- Email: [email protected]
Image Fusion System Using PACS for MRI, CT, and PET Images
PII S1095-0397(99)00018-7
Clinical Positron Imaging Vol. 2, No. 3, 137–143. 1999 Copyright 1999 Elsevier Science Inc. Printed in the USA. All rights reserved. 1095-0397/99 $–see front matter
ORIGINAL ARTICLE
Image Fusion System Using PACS for MRI, CT, and PET Images Saleh Alyafei, BHSc1, Tomio Inoue, MD1, Hong Zhang, MD1, Khalil Ahmed, MD1, Noboru Oriuchi, MD1, Noriko Sato, MD1, Hideki Suzuki, MD2, Keigo Endo, MD1 1
Department of Nuclear Medicine and Diagnostic Radiology, Gunma University School of Medicine. Showa machi 3-39-22, Maebashi, Gunma 371-8511, Japan; 2 Department of Medical Information, Gunma Prefectural College of Health Sciences, Gunma, Japan This paper presents and evaluates Picture Archiving and Communication System (PACS) and its interaction with the image fusion applications using positron emission tomography (PET), CT, and MRI, as well as some clinical applications of fusion images. A network connections between medium-sized PACS involving CT, MRI, SPECT, and PET were developed. Image registration and fusion was achieved in the PET’s workstation by Advanced Visual System (AVS) software. Reconstructed datasets of CT, MRI, SPECT, and PET were transferred and archived in PACS servers. Series of anatomical images of CT and MRI were fused with metabolical images of PET with 18F labeled fluoro-2-deoxy-D-glucose (FDG). Throughput rate of image data, as well as clinical applications of fusion images were evaluated and correlated with phantom studies. The average throughput rate of archiving and processing steps was 0.45 Mbps, 0.77 Mbps, respectively. Image fusion experiment using phantom and patients’ data showed high accuracy in all directions. The combination of PACS and image fusion provided very useful clinical tools and made it quite easy to maximize the benefit from the diagnostic imaging. (Clin Pos Imag 1999;2:137–143) 1999 Elsevier Science Inc. All rights reserved. Key Words: Picture Archiving and Communication System; PACS; Image Fusion.
Introduction
P
icture Archiving and Communication Systems (PACS) is becoming an integral part of most modern medical imaging departments. Tremendous volume of images accumulates in PACS database from different modalities such as CT, MRI, SPECT, and PET.1 PACS is the infrastructure supporting advances in many areas, such as in 1) hospital operations, it gives faster and more widely available access to image and reports offer the potential for savings in clinician time and length of stay; 2) radiology operations, it provides timely access to the images to improve the patient flow, throughput and image reporting; 3) case management, it can provide image data for systems streamlining the presentation of case review conferences; 4) teaching files, both educational and clinical goals are enhanced by convenient access to teaching cases during reading sessions. An on-line teaching file
Address correspondence to: Saleh Alyafei, Department of Nuclear Medicine, Gunma University School of Medicine, Showa machi 3-3922, Maebashi, Gunma 371-8511, Japan.
could provide access to relevant reference images for comparison during reading sessions; and 5) PACS promote and incorporate research results of multimodality fusion, into clinical use. PACS is the foundation resource of image registration and fusion applications. Multimodality image fusion application requires standardized parameters to be clinically useful. These parameters should cover interface of digital image transmission, filing and retrieval, image compression, and multi-image database. In contrast to these facts, the Digital Imaging and Communications in Medicine (DICOM) V3.0 was developed to aid in the acquisition and transport of images via PACS.2–3 Also, PACS is eliminating the need of multiple single-connections between workstation of image fusion and each workstation of CT, MRI, nuclear medicine, and PET. This paper describes and evaluates image fusion method using the advantage of a DICOM based PACS. Also, it presents some clinical applications of registration and image fusion of multimodality such as CT, MRI, and PET. 137
138 Clinical Positron Imaging, Volume 2, Number 3
Figure 1. Current PACS configuration, the thin lines represent the Ethernet network (10 Mbps), and the thick lines represent the ATM fiber connections (155 Mbps).
Materials and Methods Background and Configuration of PACS and Fusion System We adopted a hospital information system (HIS) in 1988 followed by a reporting system (RS) in 1991. One year later a medium-sized PACS initially installed, was linked by fiber distributed data interface (FDDI, 100 Mbps), optical loop local area network (LAN). Three CT scanners (TCT-900S, X-Force, Toshiba, Tokyo), two computed radiography (CR) (TCR, Toshiba), two MRIs (MAGNETOM-Simens, Signa-GE), two gamma cameras (PRISM 2000, PRISM 3000), and one whole body PET scanner (Shimadzu SET2400W, Kyoto, Japan) were connected to PACS. In 1997 PACS was updated and remodeled with a new main PACS server computer (UNIX workstation, EWS4800/320px, 128 MB RAM, 250 MHz, 14 GB HD, and 78 GB MO). Three PACS DICOM servers (UNIX workstation, EWS4800/ 460px, 128 MB, 250 MHz, 7 GB HD, and 78 GB MO), five gateway computers, and eight display terminal computers (PC pentium 133 MHz, 80 MB) were components of the new PACS. Images of all modalities were converted to DICOM 3.0 formats. Transmission control protocol/internet protocol (TCP/IP) traffic was used because of its support in the DICOM 3.0 specification. A new network of asynchronous transfer mode (ATM) was established with direct connection to the PACS controller computer and via Ethernet to the gateway computers and image display terminals (Figure 1).
The necessary on-line storage capacity based on the yearly statistics was computed. The approximate annual volume data production of two CT scanners, two MRI units, two SPECT gamma cameras, and one PET scanner have been included (Figure 2). Image compression of 1.5–2 factor has been applied. The compressed images are retrieved from the archive to the requester of the images. The latter would locally decompress the images to the original resolution, as such compression method would increase the throughput rate of the data and increase the storage capacity of PACS archive. The highest amount of storage was needed for the CT. The SPECT and PET images required the least amounts of storage in the PACS servers. One slice of CT image consisted of 512 * 512 pixels requires 524 KBytes. MRI’s slice consisted of 256 * 256 pixels requires 131 KBytes, PET’s slice consisted of 128 * 128 pixels requires 33 KBytes, and SPECT’s slice consisted of 64 * 64 pixels requires 8.2 KBytes. AVS (Advanced Visual Systems, Inc., Waltham, MA) software was installed in PET workstation, SET2400W (Digital Unix Ver. 3.2G, Compac Alpha Station 600, 5/333, 128 MB, 333 MHz) to determine the clinical potential of registration and image fusion. Patients underwent series of diagnostic examinations using CT, MRI, SPECT, or PET. The reconstructed images were transferred to PACS, then to SET2400W workstation. A method of three-dimensional registrations of CT, MRI, and PET datasets has been carried out and assessed.4–6 The current generation of PET scanners provides functional data with a spatial resolution of 3–6 mm. A
Image Fusion System Using PACS/Alyafei et al. 139
Figure 2. The approximate annual volume data. Two CT scanners, 2 MRI units, 2 SPECT gamma cameras, and 1 PET scanner were included.
high target-to-background ratio of smaller lesions less than 1 cm in diameter is often observed,7–8 because of the high spatial resolution of PET and the high tumor affinity of FDG in these cases, the identification of lesion localization based on only FDG PET is difficult. Fusion of metabolic FDG images of PET to the anatomical information available from MRI or CT scans with higher spatial resolution seem to be desirable.9
These datasets were transferred from CT, MRI, SPECT, and PET to PACS server (archiving step), as well as from PACS sever to the AVS workstation, (processing step). The average throughput measurements of each step were recorded, by copying a static buffer across the network. The system time is recorded before the actual transfer and once again after the transfer is completed.
Throughput Rate Measurement
Phantom Experiments
CT, MRI, and SPECT configuration consisted of one ATM switch (AS 155), one PACS server, with a direct ATM connection to the switch using the user network interface (UNI, Ver. 3.1). All the ATM interfaces operate at rate of 155 Mbps. The workstation of PET was connected to the PACS DICOM server via Ethernet line. Three modality input terminals (MITs), connected by Ethernet to the router, that has both Ethernet and ATM connections (AR 155, 3COM) (Figure 1). We have performed several transfer tests of patients’ data and phantom’s data. These tests were performed with a focus on a clinical application. Datasets of CT images of 3 patients, (total of 42 MB), MRI images of 2 patients (23 MB), nuclear medicine images of 2 patients (6.7 MB), and datasets of PET images of 2 patients (6 MB), as well as large volumes of phantom’s data of each modality were involved in the transferred experiment.
The basic experiments using sphere phantoms were performed to assess the accuracy of volume measurements by CT, MRI, SPECT, and PET, and also to evaluate the validity of image fusion and registration methods. Spheres of 5, 7, 13, 18, 25, and 40 mm in diameter were filled with a solution of uniform activity of 18F for the PET study, 99mTcO42 for the SPECT study. They were filled with air for the evaluation of CT and MRI. The spheres were positioned in a circle of z5 cm radius and placed in a 20-cm diameter phantom, with uniform distribution of attenuating medium such as water. All slices were reconstructed with a 10-mm thickness. A region of interest (ROI) on each sphere boundary was drawn on the transaxial images of CT, MRI, SPECT, and PET. The number of pixels inside each boundary was then computed. Each sphere’s area was calculated in mm2. The sphere’s volumes were determined by
140 Clinical Positron Imaging, Volume 2, Number 3
multiplying each sphere’s area with the slice thickness. The degree of misregistration of the spheres was evaluated by calculating the distance between the center of gravity of each spheres on PET image with the corresponding spheres on the CT, and MR image.
dard deviations of the PET voxels within the partitions defined by a segmentation of the CT/MRI intensities into 256 levels.
Clinical Application
Throughput Rate Measurements
Two patients with malignant tumors underwent CT to obtain anatomical information about the location and boundary of the lesions. In addition, each patient received PET scan, following the injection of FDG to assess the metabolic activity in the lesions. The patient was in supine position and the abdominal region was imaged with 3 mm contiguous transverse slices. Scans were performed before and following 150 cc injection of contrast medium to delineate tumor boundaries from normal tissue (scan parameters of 120 kVp and 200 mA, 512 * 512 matrix). The images were saved in the processing workstation of CT. The whole body PET scanner (Shimadzu, Kyoto, Japan), that provides 63 imaging slices with an axial field of view of 200 mm, was used for FDG PET study. A transmission scan for attenuation correction was conducted with a rotating 67 Ge-germanium rod source (z350–400 million counts). Patients were then injected with 185 Mbq of FDG. Following 40 minute incorporation period, an 8 minute emission scan was acquired. Transverse section images were reconstructed with a conventional filtered back-projection algorithm to yield a final in-plane resolution of z4.3 mm (full width at half maximum, FWHM) and an axial resolution of z4.3 mm FWHM. All cross-section image-sets were reconstructed initially on each acquisition system using standard vendorsupplied reconstruction algorithm. Image fusions were performed with PET workstation, running Digital UNIX and the native MIT X-windows graphical user interface integrated with AVS. Reconstructed CT, and MRI datasets were transferred from PACS server to the workstation of PET using TCP/IP. PET datasets were constructed in the workstation of PET, then converted to DICOM 3 format. An AVS module was developed to reconstruct the incoming transaxial slices into 3-dimensional slices, transverse, coronal, and sagittal. These three views were simultaneously used in assisting the fusion images process. The algorithm used for image registration was the intensity value matching method developed by Wood et al.5 It is based on the concept that two image volumes are in registration when the variance of the pixel-by-pixel ratio is a minimum. This assumes the spatial intensity patterns are similar for the two volumes, that is well justified for intermodality registration. For PET and CT/MRI registration the assumption is only approximately true. The concept of minimum variance of the pixel-by-pixel ratio was extended to minimize a weighted sum of stan-
The average throughput rates of CT, MRI, and SPECT datasets from imaging modality to main PACS server (archiving) were 0.58 Mbps, 0.43 Mbps, and 0.33 Mbps, respectively (Fig. 3A). The average throughput rate of PET datasets to the DICOM PACS server was 3.75 Mbps. The average throughput rates to the PET workstation (processing) from the main PACS server were 1.2 Mbps, 0.49 Mbps, and 0.62 Mbps for CT, MRI, and SPECT data, respectively (Figure 3B).
Results
Phantom Experiments The volume of spheres determined by CT, MRI, SPECT, and PET was compared to the actual volume. Regarding the maximum volume of 126 ml, the measured volumes by CT, MRI, SPECT, and PET were 129 ml, 130 ml, 179 ml, and 146 ml, respectively. Regarding the minimum volume of 2 ml, the volumes determined by CT, MRI, SPECT, and PET were 3 ml, 3 ml, 0 ml, and 15 ml, respectively. Regarding the sphere with a 20-mm radius, the overestimation of the measurement of radius by CT, MRI, SPECT, and PET compared to the radius of the actual sphere were 3%, 3%, 43%, and 36%, respectively. A transaxial fused image showed an excellent registration of PET and MRI images in X-and-Y directions (Figure 4). The degree of misregistration was relatively small always (,4 mm).
Clinical Application A fusion image of a 60-year-old male patient with prostate cancer was obtained. Transverse CT slice through the pelvic area revealed lymphnode swelling, and bone metastasis in sacrum and iliac bone (Figure 5A). The FDG-PET image was registered using AVS (Figure 5B), and sliced to match the CT image. PET/CT fused 2-dimensional images were produced (Figure 5C). With the PET scan alone, the tumors were readily apparent but it was difficult to decide the location and the nature of these tumors due to the lack of anatomical information on the PET scan. By the fusion images of PET and CT, we could identify clearly the anatomical location of abnormal FDG uptake. Another patient is an 81-years-old male with lung cancer, and left pleural thickness after the irradiation therapy. Because tumor marker (CEA) was increasing, recurrent lung cancer was suspected. On the CT image (Figures 6A and 6D), there is an apical pleural thickness of the left lung, suggesting cancer involvement or
Image Fusion System Using PACS/Alyafei et al. 141
Figure 3. (A) The average throughput rates in Mbps of CT, MRI, SPECT, and PET datasets from imaging modality to PACS servers (archiving). (B) The average throughput rates in Mbps from the main PACS server to the PET workstation (processing).
chronic inflammation. The corresponding PET scan (Figures 6B and 6E) revealed high uptake around the left pleural cavity with a hot spot in the median portion. The fused image demonstrating good corresponding between the area of cancer infiltration on the PET image and the area of pleural thickness on the CT image (Figures 6C and 6F).
Discussion PACS has tremendous benefits and values outside of diagnostic department as well as internally. The biggest benefit derived from an ideal PACS is to break the physical as well as time barrier for information exchange. PACS is playing a vital role by contributing the fundamental requirements to achieve clinical setup for image fusion in our institute. Most communications providers refer to bandwidth when talking about data. Bandwidth refers to the amount of information that can be passed through a given connection, usually measured in Kilobits/second (Kbps) or Megabits/second (Mbps). The larger the
bandwidth, the more data can be passed through in a measured second. In our throughput rate experiment, the highest throughput rate was between PET and DICOM PACS server, almost 50% faster than other network connections. This is probably due to the fact that PET workstation and DICOM PACS server have a singular connection, compared with the multiple connections of CT, MRI, and SPECT with the main PACS server. The nature of the multiple connections of the main PACS server, causes high data traffic, that slows the throughput rate. Another reason could be that the main PACS server is using the Ethernet branches to connect MIT computers with ATM router. A bottleneck may occurs within these connections, and collisions of image data is highly happening.10 By measuring the throughput rate, the biggest drawback to the image transfer is not the network but workstations. Even Ethernet cable can operate at 10 Mbps and theoretically networks have unlimited potential transfer speed if they are constructed of fiber. Workstations are limited by the operating system and hardware utilized for internal communications as well
Figure 4. (A) Transaxial image of a phantom acquired by MRI. (B) A corresponding co-registered transaxial image of the same phantom acquired by PET scanner. (C) A transaxial fused image slice showing excellent registration of PET and MRI images in X-and-Y directions.
Figure 5. A 60-year-old male patient with intrapelvic lymphnode metastasis. Fusion images (C) with CT (A) and FDG PET (B) provided more accurate information of the localization of abnormal FDG uptake.
Figure 6. A 81-year-old male with recurrent lung cancer. Fusion images (C, F) of CT (A, D) and FDG PET (B, E) clearly showed the recurrent tumor.
Image Fusion System Using PACS/Alyafei et al. 143
as network connections. Over all, the throughput rates of our PACS were satisfactory. Essentially an archive’s function is to identify, store and protect data. To create a functioning archive, it requires a complete functional analysis and an assessment of how much space is needed for storage. Any PACS must be able to perform multiple tasks, such as: 1) identify and store all images in a common file format; 2) allow unlimited access to authorized users; 3) distribute images as needed in a timely manner; and 4) should have 100% confidence of image location and erase any unnecessary and duplicate information. All of this must be done in a secure environment. The archive media of PACS is what make it all happens. Our PACS storage capacity was designed to compile our workload for 2–3 years. The storage capacity could be increased easily and inexpensively, by adding more MOs whenever it is required. The results of volumetric measurement by CT/MRI images reproduced using AVS showed good correlation with the actual sphere’s volume, that suggests high accuracy of volumetric data may be achieved. We have evaluated the spatial resolution of the phantom images of MRI and PET by registration and fusion of these images. The overlapping spheres’ area of PET images was always bigger than the area of the MRI images, especially with low volumes. These differences may be due to the spatial resolution of the MRI and PET scanner. It may be also due to partial volume effects. This error could effect on the accuracy of registration and image fusion. Further investigations of volume, overlapping comparison and correction will be necessary to assess the image registration and fusion. The accuracy of the registration technique was evaluated by two basic approaches, quantitatively through a phantom study and by evaluation of the clinical images. The phantom study indicated that this technique was accurate within 2–4 mm in all directions. The patients’ data of fusion images with FDG PET and CT successfully provided us more accurate information about the localization of abnormal FDG uptake (Figs. 5, 6). Especially fusion image of PET and CT/MRI is quite useful for differentiation between lesions and normal structures,11 as well as for the diagnosis of recurrent tumor after the surgical interventions, because the deformed structure of treated region made it difficult to differentiate the tumor recurrence from reactive inflammation.12
Conclusion We have described the structure of our PACS and evaluated its performance by measuring the throughput rate and the storage capacity. The combination of PACS and image fusion provided very useful clinical tools and made it quite easy to maximize the benefit from the diagnostic imaging.
References 1. Badaoui, S.; Chameroy, V.; Aubry, F. A database manager of biomedical images. Med. Inform. 18:23–33; 1993. 2. ACR/NEMA-DICOM: Digital image and communication in medicine version 3.0, Standards Publication No. PS 3.13.9. Washington, DC: National Electrical Manufacturers Assoc., 1992–1993. 3. Prior, F.W. Specifying DICOM complince for modality interfaces. Radiographics 13:1381–1388; 1993. 4. Woods, R.P.; Cherry, S.R.; Mazziotta, J.C. Rapid automated algorithm for aligning and reslicing PET images. J. Comput. Assist. Tomogr. 16:620–633; 1992. 5. Levin, D.N.; Hu, X.; Tan, KM; et al. The brain: integrated three-dimensional display of MR and PET images. Radiology 172:783–789; 1989. 6. Hu, X.; Tan, K.M.; Levin, D.N.; Pelizzari, C.A.; Chen, G.T.Y. A volume-rendering technique for integrated three-dimensional display of MR and PET data. In: Hokne, K.H., ed. 3D imaging in medicine, Vol. F60. Berlin, Heidelberg: Springer-Verlag; 1990: 379–397. 7. Wahl, R.L.; Hutchins, G.D.; Buchsbaum, D.J.; Liebert, M.; Grossman, H.B.; Fisher, S. Fluorine-18-2-deoxy-2-fluoroD-glucose (FDG) uptake into human tumor xenografts: feasibility studies for cancer imaging with PET. Cancer 67:1544–1550; 1991. 8. Wahl, R.L.; Cody, R.L.; Hutchins, G.D.; Mudgett, E.E. Primary and metastatic breast carcinoma: initial clinical evaluation with PET with radilabeled glucose analogue 2[F-18] fuoro-2-deoxy-D-glucose. Radiology 179:765–770; 1991. 9. Wahl, R.L.; Meyer, C.; Koeppe, R.; Aisen, A.; Quint, L.E. Where is the tumor imaged on FDG PET scanning as a “hot spot?” Method for metabolic/anatomic registration [Abstract]. Radiology 181:151; 1991. 10. Inoue, T.; Koyama, Y.; Suzuki, H.; et al. Introducing a downsized, open computer system with asynchronous transfer mode in PACS and Its effect on image data transfer. Radiation Medicine 16:399–402; 1998. 11. Uematsu, H.; Sadato, N.; Yonekura, Y.; et al. Coregistration of FDG PET and MRI of the head and neck using normal distribution of FDG. J. Nucl. Med. 39:2121–2127; 1998. 12. Inoue, T.; Kim, E.E.; Komaki, R.; et al. Detecting recurrent or residual lung cancer with FDG-PET. J. Nucl. Med. 36:788–793; 1995.
Clinical Positron Imaging Vol. 2, No. 3, 137–143. 1999 Copyright 1999 Elsevier Science Inc. Printed in the USA. All rights reserved. 1095-0397/99 $–see front matter
ORIGINAL ARTICLE
Image Fusion System Using PACS for MRI, CT, and PET Images Saleh Alyafei, BHSc1, Tomio Inoue, MD1, Hong Zhang, MD1, Khalil Ahmed, MD1, Noboru Oriuchi, MD1, Noriko Sato, MD1, Hideki Suzuki, MD2, Keigo Endo, MD1 1
Department of Nuclear Medicine and Diagnostic Radiology, Gunma University School of Medicine. Showa machi 3-39-22, Maebashi, Gunma 371-8511, Japan; 2 Department of Medical Information, Gunma Prefectural College of Health Sciences, Gunma, Japan This paper presents and evaluates Picture Archiving and Communication System (PACS) and its interaction with the image fusion applications using positron emission tomography (PET), CT, and MRI, as well as some clinical applications of fusion images. A network connections between medium-sized PACS involving CT, MRI, SPECT, and PET were developed. Image registration and fusion was achieved in the PET’s workstation by Advanced Visual System (AVS) software. Reconstructed datasets of CT, MRI, SPECT, and PET were transferred and archived in PACS servers. Series of anatomical images of CT and MRI were fused with metabolical images of PET with 18F labeled fluoro-2-deoxy-D-glucose (FDG). Throughput rate of image data, as well as clinical applications of fusion images were evaluated and correlated with phantom studies. The average throughput rate of archiving and processing steps was 0.45 Mbps, 0.77 Mbps, respectively. Image fusion experiment using phantom and patients’ data showed high accuracy in all directions. The combination of PACS and image fusion provided very useful clinical tools and made it quite easy to maximize the benefit from the diagnostic imaging. (Clin Pos Imag 1999;2:137–143) 1999 Elsevier Science Inc. All rights reserved. Key Words: Picture Archiving and Communication System; PACS; Image Fusion.
Introduction
P
icture Archiving and Communication Systems (PACS) is becoming an integral part of most modern medical imaging departments. Tremendous volume of images accumulates in PACS database from different modalities such as CT, MRI, SPECT, and PET.1 PACS is the infrastructure supporting advances in many areas, such as in 1) hospital operations, it gives faster and more widely available access to image and reports offer the potential for savings in clinician time and length of stay; 2) radiology operations, it provides timely access to the images to improve the patient flow, throughput and image reporting; 3) case management, it can provide image data for systems streamlining the presentation of case review conferences; 4) teaching files, both educational and clinical goals are enhanced by convenient access to teaching cases during reading sessions. An on-line teaching file
Address correspondence to: Saleh Alyafei, Department of Nuclear Medicine, Gunma University School of Medicine, Showa machi 3-3922, Maebashi, Gunma 371-8511, Japan.
could provide access to relevant reference images for comparison during reading sessions; and 5) PACS promote and incorporate research results of multimodality fusion, into clinical use. PACS is the foundation resource of image registration and fusion applications. Multimodality image fusion application requires standardized parameters to be clinically useful. These parameters should cover interface of digital image transmission, filing and retrieval, image compression, and multi-image database. In contrast to these facts, the Digital Imaging and Communications in Medicine (DICOM) V3.0 was developed to aid in the acquisition and transport of images via PACS.2–3 Also, PACS is eliminating the need of multiple single-connections between workstation of image fusion and each workstation of CT, MRI, nuclear medicine, and PET. This paper describes and evaluates image fusion method using the advantage of a DICOM based PACS. Also, it presents some clinical applications of registration and image fusion of multimodality such as CT, MRI, and PET. 137
138 Clinical Positron Imaging, Volume 2, Number 3
Figure 1. Current PACS configuration, the thin lines represent the Ethernet network (10 Mbps), and the thick lines represent the ATM fiber connections (155 Mbps).
Materials and Methods Background and Configuration of PACS and Fusion System We adopted a hospital information system (HIS) in 1988 followed by a reporting system (RS) in 1991. One year later a medium-sized PACS initially installed, was linked by fiber distributed data interface (FDDI, 100 Mbps), optical loop local area network (LAN). Three CT scanners (TCT-900S, X-Force, Toshiba, Tokyo), two computed radiography (CR) (TCR, Toshiba), two MRIs (MAGNETOM-Simens, Signa-GE), two gamma cameras (PRISM 2000, PRISM 3000), and one whole body PET scanner (Shimadzu SET2400W, Kyoto, Japan) were connected to PACS. In 1997 PACS was updated and remodeled with a new main PACS server computer (UNIX workstation, EWS4800/320px, 128 MB RAM, 250 MHz, 14 GB HD, and 78 GB MO). Three PACS DICOM servers (UNIX workstation, EWS4800/ 460px, 128 MB, 250 MHz, 7 GB HD, and 78 GB MO), five gateway computers, and eight display terminal computers (PC pentium 133 MHz, 80 MB) were components of the new PACS. Images of all modalities were converted to DICOM 3.0 formats. Transmission control protocol/internet protocol (TCP/IP) traffic was used because of its support in the DICOM 3.0 specification. A new network of asynchronous transfer mode (ATM) was established with direct connection to the PACS controller computer and via Ethernet to the gateway computers and image display terminals (Figure 1).
The necessary on-line storage capacity based on the yearly statistics was computed. The approximate annual volume data production of two CT scanners, two MRI units, two SPECT gamma cameras, and one PET scanner have been included (Figure 2). Image compression of 1.5–2 factor has been applied. The compressed images are retrieved from the archive to the requester of the images. The latter would locally decompress the images to the original resolution, as such compression method would increase the throughput rate of the data and increase the storage capacity of PACS archive. The highest amount of storage was needed for the CT. The SPECT and PET images required the least amounts of storage in the PACS servers. One slice of CT image consisted of 512 * 512 pixels requires 524 KBytes. MRI’s slice consisted of 256 * 256 pixels requires 131 KBytes, PET’s slice consisted of 128 * 128 pixels requires 33 KBytes, and SPECT’s slice consisted of 64 * 64 pixels requires 8.2 KBytes. AVS (Advanced Visual Systems, Inc., Waltham, MA) software was installed in PET workstation, SET2400W (Digital Unix Ver. 3.2G, Compac Alpha Station 600, 5/333, 128 MB, 333 MHz) to determine the clinical potential of registration and image fusion. Patients underwent series of diagnostic examinations using CT, MRI, SPECT, or PET. The reconstructed images were transferred to PACS, then to SET2400W workstation. A method of three-dimensional registrations of CT, MRI, and PET datasets has been carried out and assessed.4–6 The current generation of PET scanners provides functional data with a spatial resolution of 3–6 mm. A
Image Fusion System Using PACS/Alyafei et al. 139
Figure 2. The approximate annual volume data. Two CT scanners, 2 MRI units, 2 SPECT gamma cameras, and 1 PET scanner were included.
high target-to-background ratio of smaller lesions less than 1 cm in diameter is often observed,7–8 because of the high spatial resolution of PET and the high tumor affinity of FDG in these cases, the identification of lesion localization based on only FDG PET is difficult. Fusion of metabolic FDG images of PET to the anatomical information available from MRI or CT scans with higher spatial resolution seem to be desirable.9
These datasets were transferred from CT, MRI, SPECT, and PET to PACS server (archiving step), as well as from PACS sever to the AVS workstation, (processing step). The average throughput measurements of each step were recorded, by copying a static buffer across the network. The system time is recorded before the actual transfer and once again after the transfer is completed.
Throughput Rate Measurement
Phantom Experiments
CT, MRI, and SPECT configuration consisted of one ATM switch (AS 155), one PACS server, with a direct ATM connection to the switch using the user network interface (UNI, Ver. 3.1). All the ATM interfaces operate at rate of 155 Mbps. The workstation of PET was connected to the PACS DICOM server via Ethernet line. Three modality input terminals (MITs), connected by Ethernet to the router, that has both Ethernet and ATM connections (AR 155, 3COM) (Figure 1). We have performed several transfer tests of patients’ data and phantom’s data. These tests were performed with a focus on a clinical application. Datasets of CT images of 3 patients, (total of 42 MB), MRI images of 2 patients (23 MB), nuclear medicine images of 2 patients (6.7 MB), and datasets of PET images of 2 patients (6 MB), as well as large volumes of phantom’s data of each modality were involved in the transferred experiment.
The basic experiments using sphere phantoms were performed to assess the accuracy of volume measurements by CT, MRI, SPECT, and PET, and also to evaluate the validity of image fusion and registration methods. Spheres of 5, 7, 13, 18, 25, and 40 mm in diameter were filled with a solution of uniform activity of 18F for the PET study, 99mTcO42 for the SPECT study. They were filled with air for the evaluation of CT and MRI. The spheres were positioned in a circle of z5 cm radius and placed in a 20-cm diameter phantom, with uniform distribution of attenuating medium such as water. All slices were reconstructed with a 10-mm thickness. A region of interest (ROI) on each sphere boundary was drawn on the transaxial images of CT, MRI, SPECT, and PET. The number of pixels inside each boundary was then computed. Each sphere’s area was calculated in mm2. The sphere’s volumes were determined by
140 Clinical Positron Imaging, Volume 2, Number 3
multiplying each sphere’s area with the slice thickness. The degree of misregistration of the spheres was evaluated by calculating the distance between the center of gravity of each spheres on PET image with the corresponding spheres on the CT, and MR image.
dard deviations of the PET voxels within the partitions defined by a segmentation of the CT/MRI intensities into 256 levels.
Clinical Application
Throughput Rate Measurements
Two patients with malignant tumors underwent CT to obtain anatomical information about the location and boundary of the lesions. In addition, each patient received PET scan, following the injection of FDG to assess the metabolic activity in the lesions. The patient was in supine position and the abdominal region was imaged with 3 mm contiguous transverse slices. Scans were performed before and following 150 cc injection of contrast medium to delineate tumor boundaries from normal tissue (scan parameters of 120 kVp and 200 mA, 512 * 512 matrix). The images were saved in the processing workstation of CT. The whole body PET scanner (Shimadzu, Kyoto, Japan), that provides 63 imaging slices with an axial field of view of 200 mm, was used for FDG PET study. A transmission scan for attenuation correction was conducted with a rotating 67 Ge-germanium rod source (z350–400 million counts). Patients were then injected with 185 Mbq of FDG. Following 40 minute incorporation period, an 8 minute emission scan was acquired. Transverse section images were reconstructed with a conventional filtered back-projection algorithm to yield a final in-plane resolution of z4.3 mm (full width at half maximum, FWHM) and an axial resolution of z4.3 mm FWHM. All cross-section image-sets were reconstructed initially on each acquisition system using standard vendorsupplied reconstruction algorithm. Image fusions were performed with PET workstation, running Digital UNIX and the native MIT X-windows graphical user interface integrated with AVS. Reconstructed CT, and MRI datasets were transferred from PACS server to the workstation of PET using TCP/IP. PET datasets were constructed in the workstation of PET, then converted to DICOM 3 format. An AVS module was developed to reconstruct the incoming transaxial slices into 3-dimensional slices, transverse, coronal, and sagittal. These three views were simultaneously used in assisting the fusion images process. The algorithm used for image registration was the intensity value matching method developed by Wood et al.5 It is based on the concept that two image volumes are in registration when the variance of the pixel-by-pixel ratio is a minimum. This assumes the spatial intensity patterns are similar for the two volumes, that is well justified for intermodality registration. For PET and CT/MRI registration the assumption is only approximately true. The concept of minimum variance of the pixel-by-pixel ratio was extended to minimize a weighted sum of stan-
The average throughput rates of CT, MRI, and SPECT datasets from imaging modality to main PACS server (archiving) were 0.58 Mbps, 0.43 Mbps, and 0.33 Mbps, respectively (Fig. 3A). The average throughput rate of PET datasets to the DICOM PACS server was 3.75 Mbps. The average throughput rates to the PET workstation (processing) from the main PACS server were 1.2 Mbps, 0.49 Mbps, and 0.62 Mbps for CT, MRI, and SPECT data, respectively (Figure 3B).
Results
Phantom Experiments The volume of spheres determined by CT, MRI, SPECT, and PET was compared to the actual volume. Regarding the maximum volume of 126 ml, the measured volumes by CT, MRI, SPECT, and PET were 129 ml, 130 ml, 179 ml, and 146 ml, respectively. Regarding the minimum volume of 2 ml, the volumes determined by CT, MRI, SPECT, and PET were 3 ml, 3 ml, 0 ml, and 15 ml, respectively. Regarding the sphere with a 20-mm radius, the overestimation of the measurement of radius by CT, MRI, SPECT, and PET compared to the radius of the actual sphere were 3%, 3%, 43%, and 36%, respectively. A transaxial fused image showed an excellent registration of PET and MRI images in X-and-Y directions (Figure 4). The degree of misregistration was relatively small always (,4 mm).
Clinical Application A fusion image of a 60-year-old male patient with prostate cancer was obtained. Transverse CT slice through the pelvic area revealed lymphnode swelling, and bone metastasis in sacrum and iliac bone (Figure 5A). The FDG-PET image was registered using AVS (Figure 5B), and sliced to match the CT image. PET/CT fused 2-dimensional images were produced (Figure 5C). With the PET scan alone, the tumors were readily apparent but it was difficult to decide the location and the nature of these tumors due to the lack of anatomical information on the PET scan. By the fusion images of PET and CT, we could identify clearly the anatomical location of abnormal FDG uptake. Another patient is an 81-years-old male with lung cancer, and left pleural thickness after the irradiation therapy. Because tumor marker (CEA) was increasing, recurrent lung cancer was suspected. On the CT image (Figures 6A and 6D), there is an apical pleural thickness of the left lung, suggesting cancer involvement or
Image Fusion System Using PACS/Alyafei et al. 141
Figure 3. (A) The average throughput rates in Mbps of CT, MRI, SPECT, and PET datasets from imaging modality to PACS servers (archiving). (B) The average throughput rates in Mbps from the main PACS server to the PET workstation (processing).
chronic inflammation. The corresponding PET scan (Figures 6B and 6E) revealed high uptake around the left pleural cavity with a hot spot in the median portion. The fused image demonstrating good corresponding between the area of cancer infiltration on the PET image and the area of pleural thickness on the CT image (Figures 6C and 6F).
Discussion PACS has tremendous benefits and values outside of diagnostic department as well as internally. The biggest benefit derived from an ideal PACS is to break the physical as well as time barrier for information exchange. PACS is playing a vital role by contributing the fundamental requirements to achieve clinical setup for image fusion in our institute. Most communications providers refer to bandwidth when talking about data. Bandwidth refers to the amount of information that can be passed through a given connection, usually measured in Kilobits/second (Kbps) or Megabits/second (Mbps). The larger the
bandwidth, the more data can be passed through in a measured second. In our throughput rate experiment, the highest throughput rate was between PET and DICOM PACS server, almost 50% faster than other network connections. This is probably due to the fact that PET workstation and DICOM PACS server have a singular connection, compared with the multiple connections of CT, MRI, and SPECT with the main PACS server. The nature of the multiple connections of the main PACS server, causes high data traffic, that slows the throughput rate. Another reason could be that the main PACS server is using the Ethernet branches to connect MIT computers with ATM router. A bottleneck may occurs within these connections, and collisions of image data is highly happening.10 By measuring the throughput rate, the biggest drawback to the image transfer is not the network but workstations. Even Ethernet cable can operate at 10 Mbps and theoretically networks have unlimited potential transfer speed if they are constructed of fiber. Workstations are limited by the operating system and hardware utilized for internal communications as well
Figure 4. (A) Transaxial image of a phantom acquired by MRI. (B) A corresponding co-registered transaxial image of the same phantom acquired by PET scanner. (C) A transaxial fused image slice showing excellent registration of PET and MRI images in X-and-Y directions.
Figure 5. A 60-year-old male patient with intrapelvic lymphnode metastasis. Fusion images (C) with CT (A) and FDG PET (B) provided more accurate information of the localization of abnormal FDG uptake.
Figure 6. A 81-year-old male with recurrent lung cancer. Fusion images (C, F) of CT (A, D) and FDG PET (B, E) clearly showed the recurrent tumor.
Image Fusion System Using PACS/Alyafei et al. 143
as network connections. Over all, the throughput rates of our PACS were satisfactory. Essentially an archive’s function is to identify, store and protect data. To create a functioning archive, it requires a complete functional analysis and an assessment of how much space is needed for storage. Any PACS must be able to perform multiple tasks, such as: 1) identify and store all images in a common file format; 2) allow unlimited access to authorized users; 3) distribute images as needed in a timely manner; and 4) should have 100% confidence of image location and erase any unnecessary and duplicate information. All of this must be done in a secure environment. The archive media of PACS is what make it all happens. Our PACS storage capacity was designed to compile our workload for 2–3 years. The storage capacity could be increased easily and inexpensively, by adding more MOs whenever it is required. The results of volumetric measurement by CT/MRI images reproduced using AVS showed good correlation with the actual sphere’s volume, that suggests high accuracy of volumetric data may be achieved. We have evaluated the spatial resolution of the phantom images of MRI and PET by registration and fusion of these images. The overlapping spheres’ area of PET images was always bigger than the area of the MRI images, especially with low volumes. These differences may be due to the spatial resolution of the MRI and PET scanner. It may be also due to partial volume effects. This error could effect on the accuracy of registration and image fusion. Further investigations of volume, overlapping comparison and correction will be necessary to assess the image registration and fusion. The accuracy of the registration technique was evaluated by two basic approaches, quantitatively through a phantom study and by evaluation of the clinical images. The phantom study indicated that this technique was accurate within 2–4 mm in all directions. The patients’ data of fusion images with FDG PET and CT successfully provided us more accurate information about the localization of abnormal FDG uptake (Figs. 5, 6). Especially fusion image of PET and CT/MRI is quite useful for differentiation between lesions and normal structures,11 as well as for the diagnosis of recurrent tumor after the surgical interventions, because the deformed structure of treated region made it difficult to differentiate the tumor recurrence from reactive inflammation.12
Conclusion We have described the structure of our PACS and evaluated its performance by measuring the throughput rate and the storage capacity. The combination of PACS and image fusion provided very useful clinical tools and made it quite easy to maximize the benefit from the diagnostic imaging.
References 1. Badaoui, S.; Chameroy, V.; Aubry, F. A database manager of biomedical images. Med. Inform. 18:23–33; 1993. 2. ACR/NEMA-DICOM: Digital image and communication in medicine version 3.0, Standards Publication No. PS 3.13.9. Washington, DC: National Electrical Manufacturers Assoc., 1992–1993. 3. Prior, F.W. Specifying DICOM complince for modality interfaces. Radiographics 13:1381–1388; 1993. 4. Woods, R.P.; Cherry, S.R.; Mazziotta, J.C. Rapid automated algorithm for aligning and reslicing PET images. J. Comput. Assist. Tomogr. 16:620–633; 1992. 5. Levin, D.N.; Hu, X.; Tan, KM; et al. The brain: integrated three-dimensional display of MR and PET images. Radiology 172:783–789; 1989. 6. Hu, X.; Tan, K.M.; Levin, D.N.; Pelizzari, C.A.; Chen, G.T.Y. A volume-rendering technique for integrated three-dimensional display of MR and PET data. In: Hokne, K.H., ed. 3D imaging in medicine, Vol. F60. Berlin, Heidelberg: Springer-Verlag; 1990: 379–397. 7. Wahl, R.L.; Hutchins, G.D.; Buchsbaum, D.J.; Liebert, M.; Grossman, H.B.; Fisher, S. Fluorine-18-2-deoxy-2-fluoroD-glucose (FDG) uptake into human tumor xenografts: feasibility studies for cancer imaging with PET. Cancer 67:1544–1550; 1991. 8. Wahl, R.L.; Cody, R.L.; Hutchins, G.D.; Mudgett, E.E. Primary and metastatic breast carcinoma: initial clinical evaluation with PET with radilabeled glucose analogue 2[F-18] fuoro-2-deoxy-D-glucose. Radiology 179:765–770; 1991. 9. Wahl, R.L.; Meyer, C.; Koeppe, R.; Aisen, A.; Quint, L.E. Where is the tumor imaged on FDG PET scanning as a “hot spot?” Method for metabolic/anatomic registration [Abstract]. Radiology 181:151; 1991. 10. Inoue, T.; Koyama, Y.; Suzuki, H.; et al. Introducing a downsized, open computer system with asynchronous transfer mode in PACS and Its effect on image data transfer. Radiation Medicine 16:399–402; 1998. 11. Uematsu, H.; Sadato, N.; Yonekura, Y.; et al. Coregistration of FDG PET and MRI of the head and neck using normal distribution of FDG. J. Nucl. Med. 39:2121–2127; 1998. 12. Inoue, T.; Kim, E.E.; Komaki, R.; et al. Detecting recurrent or residual lung cancer with FDG-PET. J. Nucl. Med. 36:788–793; 1995.