Java-based remote viewing and processing of nuclear medicine images: toward "the imaging department without walls".

UNLABELLED: In nuclear medicine practice, images often need to be reviewed and reports prepared from locations outside the department, usually in the form of hard copy. Although hard-copy images are simple and portable, they do not offer electronic data search and image manipulation capabilities. On the other hand, picture archiving and communication systems or dedicated workstations cannot be easily deployed at numerous locations. To solve this problem, we propose a Java-based remote viewing station (JaRViS) for the reading and reporting of nuclear medicine images using Internet browser technology. METHODS: JaRViS interfaces to the clinical patient database of a nuclear medicine workstation. All JaRViS software resides on a nuclear medicine department server. The contents of the clinical database can be searched by a browser interface after providing a password. Compressed images with the Java applet and color lookup tables are downloaded on the client side. This paradigm does not require nuclear medicine software to reside on remote computers, which simplifies support and deployment of such a system. To enable versatile reporting of the images, color tables and thresholds can be interactively manipulated and images can be displayed in a variety of layouts. Image filtering, frame grouping (adding frames), and movie display are available. Tomographic mode displays are supported, including gated SPECT. RESULTS: The time to display 14 lung perfusion images in 128 x 128 matrix together with the Java applet and color lookup tables over a V.90 modem is <1 min. SPECT and PET slice reorientation is interactive (<1 s). JaRViS could run on a Windows 95/98/NT or a Macintosh platform with Netscape Communicator or Microsoft Intemet Explorer. The performance of Java code for bilinear interpolation, cine display, and filtering approaches that of a standard imaging workstation. CONCLUSION: It is feasible to set up a remote nuclear medicine viewing station using Java and an Internet or intranet browser. Images can be made easily and cost-effectively available to referring physicians and ambulatory clinics within and outside of the hospital, providing a convenient alternative to film media. We also find this system useful in home reporting of emergency procedures such as lung ventilation-perfusion scans or dynamic studies.

Automated alignment and sizing of myocardial stress and rest scans to three-dimensional normal templates using an image registration algorithm.

UNLABELLED: To optimize the interpretation of myocardial SPECT, we developed an automated method for alignment, sizing and quantification of images using three-dimensional reference templates. METHODS: Stress and rest reference templates were built using a hybrid three-dimensional image registration scheme based on principal-axes and simplex-minimization techniques. Normal patient studies were correlated to a common orientation, position and size. Aligned volumes were added to each other to create amalgamated templates. Separate templates were built for normal stress and rest SPECT 99mTc-sestamibi scans of 23 men and 15 women. The same algorithm was used to correlate abnormal test-patient studies with respective normal templates. The robustness of the fitting algorithm was evaluated by registering data with simulated defects and by repeated registrations after arbitrary misalignment of images. To quantify regional count distribution, 18 three-dimensional segments were outlined on the templates, and counts in the segment were evaluated for all test patients. RESULTS: Our technique provided accurate and reproducible alignment of the images and compensated for varying dimensions of the myocardium by adjusting scaling parameters. The algorithm successfully registered both normal and abnormal studies. The mean registration errors caused by simulated defects were 1.5 mm for position, 1.3 degrees for tilt and 5.3% for sizing (stress images), and 1.4 mm, 2.0 degrees and 3.7% (rest images); these errors were below the limits of visual assessment. CONCLUSION: Automated three-dimensional image fitting to normal templates can be used for reproducible quantification of myocardial SPECT, eliminating operator-dependence of the results.

Optimization of automated quantification of 123I-IBZM uptake in the striatum applied to parkinsonism.

UNLABELLED: Evaluation of therapies for parkinsonism by dopamine receptor SPECT requires a reproducible, optimized quantitation technique. This study presents a new, objective, automated technique for semiquantitative analysis of dopamine receptor density, as applied to the differential diagnosis of parkinsonism. METHODS: Dopamine receptor density measured by 123I-iodobenzamide (IBZM) SPECT was retrospectively analyzed in nonidiopathic parkinsonism (NIPS), in Parkinson's disease (PD), and in healthy volunteers (n = 19, 38, and 13, respectively). A mean template was created from coregistered control studies. Registration errors were assessed using studies with simulated binding deficits. Patient studies were registered to the mean template, and striatal binding was calculated from a corresponding map of 3-dimensional regions of interest (ROIs). The striatal binding ratio and deficits determined by voxelwise comparison with the normal template were investigated and tested with various 3-dimensional ROI sizes and positions. Separation of patient groups was determined by tscore after automatically processing all studies. Results were compared with manual ROI analyses. RESULTS: The automatic method was completely reproducible in 64 of 70 cases. The best diagnostic discriminator was the minimum binding ratio of the 2 striatal nuclei, with the following values: NIPS, 1.33+/-0.13; PD, 1.50+/-0.12; healthy volunteers, 1.49+/-0.08 (+/-SD). The deficit size from voxelwise analysis was: NIPS, 20.5+/-8.2 mL; PD, 9.5+/-8.3; healthy volunteers, 8.9+/-6.0 (+/-SD). The accuracy, measured by receiver operating characteristic areas, was 0.85+/-0.05, 0.77+/-0.06, and 0.80+/-0.06 (+/-SE) for the optimal predictor (automated) and 2 blinded observers (manual), respectively. CONCLUSION: A new 3-dimensional, automated technique has been developed to semiquantitate receptor density that dramatically improves reproducibility. The optimal diagnostic discriminator of parkinsonism determined by the automatic technique has good accuracy compared with the manual technique.

Automatic three-dimensional multimodality registration using radionuclide transmission CT attenuation maps: a phantom study.

UNLABELLED: Coregistration of images from a single subject, acquired by different modalities, is important in clinical diagnosis, surgery and therapy planning. The purpose of this study was to evaluate, using a physical torso phantom, a novel, fully automated method for three-dimensional image registration of CT and SPECT, using radionuclide transmission (RNT) attenuation maps. METHODS: We obtained CT scans and SPECT scans paired with RNT maps of an anthropomorphic cardiac phantom. RNT attenuation maps were acquired using an uncollimated 99mTc-filled flood source. RNT and SPECT scans were acquired in the same spatial orientation (usual clinical practice in nonuniform attenuation correction). In addition, CT attenuation maps (CTMAPs) for 99mTc SPECT were generated from CT by linear energy scaling. RNT maps were registered to CT and CTMAPs by iterative simplex minimization of count difference and uniformity index (sum of RNT map intensity variances corresponding to each intensity level in the CT volume). In each iteration, three shifts and three angles were adjusted. To register SPECT to CT, we applied the RNT transformation parameters to SPECT. RESULTS: RNT maps could be registered to CT and CTMAP images using both criteria. The average three-dimensional distance between landmark and automated registration was 2.5 +/- 1.2 mm for count difference and 3.3 +/- 1.3 mm for uniformity index. The three-dimensional reproducibility errors were 1.2 +/- 0.7 mm for count difference, 2.1 +/- 0.5 mm for uniformity index and 2.3 +/- 1.0 mm for manual marker registration. The minimization of uniformity index was robust when up to 50% CT or RNT slices were missing and was not affected significantly (<2 mm) by realistic variation in CT values (+/- 12 Hounsfield units). CONCLUSION: In addition to typical use in nonuniform attenuation correction, RNT maps can be used for fully automated three-dimensional registration of SPECT to CT. Such registration is not affected by features and quality of SPECT images and avoids difficulties associated with fiducial markers. Our method can be applied to SPECT-CT registration of various organs, such as brain, heart, lungs, breasts and abdomen, including oncological scans.

Three-dimensional demarcation of perfusion zones corresponding to specific coronary arteries: application for automated interpretation of myocardial SPECT.

UNLABELLED: In this study, three-dimensional maps of specific coronary artery territories were derived and combined with normal distribution maps as a reference for automated characterization of defects, including location and size. METHODS: One hundred sixty-eight 99mTc-sestamibi myocardial perfusion SPECT scans from normal patients and patients with single-vessel disease were selected according to angiographic data. Five separate groups were established for men and women: normal, proximal left anterior descending (PLAD), distal left anterior descending (DLAD), right coronary artery (RCA) and left circumflex (LCx). All myocardial perfusion studies were aligned and sized to the same three-dimensional orientation using a previously developed automated image registration technique. Mean and variation three-dimensional templates were constructed from stress images in each group. Normal templates were demarcated with hypoperfusion regions obtained from disease templates. The defects were detected in the individual patient's images by a region-growing algorithm which identified abnormal voxels by comparison to the corresponding voxels in the mean and variation templates. RESULTS: Defects were quantified with respect to volume, location relative to the expected hypoperfusion zones and severity index. Abnormal regions could be marked directly on tomographic slices and visualized in various orientations. Single defects greater than 2% of the myocardium positioned within demarcated perfusion territories were detected in 105/119 abnormal patients and in 3/49 normal patients. CONCLUSION: Maps of myocardial perfusion zones created from images of angiographically selected patients provide a reference for automated localization of myocardial perfusion defects. A template-based region-growing is a robust technique for volumetric quantification and localization of abnormal regions.

Pulmonary uptake of technetium-99m-sestamibi induced by dipyridamole-based stress or exercise.

UNLABELLED: On poststress images with 99mTc-sestamibi (MIBI), increased lung uptake of the radiotracer may reflect severe or multivessel coronary artery disease. METHODS: We measured pulmonary/myocardial ratios of MIBI at standardized times on immediate poststress acquisitions and on delayed tomographic acquisitions. In 1500 sequential patients referred for rest and stress myocardial tomography, ancillary planar images were obtained 4 min postinjection at peak stress with exercise, either alone (exercise, n = 674), or after intravenous dipyridamole (dipyridamole, n = 826). RESULTS: Based on 95% confidence limits in the angiographic normals, high values for immediate acquisitions were found in 17% of dipyridamole studies and 15% of exercise studies. High values for delayed acquisitions were found in 10% of dipyridamole studies and 9% of exercise studies. For both stress modes, increased values were related (p < 0.001) to ischemic perfusion defects for immediate images, to fixed defects for delayed images, and to ventricular dilation in both cases. By logistic regression analysis, body weight and history of infarction were also minor independent determinants (p < 0.01) of delayed acquisitions. In a subset of 250 cases with angiographic correlation (163 with dipyridamole; 87 with exercise), immediate lung uptake was highly correlated with ventricular dysfunction and with coronary stenoses (p < 0.0001). Relationships were similar to those in a historic control series imaged with 201TI. Values for delayed poststress images, and for corresponding rest images, showed strong relationships to ventricular dysfunction but not to stenosis severity. CONCLUSION: The relationships of immediate lung uptake to scintigraphic and angiographic disease patterns suggest its possible diagnostic use as an indicator of stress-induced ventricular decompensation.

The cortical representation of somatosensory evoked potentials of the phrenic nerve.

Respiratory electrophysiological studies are of essential value in diagnosing and managing patients with respiratory failure, but assessment of the sensory phrenic nerve fibres has been neglected. We recorded phrenic nerve somatosensory evoked potentials (SSEPs) by combining neurophysiological and neuroimaging techniques in three healthy subjects. Evoked potentials of the phrenic nerve showed the highest amplitude at CP3, determined by the modified 10-20 EEG system, and occurred at a constant latency, PI at 12.0 +/- 0.6 ms, and NI at 17.3 +/- 0.8 ms. Single photon emission computer tomography (SPECT) performed during phrenic nerve stimulation revealed focal neuronal activation in the somatosensory pathways. Intravenously administered Tc-99m Ethyl Cysteinate Dimer (ECD) was used as a blood flow tracer to obtain baseline and activated images. After image registration, baseline images were compared voxel-by-voxel with the activation images. The mean inter-subject summation image of the activated state was compared with that of the baseline state using ten normal subjects. The extent of the total voxel volume increase on the mean images of the 3 activated SPECT images was 0.7%, and a mean signal increase of 22%. For further anatomic localization of regional increases in signal, the magnetic resonance image (MRI) scan of each subject was registered and superimposed on the activated stage SPECT image. This method may be used clinically to study the pathophysiology of impaired central respiratory drive.

Variability and accuracy of measurements of prostate brachytherapy seed position in vitro using three-dimensional ultrasound: an intra- and inter-observer study.

This paper is a step in investigating whether three-dimensional (3D) ultrasound can be used intraoperatively to replace Computed Tomography (CT) for localization of brachytherapy seeds. In order to quantify the accuracy and variability of seed localization without introducing effects due to tissues, we first report our results with test phantoms. An inter- and intra-observer study was performed to assess the variability of 2 3D ultrasound scan acquisition methods: Tilt 3D scanning and pull-back 3D scanning. Seven observers measured the positions of gold seed markers in an agar phantom twice in each of the three orthogonal image planes. An analysis of variance (ANOVA) was performed to determine the intra- and inter-observer standard errors of measurement (SEM) and the minimum detectable changes in marker position (deltap). Average intra- and inter-observer SEMs for the tilt scan 3D image were 0.36 and 0.40 mm, respectively. Measurements of the pull-back scan 3D image yielded average intra- and inter-observer SEM of 0.46 and 0.49 mm, respectively. A paired difference analysis showed that the lower SEM for the tilt 3D scan image were statistically significant at a significance level of alpha= 0.05. The accuracy of the US measurements was tested by determining marker coordinates from CT images of the phantom in a stereotactic head frame. CT coordinates were matched to the ultrasound (US) coordinates by means of an affine transform. Average matching errors in x, y, and z were 0.02, 0.10, and -0.02 mm, respectively.

Evaluation of linear registration algorithms for brain SPECT and the errors due to hypoperfusion lesions.

The semiquantitative analysis of perfusion single-photon emission computed tomography (SPECT) images requires a reproducible, objective method. Automated spatial standardization (registration) of images is a prerequisite to this goal. A source of registration error is the presence of hypoperfusion defects, which was evaluated in this study with simulated lesions. The brain perfusion images measured by 99mTc-HMPAO SPECT from 21 patients with probable Alzheimer's disease and 35 control subjects were retrospectively analyzed. An automatic segmentation method was developed to remove external activity. Three registration methods, robust least squares, normalized mutual information (NMI), and count difference were implemented and the effects of simulated defects were compared. The tested registration methods required segmentation of the cerebrum from external activity, and the automatic and manual methods differed by a three-dimensional displacement of 1.4+/-1.1 mm. NMI registration proved to be least adversely effected by simulated defects with 3 mm average displacement caused by severe defects. The error in quantifying the patient-template parietal ratio due to misregistration was 2.0% for large defects (70% hypoperfusion) and 0.5% for smaller defects (85% hypoperfusion).

Evaluation of voxel-based registration of 3-D power Doppler ultrasound and 3-D magnetic resonance angiographic images of carotid arteries.

Spatial registration and fusion of ultrasound (US) images with other modalities may aid clinical interpretation. We implemented and evaluated on patient data an automated retrospective registration of magnetic resonance angiography (MRA) carotid bifurcation images with 3-D power Doppler ultrasound (PD US) and indirectly with 3-D B-mode US. Volumes were initially thresholded to reduce the uncorrelated noise signals. The registration algorithm subsequently maximized the mutual information measure between the PD US and 3-D MRA via iterative simplex search to find best "rigid body" transformation. We rated the performance of the algorithm visually on (n = 5) clinical MRA and 3-D PD US datasets. We also evaluated quantitatively the effect of thresholding, initial misalignment of the paired volumes and the reproducibility registration. We investigated the effect of image artefacts by simulation experiments. Preregistration misalignments of up to 5 mm in the transaxial plane, up to 10 mm along the axis of the carotids and up to 40 degrees resulted in 107 of 110 successful registrations, with translational and rotational errors of 0.32 mm +/- 0.3 mm and 1.6 +/- 2.1 degrees. The algorithm was not affected by missing arterial segments of up to 8 mm in length. The average registration time was 4 min. We conclude that the algorithm could be applied to 3-D US PD and MRA data for automated multimodality registration of carotid vessels without the use of fiducials.

Comparing the image quality of myocardial perfusion agents in the clinical laboratory: small test groups and large reference populations.

Myocardial perfusion studies have been performed for 6 years using technetium-99m (99Tcm)-sestamibi (MIBI). In this study we evaluated a newer agent, 99Tcm-tetrofosmin (TF), on a trial basis for 2 weeks, and compared the results from each week to those in an adjacent week of MIBI use. The routine protocol included weight-based dosing, frequent dipyridamole use, and separate-day rest and stress wherever possible. During the first week, TF was used with 'usual' image timing, i.e. stress tomography performed 30-60 min after stress, and ancillary immediate images performed 4 min after injection. For the second week, 'early' tomography was performed 15-30 min after stress. TF scans (n = 53) were compared with MIBI scans for the adjacent weeks (n = 54) and with a historical reference series (n = 1800). Blinded analysis was made of tomographic image quality, peak myocardial counts and background activity (lung and abdomen) on immediate and delayed acquisitions and on tomographic reconstructions. The TF and MIBI test groups were similar with respect to gender, weight, stress protocol, tracer doses, imaging times and scintigraphic findings. Using analysis of variance, the tomographic quality scores were similar for the two observers, with stress>rest (P<0.0001), 'usual'>'early' (P<0.001) and MIBI>TF (P<0.05). Myocardial counts were approximately 20% higher with the MIBI test group than with TF at all times after stress (P=0.001), and were similar to the reference population. MIBI with usual timing gave more favourable stress abdominal background ratios than the other three agent/timing combinations. Satisfactory images could be obtained with TF, but no apparent advantage over MIBI could be attained with earlier post-stress imaging. Subtle advantages for MIBI over TF were suggested by comparison of the small test groups. In our local imaging context, these conclusions were reinforced by a large control series.

Comparison of ordered subsets expectation maximization and Chang's attenuation correction method in quantitative cardiac SPET: a phantom study.

Photon attenuation is one of the primary causes of artifacts in cardiac single photon emission tomography (SPET). Several attenuation correction algorithms have been proposed. The aim of this study was to compare the effect of using the ordered subsets expectation maximization (OSEM) reconstruction algorithm and Chang's non-uniform attenuation correction method on quantitative cardiac SPET. We performed SPET scans of an anthropomorphic phantom simulating normal and abnormal myocardial studies. Attenuation maps of the phantom were obtained from computed tomographic images. The SPET projection data were corrected for attenuation using OSEM reconstruction, as well as Chang's method. For each defect scan and attenuation correction method, we calculated three quantitative parameters: average radial maximum (ARM) ratio of the defect-to-normal area, maximum defect contrast (MDC) and defect volume, using automated three-dimensional quantitation. The differences between the two methods were less than 4% for defect-to-normal ARM ratio, 19% for MDC and 13% for defect volume. These differences are within the range of estimated statistical variation of SPET. The calculation times of the two methods were comparable. For all SPET studies, OSEM attenuation correction gave a more correct activity distribution, with respect to both the homogeneity of the radiotracer and the shape of the cardiac insert. The difference in uniformity between OSEM and Chang's method was quantified by segmental analysis and found to be less than 8% for the normal study. In conclusion, OSEM and Chang's attenuation correction are quantitatively equivalent, with comparable calculation times. OSEM reconstruction gives a more correct activity distribution and is therefore preferred.

Automated three-dimensional quantification of myocardial perfusion and brain SPECT.

To allow automated and objective reading of nuclear medicine tomography, we have developed a set of tools for clinical analysis of myocardial perfusion tomography (PERFIT) and Brain SPECT/PET (BRASS). We exploit algorithms for image registration and use three-dimensional (3D) "normal models" for individual patient comparisons to composite datasets on a "voxel-by-voxel basis" in order to automatically determine the statistically significant abnormalities. A multistage, 3D iterative inter-subject registration of patient images to normal templates is applied, including automated masking of the external activity before final fit. In separate projects, the software has been applied to the analysis of myocardial perfusion SPECT, as well as brain SPECT and PET data. Automatic reading was consistent with visual analysis; it can be applied to the whole spectrum of clinical images, and aid physicians in the daily interpretation of tomographic nuclear medicine images.

Investigation of measures to reduce interfering abdominal activity on rest myocardial images with Tc-99m sestamibi.

With the new myocardial agent Tc-99m sestamibi, relatively high abdominal uptake represents a major limiting factor. The effect of a standard feeding (commercial milkshake taken immediately after injection), and posture (standing versus sitting for 10 minutes postinjection) on the resting biodistribution of sestamibi was investigated in patients receiving 3 to 5 MBq/kg injections as part of rest-stress tomography. Ancillary anterior 1-minute images of heart and abdomen were obtained at 15 minutes postinjection and 90 to 140 minutes postinjection in 32 patients, randomized to feeding and postural treatments. Feeding decreased the activity in the gallbladder at both 15 and 110 minutes, but had no effect on liver parenchyma; activity in a background rectangle immediately beneath the heart was decreased by feeding only on the 15-minute images. An effect of posture was not apparent. Further study of the acute effects of a milkshake in eight patients showed a prompt reduction of 26% in a subdiaphragmatic background rectangle, but a more gradual decline in gallbladder counts. Thus, feeding of lipid after injection is not an essential component of sestamibi imaging protocols; oral administration of fluid immediately before imaging may help reduce interfering gastric activity.

Brain mapping of median nerve somatosensory evoked potentials with combined 99mTc-ECD single-photon emission tomography and magnetic resonance imaging.

Single-photon emission tomography (SPET) was performed during electrical median nerve stimulation and used to detect focal neuronal activation in the somatosensory pathways. Intravenously administered technetium-99m ethyl cysteinate dimer (ECD) was used as a blood flow tracer to obtain baseline and activated images in each of three subjects. After image registration, baseline images were compared voxel by voxel with the activation images. In addition, the mean summation of the activated-state images of the subjects was compared with the mean summation of the baseline-state images of ten normal subjects. Discrete brain regions occupying 0.9%-1.6% of total brain volume showed an increase in signal from 33.6% to 35.0%. For further anatomical localization of regional increases in signal, the MRI scan of each subject was registered and superimposed on the activated-state SPET image. This method may be used to localize lesions in various disorders of the central nervous system.