Imaging Informatics: 25 Years of Progress.

The science and applications of informatics in medical imaging have advanced dramatically in the past 25 years. This article provides a selective overview of key developments in medical imaging informatics. Advances in standards and technologies for compression and transmission of digital images have enabled Picture Archiving and Communications Systems (PACS) and teleradiology. Research in speech recognition, structured reporting, ontologies, and natural language processing has improved the ability to generate and analyze the reports of imaging procedures. Informatics has provided tools to address workflow and ergonomic issues engendered by the growing volume of medical image information. Research in computeraided detection and diagnosis of abnormalities in medical images has opened new avenues to improve patient care. The growing number of medical-imaging examinations and their large volumes of information create a natural platform for "big data" analytics, particularly when joined with high-dimensional genomic data. Radiogenomics investigates relationships between a disease's genetic and gene-expression characteristics and its imaging phenotype; this emerging field promises to help us better understand disease biology, prognosis, and treatment options. The next 25 years offer remarkable opportunities for informatics and medical imaging together to lead to further advances in both disciplines and to improve health.

Antipsychotic agents used to augment clozapine during long-term inpatient hospitalizations.

INTRODUCTION: Literature assessing effective clozapine augmentation strategies is limited. The aim of this retrospective evaluation was to examine antipsychotics used to augment clozapine and assess whether an augmentation antipsychotic would continue at discharge. METHODS: Demographic, clinical and pharmacy data were collected retrospectively if patients had received clozapine plus an antipsychotic used for augmentation. The dose of the augmentation agent, length of augmentation therapy, and concomitant medications were collected. RESULTS: Of the 49 patients (mean age 45.3±12.1 years), 27 (55%) were male. The mean clozapine dose at discharge was 406.1±121.8 mg. When a first generation antipsychotic (FGA) was selected initially to augment clozapine there was a greater likelihood it would be continued until discharge compared to a second generation antipsychotic (SGA) (78 vs. 50%, OR=3.6, 95% CI 1.03-12.6). FGAs (3.2%) compared to SGAs (35%) were less likely to be discontinued due to a documented lack of benefit when first selected to augment clozapine (OR=16.2, 95% CI 2-131.3). Electroconvulsive therapy plus clozapine was found to be beneficial in several patients (n=14) who failed at least 1 augmentation strategy. DISCUSSION: This real world data suggests that adding an antipsychotic to clozapine is a reasonable approach to those who do not fully respond to clozapine monotherapy. While comparisons of all agents could not be conducted from this small retrospective study, these data suggest FGAs should be investigated in future studies as potential agents to successfully augment clozapine therapy.

Randomized testing of alternative survey formats using anonymous volunteers on the World Wide Web.

Consenting visitors to a health survey Web site were randomly assigned to a "matrix" presentation or an "expanded" presentation of survey response options. Among 4,208 visitors to the site over 3 months, 1,615 (38 percent) participated by giving consent and completing the survey. During a pre-trial period, when consent was not required, 914 of 1,667 visitors (55 percent) participated (odds ratio 1.9, P<0.0001). Mean response times were 5.07 minutes for the matrix format and 5.22 minutes for the expanded format (P=0.16). Neither health status scores nor alpha reliability coefficients were substantially influenced by the survey format, but health status scores varied with age and gender as expected from U.S. population norms. In conclusion, presenting response options in a matrix format may not substantially speed survey completion. This study demonstrates a method for rapidly evaluating interface design alternatives using anonymous Web volunteers who have provided informed consent.

A Bayesian network for diagnosis of primary bone tumors.

The authors developed a Bayesian network to differentiate among five benign and five malignant neoplasms of the appendicular skeleton using the patient's age and sex and 17 radiographic characteristics. In preliminary evaluation with physicians in training, the model identified the correct diagnosis in 19 cases (68%), and included the correct diagnosis among the two most probable diagnoses in 25 cases (89%). Bayesian networks can capture and apply knowledge of primary bone neoplasms. Further testing and refinement of the model are underway.

Design and implementation of an Internet-based health information resource.

In support of its institutional mission of community service, the Medical College of Wisconsin created a digital library of health information. This Internet-based resource, named MCW HealthLink, provides health news and information geared to patients, their family members, and the community at large. Information is delivered through a World Wide Web site and an e-mail newsletter. The web and e-mail components are integrated to maximize their utility to readers. This article describes the philosophy, design and implementation of the resource. Ongoing efforts include development of a health information resource tailored to the needs of individual readers.

Potential use of extensible markup language for radiology reporting: a tutorial.

One of the main goals of radiology is to communicate imaging information to aid in patient management. Information standards can facilitate communication and help realize this goal. Extensible Markup Language (XML) is a new information transmission standard that was developed to meet the growing need for robust, large-scale World Wide Web applications. XML notation provides a compact document representation scheme that allows radiology reports to be transmitted over the Web as universally understandable, self-defining documents. XML documents can include a report-specific document type definition (DTD) that defines the allowable data fields and values. XML may also be used to generate data entry forms for radiology reporting and help physicians improve the efficiency of the reporting process. XML documents can be used to store reporting results directly, thus allowing pertinent data to be shared on the Web. An XML-based approach can allow users to link information to entities outside the information systems of a given institution. XML-based methods and applications have the potential to promote development of robust radiology reporting systems and integration with broader, enterprise-wide information systems.

Promoting the online use of radiology appropriateness criteria.

Radiology appropriateness criteria and practice guidelines seek to promote the cost-effective use of radiology procedures and interventions and can be most useful when integrated with electronic patient records and order-entry systems. The task of translating practice guidelines into computer-based formats can highlight deficiencies and lead to revisions that make them more useful. Computer-based practice guidelines can include additional didactic material, such as images, videos, sounds, simulations, and links to bibliographic databases. Given patient data, information systems can select the most appropriate intervention automatically; some systems can function autonomously. Knowledge representation schemes can make appropriateness criteria available across a wide variety of computer platforms. Internet-based tools can allow developers to collaborate across an institution or around the globe. Information systems can bring appropriateness criteria to physicians at the point of care. The use of standardized approaches is important to ensure that appropriateness criteria reach the broadest possible audience and that such efforts can be incorporated easily into automated systems.

Applicability of American College of Radiology appropriateness criteria in a general internal medicine clinic.

OBJECTIVE: The American College of Radiology (ACR) Appropriateness Criteria for Imaging and Treatment Decisions are intended to help radiologists and referring physicians use imaging procedures appropriately and cost-effectively, but these criteria have not undergone empirical testing. To assess how readily the ACR appropriateness criteria can be applied to clinical practice, we retrospectively examined their applicability in a general ambulatory care setting. MATERIALS AND METHODS: From all requests during an 8-month period for noninterventional CT, sonography, MR imaging, and nuclear medicine imaging procedures received from a general internal medicine clinic, we excluded cases for which relevant clinic notes were unavailable or incomplete. Three experienced radiologists classified cases by consensus, using data from radiology requests and clinic notes. Cases were classified as a "complete match" if the features matched a clinical condition and variant included in the ACR appropriateness criteria; as a "partial match" if the features matched a clinical condition but did not match all features of a variant; or as "unmatched." RESULTS: Of 316 cases, there were 202 complete matches (64%) and 37 partial matches (12%). Of the 77 unmatched cases (24%), 14 pertained to asymptomatic patients. CONCLUSION: The ACR appropriateness criteria could be applied to 76% of the imaging procedure requests that we received from a general internal medicine clinic. These results suggest that the ACR appropriateness criteria can be applied to diagnostic imaging in a general ambulatory-care setting.

Standard Generalized Markup Language for self-defining structured reports.

Structured reporting is the process of using standardized data elements and predetermined data-entry formats to record observations. The Standard Generalized Markup Language (SGML; International Standards Organization (ISO) 8879:1986)--an open, internationally accepted standard for document interchange was used to encode medical observations acquired in an Internet-based structured reporting system. The resulting report is self-documenting: it includes a definition of its allowable data fields and values encoded as a report-specific SGML document type definition (DTD). The data-entry forms, DTD, and report document instances are based on report specifications written in a simple, SGML-based language designed for that purpose. Reporting concepts can be linked with those of external vocabularies such as the Unified Medical Language System (UMLS) Metathesaurus. The use of open standards such as SGML is an important step in the creation of open, universally comprehensible structured reports.

Superior temporal gyrus and the course of early schizophrenia: progressive, static, or reversible?

Accumulating evidence suggests alterations in brain structure, especially in the prefrontal and temporal cortex, in schizophrenia. Previous studies examining the progression of brain structural alterations in schizophrenia have led to conflicting results. Morphometric studies of the superior temporal gyrus (STG) volumes were conducted in a series of neuroleptic-naive first-episode schizophrenic patients, non-schizophrenic first-episode psychotic patients, and matched healthy controls. Three-dimensional MRI scans were carried out in these subjects before and after one year of treatment. Volume reductions were seen at baseline in the left superior temporal gyrus (adjusted for intracranial volume) in both of the patient groups. Pretreatment illness duration was inversely related to the volume of the left superior temporal gyrus; this relation was confined to males. One-year follow-up MRI investigations in a smaller subset of patients suggested that the STG volume reductions may be reversible. No significant changes were noted in the STG volumes in matched healthy controls who were also scanned at baseline as well as at one-year follow-up. These findings have implications for understanding the nature of the neuropathological processes in early schizophrenia, as well as the potential impact of early treatment.

An Internet-based ontology editor for medical appropriateness criteria.

Appropriateness criteria and practice guidelines seek to promote the cost-effectiveness use of medical interventions, and can be most useful when integrated with computer-based patient records and order-entry systems. Building an abstract model (ontology) of appropriateness criteria can require considerable effort among investigators at geographically dispersed institutions. To facilitate the construction and maintenance of ontologies for clinical appropriateness criteria, the author developed an Internet-based system for viewing and editing the knowledge model. The system, called NEON (Network-based Editor for ONtologies), uses the World Wide Web as a platform-independent user interface. NEON allows users to edit the indexing terms and the semantic network that form the ontology for a set of appropriateness criteria. Ontologies built using the system can be imported and exported using an open, internationally standardized format based on the Standard Generalized Markup Language (SGML).

A multipurpose model of radiology appropriateness criteria.

RATIONALE AND OBJECTIVES: Appropriateness criteria and practice guidelines are being developed in attempts to improve the cost-effectiveness of medical care. The authors sought to make a set of radiology appropriateness criteria usable for education, computer-based decision support, and utilization review. MODEL DEVELOPMENT: Sixty clinical conditions from the American College of Radiology's appropriateness criteria were selected. To make the information more suitable for automation, the names of the imaging procedures were standardized. Indexing terms were assigned to identify clinical conditions and to distinguish between each condition's variants. Semantic relationships between terms were defined. Information about the clinical conditions and variants, radiologic procedures, indexing terms, and relationships was encoded into a standardized language for document interchange. IMPLEMENTATION: The 1,956 rows in the appropriateness criteria tables for the 60 clinical conditions and their 212 variants were mapped into references to 163 distinct imaging procedures. The system's knowledge base included 301 indexing terms and 569 additional terms. CONCLUSION: Radiology appropriateness criteria can be indexed and encoded into a form that facilitates their use and interchange. The use of open, internationally accepted standards is an important step to make such knowledge portable and suitable for integration with evolving computer-based patient record systems.

Self-documenting structured reports using open information standards.

Structured reporting systems use standardized data elements and predetermined data-entry formats to record observations. This article describes a system for structured data entry and reporting that generates reports encoded in the Standard Generalized Markup Language (SGML), an open, internationally accepted standard for document interchange. The structured report is self-documenting: it includes a definition of its allowable data field and values encoded as a report-specific SGML document type definition (DTD). By linking its reporting concepts with those of external vocabularies such as the UMLS Metathesaurus, this system can create open, universally comprehensible structured reports.

Extensible markup language (XML) in health care: integration of structured reporting and decision support.

The Extensible Markup Language (XML) is a newly adopted Internet protocol for data interchange designed to bring the key features of the Standard Generalized Markup Language (SGML; ISO 8879:1986)--extensibility, complex structures, and validation--to the World Wide Web. In this paper, we describe an architecture that uses XML to mediate between disparate client-server systems for structured reporting and decision support.

A generalized language for platform-independent structured reporting.

Structured reporting systems allow health-care workers to record observations using predetermined data elements and formats. The author developed the Data-entry and Reporting Markup Language (DRML) to provide a generalized representational language for describing concepts to be included in structured reporting applications. DRML is based on the Standard Generalized Markup Language (SGML), an internationally accepted standard for document interchange. The use of DRML is demonstrated with the SPIDER system, which uses public-domain internet technology for structured data entry and reporting. SPIDER uses DRML documents to create structured data-entry forms, outline-format textual reports, and datasets for analysis of aggregate results. Applications of DRML include its use in radiology results reporting and a health status questionnaire. DRML allows system designers to create a wide variety of clinical reporting applications and survey instruments, and helps overcome some of the limitations seen in earlier structured reporting systems.

Appropriateness of imaging procedure requests: do radiologists agree?

OBJECTIVE: We explored the agreement among radiologists in their evaluation of the appropriateness of individual requests for imaging procedures. MATERIALS AND METHODS: We reviewed 318 noninterventional CT, sonographic, MR imaging, and nuclear medicine procedures ordered at a general internal medicine clinic during 8 months in 1995. Five subspecialty radiologists used data from the radiology request from and clinic notes to independently rate the appropriateness of each requested imaging procedure on a four-point scale. The radiologists were unaware of the results achieved by each procedure. Each case was reviewed by at least three radiologists, of whom at least one had relevant subspecialty expertise. Agreement among radiologists was analyzed using Cohen's kappa statistic and weighted kappa statistics and Cronbach's alpha statistic. RESULTS: Nonchance agreement (kappa) was .19 +/- .05; weighted kappa was .24 +/- .05. Interrater agreement was significantly greater than that expected from chance alone (p < .01). The composite score, defined as the average of the radiologists' scores for each case, showed moderate reliability, as evidenced by a value for Cronbach's alpha of 70. CONCLUSION: In the absence of explicit criteria, we found modest but statistically significant agreement among radiologists about the appropriateness of individual requests for imaging procedures. The disagreement among radiologists highlights the importance of developing well-reasoned, explicit criteria by which to judge the appropriateness of diagnostic radiology procedures. Further study is needed to elucidate the relationship between appropriateness and actual patient outcomes.

BANTER: a Bayesian network tutoring shell.

We present an educational tool for bringing the information contained in a Bayesian network to the end user in an easily intelligible form. The BANTER shell is designed to tutor users in evaluation of hypotheses and selection of optimal diagnostic procedures. BANTER can be used with any Bayesian network containing nodes that can be classified into hypotheses, observations, and diagnostic procedures. The system enables one to present various types of queries to the network, to test one's ability to select optimal diagnostic procedures, and the request explanations. We describe the system's capabilities by illustrating how it functions with two structurally different network models of real-world medical problems.

Construction of a Bayesian network for mammographic diagnosis of breast cancer.

Bayesian networks use the techniques of probability theory to reason under uncertainty, and have become an important formalism for medical decision support systems. We describe the development and validation of a Bayesian network (MammoNet) to assist in mammographic diagnosis of breast cancer. MammoNet integrates five patient-history features, two physical findings, and 15 mammographic features extracted by experienced radiologists to determine the probability of malignancy. We outline the methods and issues in the system's design, implementation, and evaluation. Bayesian networks provide a potentially useful tool for mammographic decision support.

Positive predictive value of clinical suspicion for abdominal aortic aneurysm. Implications for use of ultrasonography.

To measure the positive predictive value (PPV) of clinical suspicion of abdominal aortic aneurysm (AAA), as confirmed by ultrasonography, we reviewed the records of 343 patients at a university medical center referred to ultrasonography for newly suspected AAA. Positive predictive value was 11.1% for large aneurysms of at least 5.0 cm and 18.7% for aneurysms of at least 3.5 cm, and was higher for men and older patients. For patients under 50 years of age, PPV was only 2.6%. Ultrasonography for clinically suspected AAA has a low positive predictive yield, particularly for men under age 50 and for women.