blood flow (rCBF) abnormality. METHODS: HMPAO SPECT images obtained from six groups of subjects were compared with a normal cortical rCBF atlas, based on multivariate, voxel-by-voxel methods. In each case, a significance image was outputted, highlighting voxels with deficits of > or =3 s.d. of normal. Abnormal patterns were examined for the six groups, which comprised a further 40 normal volunteers, 18 diver controls, 50 divers with decompression illness (DCI), 34 boxers, 23 schizophrenics and 21 subjects with Alzheimer's disease. RESULTS: The percentages of abnormal cortical voxels for each group were 0.41%, 0.53%, 1.38%, 1.05%, 0.56% and 2.24%, respectively. The percentages of images in each group with at least one lesion of 10 or more connected abnormal voxels and at least 10 lesions of two or more connected voxels, respectively, were 8% and 8% (normal volunteers), 17% and 11% (diver controls), 38% and 38% (divers with DCI), 41% and 29% (boxers), 26% and 13% (schizophrenics) and 90% and 48% (subjects with Alzheimer's disease). This suggests that multiple small lesions are as common as single large lesions for divers with DCI but not for patients with Alzheimer's disease or schizophrenia. Large lesions are located predominantly in the parietal and inferior temporal regions for Alzheimer's disease, in the parietal and occipital regions for divers with DCI and boxers and in the inferior frontal region for schizophrenia. CONCLUSION: It appears that the groups considered here do have different rCBF patterns and that the significance image is a useful way of demonstrating this fact.

UK audit of relative renal function measurement using DMSA scintigraphy.

A variety of software is used to determine quantitative parameters from radionuclide imaging procedures. Knowledge of the variability of parameter values found in different hospitals is an important aspect of clinical audit of these techniques. This study investigated the variation in relative renal function measurement from static DMSA scintigraphy in the UK. Ten studies representing a range of ages and relative function values were distributed in digital form to 100 hospitals with the assistance of the gamma camera computer suppliers and regional audit coordinators. The studies were analysed at each participating hospital and details of the different techniques and computer systems used were documented. The median value of relative percentage function was assessed for each of the studies. Methods varied in terms of the view used for analysis (54% geometric mean, 46% posterior), the type of background subtraction (single region 32%, separate regions 60%, none 8%) and the definition of the renal regions of interest (73% manual, 27% semi-automatic). Eighty-eight percent of results were within two percentage points of the study median and 98% within five percentage points. There were statistically significant differences observed in the results arising from the view used for the analysis and the background subtraction protocol. The results indicate that relative renal function assessment from static DMSA scintigraphy in the UK is essentially a reliable procedure, although improvements could be made by standardizing the technique used.

An assessment of two methods for generating automatic regions of interest.

Two fully automatic methods for generating regions of interest (ROIs) for nuclear medicine images are described and assessed. One of these, involving registration of a previously defined ROI onto a new image, uses spatial information and is appropriate for two- and three-dimensional images which may be static or dynamic. The other method is based on artificial neural networks and uses temporal information. It is appropriate for dynamic images only. The registration method has been tested using 10 pairs of stress and redistribution images obtained from cardiac perfusion SPET. Regions of interest of the left ventricular muscle, defined on the stress images, were registered onto the redistribution images, where they were compared with reproducibility of manually drawn ROIs. Both methods were tested on 17 99Tcm-MAG3 kidney dynamic studies, where the original ROIs corresponding to both kidneys and the bladder were defined using the COST B2 hybrid phantom. Our results indicate that neither method is as reliable as having ROIs redrawn by the operator, although there are indications that an artificial neural network which combines the use of the spatial and temporal information could prove useful for dynamic studies.

A quantitative comparison of some FADS methods in renal dynamic studies using simulated and phantom data.

Simulated and phantom data were used to determine if factor analysis of dynamic structures (FADS) methods can be used quantitatively. FADS methods tested included variants of apex seeking, the intersection method, cluster analysis and spatial constraints. These were compared with a region-of-interest (ROI) approach. Simulated renal studies were prepared using from three to six homogeneous structures. These corresponded to two blood background structures; two structures (one pathological) for parenchyma; and two structures (one pathological) for the collecting system. Time-activity curves for background, parenchyma and collecting system were obtained for each method and compared with the true curve. A kidney phantom was modelled using a tunnelled vessel filled with chelating material A variable flow was controlled by a peristaltic pump representing the renal filtration of fluid. The glomerular filtration rate (GFR) was estimated using FADS and ROI-based methods and compared with the values measured experimentally. Most FADS methods perform well in the absence of pathology, but less well than the ROI-based method when pathology is present. Some FADS methods perform better than the ROI-based method when background estimation is a problem as in the GFR experiment. For quantitative analysis, the success of FADS depends on the validity of the underlying assumptions and on the appropriate nature of the constraints.

Functional bone imaging in the detection of ischemic osteopathies.

The efficiency and accuracy of a functional imaging technique in the detection of ischemic osteopathy is reviewed over a 10-year period. The study includes 161 patients who had trauma, 101 patients who had suspected Perthes disease, 43 patients who had dysbarism, and 22 patients who had renal transplants on steroid immunosuppressive therapy. After intravenous injection of 7.7 MBq kg-1 Tc-99m HDP data were collected on all patients during the first 20 minutes, accretion rate functional images were produced, and static bone scans performed after 3 hours. All patients were followed up until a definitive diagnosis (clinical and radiologic) of avascular necrosis or osteochondritis was confirmed or denied. Of the 327 patients studied, 114 had positive accretion rate images and 213 negative accretion rate images. There were 8 false-positive scan results and 2 false-negative scan results. The overall predictive accuracy was 97% (disease prevalence 33%) with a sensitivity rate of 98% and a specificity rate of 96%. The authors conclude that the technique is a valuable predictive diagnostic indicator for ischemic osteopathies.

CA549 and TPS patterns in the diagnosis and staging of patients with breast carcinoma.

This study compares the clinical value of the breast cancer tumour makers CA549 and TPS, and their tandem use when one or both markers indicate abnormality. For 144 patients presenting with active disease, 33 were classified as Stage I, 37 as Stage II, 40 as Stage III and 34 as Stage IV. For these patients the sensitivity of CA549 using a cut-off of 10 U/ml was 27%, 32%, 42% and 79%, respectively. The sensitivity of TPS for each stage using a cut-off of 100 U/l was 12%, 22%, 28% and 73%, respectively. At these cut-off levels, 36%, 46%, 63% and 91% of patients, respectively, have either CA549 or TPS or both markers raised. For 161 patients with diagnosed benign breast disease, the specificity of marker levels was 96% for CA549, 88% for TPS and 84% for tandem use. CA549 is shown to be superior to TPS and this was confirmed by Receiver Operating Characteristic (ROC) analysis using variable threshold levels, with the areas under the curves for all stages combined being 0.74 +/- 0.03 (ISD) and 0.66 +/- 0.03, respectively. The corresponding area under the curve for tandem use (0.75 +/- 0.03) is marginally greater than for either individual marker, although the difference with respect to CA549 is statistically insignificant.

A voxel-by-voxel multivariate analysis of cerebral perfusion defects in divers with 'bends'.

Past analysis of dysbaric-induced cerebral perfusion defects, demonstrated by 99Tcm-hexamethylpropylene amine oxime (99Tcm-HMPAO) single photon emission tomography in divers using quantitative and/or univariate techniques, has resulted i considerable controversy regarding the significance of these lesions compared to those seen in control subjects, correlations with clinical findings and the role of 99Tcm-HMPAO as a prognostic indicator in decompression sickness. We tried to address these problems by using a multivariate approach to a voxel-by-voxel analysis, involving the use of principal components, to determine ranges of normality in 50 reference controls. In subsequent images, abnormality was defined as 10 spatially connected voxels at an appropriate significance level of three standard deviations. The images of 50 divers with clinically diagnosed 'bends' were compared with those of a further 40 normal population controls with no previous history of loss of consciousness, head injury or dysbarism. The results showed that 19 of 50 divers with 'bends' and 3 of 40 population controls had significant perfusion defects, representing a significant difference between divers with dysbarism and population controls at the level P < 0.002. It is concluded that dysbarism causes significant cerebral cortical perfusion defects in affected divers both in 'silent' and symptomatic (clinically correlated) areas.

Optimization of factors affecting the state of normality of a medical image.

The purpose of this paper is to examine the first stage of the diagnostic process in medical imaging, namely determination of the state of normality, and to attempt to optimize factors contributing to this stage. An image of a given type is defined as abnormal if it does not belong to the appropriate class of normal images. All images must be pre-processed involving image registration and normalization to align and scale the images with respect to each other. Normal ranges may be determined for each voxel (or other appropriate region) from a representative normal sample using univariate analysis, obtaining mean and standard deviation images, or multivariate analysis, which accounts also for normal patterns of variation (represented as principal components). For a new image, the variation from normality (in SDs) for each region may be determined. Since the spatial distribution of this parameter is thought to be relevant, connectivity of abnormal voxels was considered as a possible factor. For the purposes of this study, SPECT images indicating regional cerebral blood flow were used. Images from 50 normal subjects formed the normal sample. A further 40 normal subjects and 200 patients referred with suspected dementia were then analyzed using the normal ranges. ROC analysis, using number of SDs as a variable threshold, was used to optimize the factors. Normalization to global values followed by multivariate analysis using four or five principal components provided optimal discrimination. Connectivity of voxels emerged as an important factor, around 10 connected voxels being optimal for this study.

Computer aided test selection (CATS) for nuclear medicine--a prototype system for renal investigations.

An expert system for renal test selection in nuclear medicine has been developed as the first stage of a collaborative project on test selection in nuclear medicine. The stages of knowledge elicitation and knowledge representation were addressed by means of a questionnaire which was completed by five experts in the field of renal nuclear medicine. A flow chart was developed from the responses and implemented using a commercially available expert system shell (Crystal 4.5). A menu specifying clinical problems, for which renal nuclear medicine is useful, is displayed to the user who is prompted for a choice. Specific aspects of the chosen problem are then shown and again a choice is requested. Selected tests, in order of expert preference, are displayed and further information on any of these is available, if required, on selection from a menu subdivided into categories such as patient preparation, preliminary investigations, etc. The system provides cross-referencing to other areas of investigation and is currently being evaluated using a structured approach commonly employed in the assessment of user interfaces.

Cerebral perfusion and psychometric testing in military amateur boxers and controls.

The objective was to compare two neurophysiological variables in active amateur boxers with non-boxing sportsmen. 41 boxers and 27 controls were given psychometric tests: 34 boxers and 34 controls underwent technetium-99m hexamethylpropyleneamineoxime single photon emission computerised tomography (Tc-99m HMPAO SPECT) cerebral perfusion scans. The controls performed better at most aspects of the psychometric tests. Boxers who had fought fewer bouts had a tendency to perform better at psychometric tests than those boxers who had fought more bouts. Tc-99m HMPAO SPECT cerebral perfusion scanning showed that controls had less aberrations in cerebral perfusion than the boxers. In conclusion, significant differences were shown in two neurophysiological variables between young amateur sportsmen who box and those who do not. The long term effects of these findings remain unknown.

Evaluation of the use of induction in the development of a medical expert system.

In an earlier study, two medical expert systems for diagnosing thyroid disorders, developed by the application of induction on a sample of previously diagnosed cases and on expert-generated rules, diagnosed a set of test cases better than an expert system developed by the more traditional method of collaboration between a knowledge engineer and an expert. In this paper, an alternative measure of the accuracy of diagnosis of each system is used to evaluate the systems. Diagnoses for every distinct case represented by a combination of indicating factors are compared with diagnoses that the expert made. The induced systems provide diagnoses for many more distinct cases, but a much higher proportion of these diagnoses are incorrect. It is argued that generalizing to unseen cases is an inappropriate use of induction algorithms. The systematic development of a decision table is a more appropriate method for devising a medical expert system.

An estimation of noise levels in HMPAO RCBF SPECT images using simulation and phantom data; comparison with results obtained from repeated normal controls.

The purpose of this work was to determine noise levels in HMPAO RCBF SPECT images. Eight simulated images of a uniform sphere of activity were made at each of three different count levels. Three images of the Amersham brain phantom were obtained at each of three count levels, roughly corresponding to the simulation levels. Image reconstruction involved a modified Shepp-Logan filter with and without attenuation correction. The scaling constant in the Budinger equation was shown to vary little over the count range used with a mean value of 23 for uncorrected phantom data and 27 for corrected phantom data, corresponding to RMS noise levels of 7%-15%. The variance due to noise was calculated as a percentage of the variance obtained for 53 normal control studies following image registration and normalization. Values of 54% for uncorrected images and 67% for corrected images were obtained. For 10 normal controls a repeated study was performed. The ratio of within-subject to (single sample) between-subject variance was determined as 73% for uncorrected images and 78% for corrected images.

A method for assessing the significance of abnormalities in HMPO brain SPECT images.

METHODS: A normal atlas for HMPAO rCBF SPECT images was obtained from images of 53 normal controls. Following image registration and normalization, a mean image was extracted, while images representing correlated normal deviants were identified using principal component analysis. These images formed the "building blocks" of the atlas. For subsequent images, the atlas was used to create a "nearest normal equivalent" image, which was compared to a residual standard deviation image to determine the significance of deviations in the new image. RESULTS: Images from 30 patients (10 with Alzheimer's disease; 12 with single or multiple infarcts; and 8 normals) were analyzed. CONCLUSION: Using an optimal decision level, 10/10 patients with Alzheimer's disease and 11/12 patients with infarcts were correctly identified, with only one false-positive resulting. We utilized a database of images obtained from normal controls to create a normal atlas.

Case control study of cerebral perfusion deficits in divers using 99Tcm hexamethylpropylene amine oxime.

In a preliminary report, Adkisson et al. (Lancet 1989; 2:119-121) used 99Tcm-hexamethylpropylene amine oxime (HMPAO) single photon emission computed tomography (SPECT) to provide evidence for cerebral perfusion deficits in 28 cases of dysbarism. The report caused concern because these deficits were found even in cases in which the clinical manifestations were limited to the spinal cord. To address this issue further, a case-control study of cerebral perfusion using 99Tcm-HMPAO SPECT is reported. Four groups of 10 subjects were studied: a) divers scanned on average 11 days after treatment of neurologic decompression illness, b) divers scanned 3-5 yr after treatment for neurologic decompression illness, c) diver controls, and d) population controls. All groups were matched for age, and the divers were further matched for general diving experience. The scans were randomized and reported blind to history. Despite a trend toward larger numbers of deficits in individuals with decompression illness, the 4 groups were statistically indistinguishable. Furthermore, no correlation was found between the location of the perfusion deficits and the clinical presentation. These results indicate that 99Tcm-HMPAO SPECT scanning requires further evaluation before clinical significance can be ascribed to perfusion deficits found in divers.