Optic Disc and Cup Image Segmentation Utilizing Contour-Based Transformation and Sequence Labeling Networks.

Optic disc (OD) and optic cup (OC) segmentation are important steps for automatic screening and diagnosing of optic nerve head abnormalities such as glaucoma. Many recent works formulated the OD and OC segmentation as a pixel classification task. However, it is hard for these methods to explicitly model the spatial relations between the labels in the output mask. Furthermore, the proportion of the background, OD and OC are unbalanced which also may result in a biased model as well as introduce more noise. To address these problems, we developed an approach that follows a coarse-to-fine segmentation process. We start with a U-Net to obtain a rough segmenting boundary and then crop the area around the boundary to form a boundary contour centered image. Second, inspired by sequence labeling tasks in natural language processing, we regard the OD and OC segmentation as a sequence labeling task and propose a novel fully convolutional network called SU-Net and combine it with the Viterbi algorithm to jointly decode the segmentation boundary. We also introduced a geometric parameter-based data augmentation method to generate more training samples in order to minimize the differences between training and test sets and reduce overfitting. Experimental results show that our method achieved state-of-the-art results on 2 datasets for both OD and OC segmentation and our method outperforms most of the ophthalmologists in terms of achieving agreement out of 6 ophthalmologists on the MESSIDOR dataset for both OD and OC segmentation. In terms of glaucoma screening, we achieved the best cup-to-disc ratio (CDR) error and area under the ROC curve (AUC) for glaucoma classification on the Drishti-GS dataset.

Knowledge-driven decision support for assessing dose distributions in radiation therapy of head and neck cancer.

PURPOSE: Clinical data that are generated through routine radiation therapy procedures can be leveraged as a source of knowledge to provide evidence-based decision support for future patients. Treatment planning in radiation therapy often relies on trial-and-error iterations, experience, judgment calls and general guidelines. The authors present a knowledge-driven decision support system that assists clinicians by reducing some of the uncertainties associated with treatment planning and provides quantified empirical estimates to help minimize the radiation dose to healthy critical structures surrounding the tumor. METHODS: A database of retrospective DICOM RT data fuels a decision support engine, which assists clinicians in selecting dose constraints and assessing dose distributions. The first step is to quantify the spatial relationships between the tumor and surrounding critical structures through features that account for distance, volume, overlap, location, shape and orientation. These features are used to identify database cases that are anatomically similar to the new patient. The dose profiles of these database cases can help clinicians to estimate an acceptable dose distribution for the new case, based on empirical evidence. Since database diversity is essential for good system performance, an infrastructure for multi-institutional collaboration was also conceptualized in order to pave the way for data sharing of protected health information. RESULTS: A set of 127 retrospective test cases was collected from a single institution in order to conduct a leave-one-out evaluation of the decision support module. In 72 % of these retrospective test cases, patients with similar tumor anatomy were also found to exhibit similar radiation dose distributions. This demonstrates the system's ability to successfully extract retrospective database cases that can estimate the new patient's dose distribution. CONCLUSION: The radiation therapy treatment planning decision support system presented here can assist clinicians in determining good dose constraints and assessing dose distributions by using knowledge gained from retrospective treatment plans.

Development of a novel imaging informatics-based system with an intelligent workflow engine (IWEIS) to support imaging-based clinical trials.

Imaging based clinical trials can benefit from a solution to efficiently collect, analyze, and distribute multimedia data at various stages within the workflow. Currently, the data management needs of these trials are typically addressed with custom-built systems. However, software development of the custom-built systems for versatile workflows can be resource-consuming. To address these challenges, we present a system with a workflow engine for imaging based clinical trials. The system enables a project coordinator to build a data collection and management system specifically related to study protocol workflow without programming. Web Access to DICOM Objects (WADO) module with novel features is integrated to further facilitate imaging related study. The system was initially evaluated by an imaging based rehabilitation clinical trial. The evaluation shows that the cost of the development of system can be much reduced compared to the custom-built system. By providing a solution to customize a system and automate the workflow, the system will save on development time and reduce errors especially for imaging clinical trials.

Design and development of an ethnically-diverse imaging informatics-based eFolder system for multiple sclerosis patients.

PURPOSE: MRI has been used to identify multiple sclerosis (MS) lesions in brain and spinal cord visually. Integrating patient information into an electronic patient record system has become key for modern patient care in medicine in recent years. Clinically, it is also necessary to track patients' progress in longitudinal studies, in order to provide comprehensive understanding of disease progression and response to treatment. As the amount of required data increases, there exists a need for an efficient systematic solution to store and analyze MS patient data, disease profiles, and disease tracking for both clinical and research purposes. METHOD: An imaging informatics based system, called MS eFolder, has been developed as an integrated patient record system for data storage and analysis of MS patients. The eFolder system, with a DICOM-based database, includes a module for lesion contouring by radiologists, a MS lesion quantification tool to quantify MS lesion volume in 3D, brain parenchyma fraction analysis, and provide quantitative analysis and tracking of volume changes in longitudinal studies. Patient data, including MR images, have been collected retrospectively at University of Southern California Medical Center (USC) and Los Angeles County Hospital (LAC). The MS eFolder utilizes web-based components, such as browser-based graphical user interface (GUI) and web-based database. The eFolder database stores patient clinical data (demographics, MS disease history, family history, etc.), MR imaging-related data found in DICOM headers, and lesion quantification results. Lesion quantification results are derived from radiologists' contours on brain MRI studies and quantified into 3-dimensional volumes and locations. Quantified results of white matter lesions are integrated into a structured report based on DICOM-SR protocol and templates. The user interface displays patient clinical information, original MR images, and viewing structured reports of quantified results. The GUI also includes a data mining tool to handle unique search queries for MS. System workflow and dataflow steps has been designed based on the IHE post-processing workflow profile, including workflow process tracking, MS lesion contouring and quantification of MR images at a post-processing workstation, and storage of quantitative results as DICOM-SR in DICOM-based storage system. The web-based GUI is designed to display zero-footprint DICOM web-accessible data objects (WADO) and the SR objects. SUMMARY: The MS eFolder system has been designed and developed as an integrated data storage and mining solution in both clinical and research environments, while providing unique features, such as quantitative lesion analysis and disease tracking over a longitudinal study. A comprehensive image and clinical data integrated database provided by MS eFolder provides a platform for treatment assessment, outcomes analysis and decision-support. The proposed system serves as a platform for future quantitative analysis derived automatically from CAD algorithms that can also be integrated within the system for individual disease tracking and future MS-related research. Ultimately the eFolder provides a decision-support infrastructure that can eventually be used as add-on value to the overall electronic medical record.

Effective staging of fibrosis by the selected texture features of liver: Which one is better, CT or MR imaging?

PURPOSE: Texture patterns of hepatic fibrosis are one of the important biomarkers to diagnose and classify chronic liver disease from initial to end stage on computed tomography (CT) or magnetic resonance (MR) images. Computer-aided diagnosis (CAD) of liver cirrhosis using texture features has become popular in recent research advances. To date, however, properly selecting effective texture features and image parameters is still mostly undetermined and not well-defined. In this study, different types of datasets acquired from CT and MR images are investigated to select the optimal parameters and features for the proper classification of fibrosis. METHODS: A total of 149 patients were scanned by multi-detector computed tomography (MDCT) and 218 patients were scanned using 1.5T and 3T superconducting MR scanners for an abdominal examination. All cases were verified by needle biopsies as the gold standard of our experiment, ranging from 0 (no fibrosis) to 5 (cirrhosis). For each case, at least four sequenced phase images are acquired by CT or MR scanners: pre-contrast, arterial, portal venous and equilibrium phase. For both imaging modalities, 15 texture features calculated from gray level co-occurrence matrix (GLCM) are extracted within an ROI in liver as one set of input vectors. Each combination of these input subsets is checked by using support vector machine (SVM) with leave-one-case-out method to differentiate fibrosis into two groups: non-cirrhosis or cirrhosis. In addition, 10 ROIs in the liver are manually selected in a disperse manner by experienced radiologist from each sequenced image and each of the 15 features are averaged across the 10 ROIs for each case to reduce the validation time. The number of input items is selected from the various combinations of 15 features, from which the accuracy rate (AR) is calculated by counting the percentage of correct answers on each combination of features aggregated to determine a liver stage score and then compared to the gold standard. RESULTS: According to the accuracy rate (AR) calculated from each combination, the optimal number of texture features to classify liver fibrosis degree ranges from 4 to 7, no matter which modality was utilized. The overall performance calculated by the average sum of maximum AR value of all 15 features is 66.83% in CT images, while 68.14%, and 71.98% in MR images, respectively; among the 15 texture features, mean gray value and entropy are the most commonly used features in all 3 imaging datasets. The correlation feature has the lowest AR value and was removed as an effective feature in all datasets. AR value tends to increase with the injection of contrast agency, and both CT and MR images reach the highest AR performance during the equilibrium phase. CONCLUSIONS: Comparing the accuracy of classification with two imaging modalities, the MR images have an advantage over CT images with regards to AR performance of the 15 selected texture features, while 3T MRI is better than 1.5T MRI to classify liver fibrosis. Finally, the texture analysis is more effective during equilibrium phase than in any of the other phased images.

Medical imaging informatics simulators: a tutorial.

PURPOSE: A medical imaging informatics infrastructure (MIII) platform is an organized method of selecting tools and synthesizing data from HIS/RIS/PACS/ePR systems with the aim of developing an imaging-based diagnosis or treatment system. Evaluation and analysis of these systems can be made more efficient by designing and implementing imaging informatics simulators. This tutorial introduces the MIII platform and provides the definition of treatment/diagnosis systems, while primarily focusing on the development of the related simulators. METHODS: A medical imaging informatics (MII) simulator in this context is defined as a system integration of many selected imaging and data components from the MIII platform and clinical treatment protocols, which can be used to simulate patient workflow and data flow starting from diagnostic procedures to the completion of treatment. In these processes, DICOM and HL-7 standards, IHE workflow profiles, and Web-based tools are emphasized. From the information collected in the database of a specific simulator, evidence-based medicine can be hypothesized to choose and integrate optimal clinical decision support components. Other relevant, selected clinical resources in addition to data and tools from the HIS/RIS/PACS and ePRs platform may also be tailored to develop the simulator. These resources can include image content indexing, 3D rendering with visualization, data grid and cloud computing, computer-aided diagnosis (CAD) methods, specialized image-assisted surgical, and radiation therapy technologies. RESULTS: Five simulators will be discussed in this tutorial. The PACS-ePR simulator with image distribution is the cradle of the other simulators. It supplies the necessary PACS-based ingredients and data security for the development of four other simulators: the data grid simulator for molecular imaging, CAD-PACS, radiation therapy simulator, and image-assisted surgery simulator. The purpose and benefits of each simulator with respect to its clinical relevance are presented. CONCLUSION: The concept, design, and development of these five simulators have been implemented in laboratory settings for education and training. Some of them have been extended to clinical applications in hospital environments.

An observer study for a computer-aided reading protocol (CARP) in the screening environment for digital mammography.

RATIONALE AND OBJECTIVES: The aims of this study were to investigate improving work flow efficiency by shortening the reading time of digital mammograms using a computer-aided reading protocol (CARP) in the screening environment and to increase detection sensitivity using CARP, compared to the current protocol, commonly referred to as the quadrant view (QV). MATERIALS AND METHODS: A total of 200 cases were selected for a receiver-operating characteristic (ROC) study to evaluate two image display work flows, CARP and QV, in the screening environment. A Web-based tool was developed for scoring, reporting, and statistical analysis. Cases were scored for and stratified by difficulty. A total of six radiologists of differing levels of training ranging from dedicated mammographers to senior radiology residents participated. Each was timed while interpreting the 200 cases in groups of 50, first using QV and then, after a washout period, using CARP. The data were analyzed using ROC and κ analysis. Interpretation times were also assessed. RESULTS: Using QV, readers' average area under the ROC curve was 0.68 (range, 0.54-0.73). Using CARP, readers' average area under the ROC curve was 0.71 (range, 0.66-0.75). There was no statistically significant difference in reader performance using either work flow. However, there was a statistically significant reduction in the average interpretation time of negative cases from 64.7 seconds using QV to 58.8 seconds using CARP. CONCLUSIONS: CARP determines the display order of regions of interest depending on computer-aided detection findings. This is a variation of traditional computer-aided detection for digital mammography that has the potential to reduce interpretation times of studies with negative findings without significantly affecting sensitivity, thus allowing improved work flow efficiency in the screening environment, in which, in most settings, the majority of cases are negative.

Racial differences in growth patterns of children assessed on the basis of bone age.

PURPOSE: To collect up-to-date data in healthy children to create a digital hand atlas (DHA) that can be used to evaluate, on the basis of the Greulich and Pyle atlas method, racial differences in skeletal growth patterns of Asian, African American, white, and Hispanic children in the United States. MATERIALS AND METHODS: This retrospective study was HIPAA compliant and approved by the institutional review board. Informed consent was obtained from all subjects or their guardians. From May 1997 to March 2008, a DHA containing 1390 hand and wrist radiographs obtained in male and female Asian, African American, white, and Hispanic children with normal skeletal development was developed. The age of subjects ranged from 1 day to 18 years. Each image was read by two pediatric radiologists working independently and without knowledge of the subject's chronologic age, and evaluation was based on their experience with the Greulich and Pyle atlas. Statistical analyses were performed with the paired-samples t test and analysis of variance to study racial differences in growth patterns. P

Automated Bone Age Assessment of Older Children using the Radius.

The Digital Hand Atlas in Assessment of Skeletal Development is a large-scale Computer Aided Diagnosis (CAD) project for automating the process of grading Skeletal Development of children from 0-18 years of age. It includes a complete collection of 1,400 normal hand X-rays of children between the ages of 0-18 years of age. Bone Age Assessment is used as an index of skeletal development for detection of growth pathologies that can be related to endocrine, malnutrition and other disease types. Previous work at the Image Processing and Informatics Lab (IPILab) allowed the bone age CAD algorithm to accurately assess bone age of children from 1 to 16 (male) or 14 (female) years of age using the Phalanges as well as the Carpal Bones. At the older ages (16(male) or 14(female) -19 years of age) the Phalanges as well as the Carpal Bones are fully developed and do not provide well-defined features for accurate bone age assessment. Therefore integration of the Radius Bone as a region of interest (ROI) is greatly needed and will significantly improve the ability to accurately assess the bone age of older children. Preliminary studies show that an integrated Bone Age CAD that utilizes the Phalanges, Carpal Bones and Radius forms a robust method for automatic bone age assessment throughout the entire age range (1-19 years of age).

A knowledge-based imaging informatics approach for managing proton beam therapy of cancer patients.

The need for a unified patient-oriented information system to handle complex proton therapy (PT) imaging and informatics data during the course of patient treatment is becoming steadily apparent due to the ever increasing demands for better diagnostic treatment planning and more accurate information. Currently, this information is scattered throughout each of the different treatment and information systems in the oncology department. Furthermore, the lack of organization with standardized methods makes it difficult and time-consuming to navigate through the maze of data, resulting in challenges during patient treatment planning. We present a methodology to develop this electronic patient record (ePR) system based on DICOM standards and perform knowledge-based medical imaging informatics research on specific clinical scenarios where patients are treated with PT. Treatment planning is similar in workflow to traditional radiation therapy (RT) methods such as intensity-modulated radiation therapy (IMRT), which utilizes a priori knowledge to drive the treatment plan in an inverse manner. In March 2006, two new RT objects were drafted in a DICOM-RT Supplement 102 specifically for ion therapy, which includes PT. The standardization of DICOM-RT-ION objects and the development of a knowledge base as well as decision-support tools that can be add-on features to the ePR DICOM-RT system were researched. This methodology can be used to extend to PT and the development of future clinical decision-making scenarios during the course of the patient's treatment that utilize "inverse treatment planning." We present the initial steps of this imaging and informatics methodology for PT and lay the foundation for development of future decision-support tools tailored to cancer patients treated with PT. By integrating decision-support knowledge and tools designed to assist in the decision-making process, a new and improved "knowledge-enhanced treatment planning" approach can be realized.

CAD-PACS integration tool kit based on DICOM secondary capture, structured report and IHE workflow profiles.

Computer aided diagnosis/detection (CAD) goes beyond subjective visual assessment of clinical images providing quantitative computer analysis of the image content, and can greatly improve clinical diagnostic outcome. Many CAD applications, including commercial and research CAD, have been developed with no ability to integrate the CAD results with a clinical picture archiving and communication system (PACS). This has hindered the extensive use of CAD for maximum benefit within a clinical environment. In this paper, we present a CAD-PACS integration toolkit that integrates CAD results with a clinical PACS. The toolkit is a software package with two versions: DICOM (digital imaging and communications in medicine)-SC (secondary capture) and DICOM-IHE (Integrating the Healthcare Enterprise). The former uses the DICOM secondary capture object model to convert the screen shot of the CAD results to a DICOM image file for PACS workstations to display, while the latter converts the CAD results to a DICOM structured report (SR) based on IHE Workflow Profiles. The DICOM-SC method is simple and easy to be implemented without ability for further data mining of CAD results, while the DICOM-IHE can be used for data mining of CAD results in the future but more complicated to implement than the DICOM-SC method.

Image-assisted knowledge discovery and decision support in radiation therapy planning.

The need for quantified knowledge and decision-support tools to handle complex radiation therapy (RT) imaging and informatics data is becoming steadily apparent. Lessons can be learned from current CAD applications in radiology. This paper proposes a methodology to develop this quantified knowledge and decision-support tools to facilitate RT treatment planning. The methodology is applied to cancer patient cases treated by intensity modulated radiation therapy (IMRT). The use of the "inverse treatment planning" and imaging intensive nature of IMRT allows for the development of such image-assisted tools for supporting decision-making thus providing better workflow efficiency and more precise dose predictions.

Automatic bone age assessment for young children from newborn to 7-year-old using carpal bones.

A computer-aided-diagnosis (CAD) method has been previously developed based on features extracted from phalangeal regions of interest (ROI) in a digital hand atlas, which can assess bone age of children from ages 7 to 18 accurately. Therefore, in order to assess the bone age of children in younger ages, the inclusion of carpal bones is necessary. However, due to various factors including the uncertain number of bones appearing, non-uniformity of soft tissue, low contrast between the bony structure and soft tissue, automatic segmentation and identification of carpal bone boundaries is an extremely challenging task. Past research works on carpal bone segmentation were performed utilizing dynamic thresholding. However, due to the limitation of the segmentation algorithm, carpal bones have not been taken into consideration in the bone age assessment procedure. In this paper, we developed and implemented a knowledge-based method for fully automatic carpal bone segmentation and morphological feature analysis. Fuzzy classification was then used to assess the bone age based on the selected features. This method has been successfully applied on all cases in which carpal bones have not overlapped. CAD results of total about 205 cases from the digital hand atlas were evaluated against subject chronological age as well as readings of two radiologists. It was found that the carpal ROI provides reliable information in determining the bone age for young children from newborn to 7-year-old.

A HIPAA-compliant architecture for securing clinical images.

The Health Insurance Portability and Accountability Act (HIPAA, instituted April 2003) Security Standards mandate health institutions to protect health information against unauthorized use or disclosure. One approach to addressing this mandate is by utilizing user access control and generating audit trails of the various authorized as well as unauthorized user access of health data. Although most current clinical image systems [e.g., picture archiving and communication system (PACS)] have components that generate log files for application debugging purposes, there is a lack of methodology to obtain and synthesize the pertinent data from the large volumes of log data generated by these multiple components within a PACS. We have designed a HIPAA-compliant architecture specifically for tracking and auditing the image workflow of clinical imaging systems such as PACS. As an initial first step, we developed HIPAA-compliant auditing system (H-CAS) based on parts of this HIPAA-compliant architecture. H-CAS was implemented within a test-bed PACS simulator located in the Image Processing and Informatics lab at the University of Southern California. Evaluation scenarios were developed where different user types performed legal and illegal access of PACS image data within each of the different components in the PACS simulator. Results were based on whether the scenarios of unauthorized access were correctly identified and documented as well as on normal operational activity. Integration and implementation pitfalls were also noted and included.

Wireless remote control of clinical image workflow: using a PDA for off-site distribution and disaster recovery.

This paper describes a picture archiving and communication system (PACS) tool based on Web technology that remotely manages medical images between a PACS archive and remote destinations. Successfully implemented in a clinical environment and also demonstrated for the past 3 years at the conferences of various organizations, including the Radiological Society of North America, this tool provides a very practical and simple way to manage a PACS, including off-site image distribution and disaster recovery. The application is robust and flexible and can be used on a standard PC workstation or a Tablet PC, but more important, it can be used with a personal digital assistant (PDA). With a PDA, the Web application becomes a powerful wireless and mobile image management tool. The application's quick and easy-to-use features allow users to perform Digital Imaging and Communications in Medicine (DICOM) queries and retrievals with a single interface, without having to worry about the underlying configuration of DICOM nodes. In addition, this frees up dedicated PACS workstations to perform their specialized roles within the PACS workflow. This tool has been used at Saint John's Health Center in Santa Monica, California, for 2 years. The average number of queries per month is 2,021, with 816 C-MOVE retrieve requests. Clinical staff members can use PDAs to manage image workflow and PACS examination distribution conveniently for off-site consultations by referring physicians and radiologists and for disaster recovery. This solution also improves radiologists' effectiveness and efficiency in health care delivery both within radiology departments and for off-site clinical coverage.

Utilizing data grid architecture for the backup and recovery of clinical image data.

Grid Computing represents the latest and most exciting technology to evolve from the familiar realm of parallel, peer-to-peer and client-server models. However, there has been limited investigation into the impact of this emerging technology in medical imaging and informatics. In particular, PACS technology, an established clinical image repository system, while having matured significantly during the past ten years, still remains weak in the area of clinical image data backup. Current solutions are expensive or time consuming and the technology is far from foolproof. Many large-scale PACS archive systems still encounter downtime for hours or days, which has the critical effect of crippling daily clinical operations. In this paper, a review of current backup solutions will be presented along with a brief introduction to grid technology. Finally, research and development utilizing the grid architecture for the recovery of clinical image data, in particular, PACS image data, will be presented. The focus of this paper is centered on applying a grid computing architecture to a DICOM environment since DICOM has become the standard for clinical image data and PACS utilizes this standard. A federation of PACS can be created allowing a failed PACS archive to recover its image data from others in the federation in a seamless fashion. The design reflects the five-layer architecture of grid computing: Fabric, Resource, Connectivity, Collective, and Application Layers. The testbed Data Grid is composed of one research laboratory and two clinical sites. The Globus 3.0 Toolkit (Co-developed by the Argonne National Laboratory and Information Sciences Institute, USC) for developing the core and user level middleware is utilized to achieve grid connectivity. The successful implementation and evaluation of utilizing data grid architecture for clinical PACS data backup and recovery will provide an understanding of the methodology for using Data Grid in clinical image data backup for PACS, as well as establishment of benchmarks for performance from future grid technology improvements. In addition, the testbed can serve as a road map for expanded research into large enterprise and federation level data grids to guarantee CA (Continuous Availability, 99.999% up time) in a variety of medical data archiving, retrieval, and distribution scenarios.

HIPAA compliant auditing system for medical images.

As an official regulation for healthcare privacy and security, Health Insurance Portability and Accountability Act (HIPAA) mandates health institutions to protect health information against unauthorized use or disclosure. One such method proposed by HIPAA Security Standards is audit trail, which records and examines health information access activities. HIPAA mandates healthcare providers to have the ability to generate audit trails on data access activities for any specific patient. Although current medical imaging systems generate activity logs, there is a lack of formal methodology to interpret these large volumes of log data and generate HIPAA compliant auditing trails. This paper outlines the design of a HIPAA compliant auditing system (HCAS) for medical images in imaging systems such as PACS and discusses the development of a security monitoring (SM) toolkit based on some of the partial components in HCAS.

Informatics in radiology (infoRAD): A complete continuous-availability PACS archive server.

The operational reliability of the picture archiving and communication system (PACS) server in a filmless hospital environment is always a major concern because server failure could cripple the entire PACS operation. A simple, low-cost, continuous-availability (CA) PACS archive server was designed and developed. The server makes use of a triple modular redundancy (TMR) system with a simple majority voting logic that automatically identifies a faulty module and removes it from service. The remaining two modules continue normal operation with no adverse effects on data flow or system performance. In addition, the server is integrated with two external mass storage devices for short- and long-term storage. Evaluation and testing of the server were conducted with laboratory experiments in which hardware failures were simulated to observe recovery time and the resumption of normal data flow. The server provides maximum uptime (99.999%) for end users while ensuring the transactional integrity of all clinical PACS data. Hardware failure has only minimal impact on performance, with no interruption of clinical data flow or loss of data. As hospital PACS become more widespread, the need for CA PACS solutions will increase. A TMR CA PACS archive server can reliably help achieve CA in this setting.

A continuous available (CA) server for medical imaging applications.

RATIONALE AND OBJECTIVES: The trend of medical imaging research and application is toward large database management and manipulation, which requires a robust image server to receive image data from sources and to deliver them to users reliably and in a timely fashion. This article describes the design, implementation, and clinical applications of a continuous available (CA) image server for these purposes. MATERIALS AND METHODS: The design of the CA image server is based on the concept of a triple modular redundancy server with three redundant server modules. Coupled with a majority voting mechanism in the three modules and failover software, the triple modular redundancy server takes care of all single points of failure hardware components in the CA image server automatically to achieve fault tolerance. Methods and procedures of evaluating the fault tolerance system reliability caused by network connectivity, motherboard, and disk storage failures are described. RESULTS: Thorough experimental results in laboratory and clinical environments verify that the image server achieves 99.999% hardware up time (or 5 minutes/year down time), satisfying the industrial terminology of hardware continuous availability. Performance of failover of the CA image server is automatically tabulated during these procedures. CONCLUSION: Applications of CA image server are extensive. Two examples are given including Picture Archiving and Communication System, and off-site back-up archive using the Application Service Provider model. As designed, the CA image server is portable, scalable, affordable, easy to install, and requires no human intervention during failover and system recovery.