Review Article CME Article
Clinical decision support systems in myocardial perfusion imaging

https://doi.org/10.1007/s12350-014-9857-9Get rights and content

Abstract

Diagnostic imaging is becoming more complicated, physicians are also required to master an ever-expanding knowledge base and take into account an ever increasing amount of patient-specific clinical information while the time available to master this knowledge base, assemble the relevant clinical data, and apply it to specific tasks is steadily shrinking. Compounding these problems, there is an ever increasing number of aging “Baby Boomers” who are becoming patients coupled with a declining number of cardiac diagnosticians experienced in interpreting these studies. Hence, it is crucial that decision support tools be developed and implemented to assist physicians in interpreting studies at a faster rate and at the highest level of up-to-date expertise. Such tools will minimize subjectivity and intra- and inter-observer variation in image interpretation, help achieve a standardized high level of performance, and reduce healthcare costs. Presently, there are many decision support systems and approaches being developed and implemented to provide greater automation and to further objectify and standardize analysis, display, integration, interpretation, and reporting of myocardial perfusion SPECT and PET studies. This review focuses on these systems and approaches.

Introduction

Heart disease continues to be the number one killer of American citizens 1 with about 600,000 fatalities annually. In an attempt to reduce the morbidity and mortality associated with cardiovascular disease, cardiac imaging has expanded in scope and complexity with nuclear cardiology myocardial perfusion imaging (MPI) procedures alone growing from about 7 million in 1999 to 11 million in 2005.2 At the same time, as diagnostic imaging is becoming more complicated, physicians are also required to master an ever-expanding knowledge base and take into account an ever increasing amount of patient-specific clinical information while the time available to master this knowledge base, assemble the relevant clinical data, and apply it to specific tasks (e.g., image interpretation, seeing a new patient) is steadily shrinking. Compounding these problems, there is an ever increasing number of aging “Baby Boomers” who are becoming patients 3 coupled with a declining number of cardiac diagnosticians 4 experienced in interpreting these studies. Without some remedy, the convergence of all these factors will inevitably lead to physician misdiagnosis and patient mismanagement. Hence, it is crucial that decision support tools be developed and implemented to assist physicians in interpreting studies at a faster rate and at the highest level of up-to-date expertise. Such tools will minimize subjectivity and intra- and inter-observer variation in image interpretation, help achieve a standardized high level of performance, and reduce healthcare costs.

Artificial intelligence (AI) is that part of computer science concerned with developing computer systems, including software programs, that exhibit the characteristics associated with intelligence in human behavior.5 In nuclear cardiology, AI techniques have been applied to two very specific tasks: (1) to automate image analysis, and (2) to aid image interpretation.

In the past two decades, image quality and clinical utility of nuclear cardiology studies have both greatly improved with major advancements in automated processing, quantification, and display. AI techniques have brought about the ability to perform completely automatic myocardial perfusion imaging (MPI) motion correction,6,7 reconstruction,8 tomographic oblique reorientation,9,10 quantification,9,11, 12, 13 and high level analysis of the results. Automated comparison to databases of normal myocardial perfusion distributions provides computer-aided diagnostic (CAD) tools to identify hypoperfused myocardium. Automated interpretation of the quantitative information in these systems is available to help train new nuclear cardiologists, aid in the analysis of complex cases, or act as a second opinion. In this issue of the Journal, the article by Johansson et al14 provides an example of the use of a CAD tool to aid in the interpretation of MPI studies. Presently, there are many other decision support systems and approaches being developed and implemented to provide greater automation and to further objectify and standardize analysis, display, integration, interpretation, and reporting of myocardial perfusion SPECT and PET studies. The rest of this review focuses on these systems and approaches.

Data-based methods for identifying a patient’s myocardial perfusion abnormalities from SPECT myocardial perfusion studies were initiated as early as 1985 15 and have been developed and commercialized by investigators at Skane University Hospital (this journal),14 Cedars-Sinai Medical Center,16, 17, 18 University of Michigan,19 Yale University,20 and Emory University.21, 22, 23 These methods all utilize a statistically defined image database of normal (and abnormal) patients to be used as a template for comparison of the MPI images of suspected cardiac patients.24 These methods have been extensively validated and compared16,17,20,21,25,26 and proven to be clinically valuable27 in standardizing and objectifying myocardial perfusion scans.

These systems use quantitative variables of myocardial perfusion compared to a defined threshold to communicate to the diagnostician whether the study is normal or abnormal. The most common variable is the LV extent of hypoperfused myocardium. Usually, if comparison of the patient’s image acquisition to the specific image protocol normal distribution demonstrates that the patient’s stress perfusion distribution exhibits more than 3% of hypoperfused LV myocardium or a summed stress score ≥428,29 the study is deemed to be abnormal. These methods are important both because they are extensively used today as a second opinion and also because the output of these programs can be utilized to form the input to AI methods for more advanced decision support.

Statistical methods can use well-understood mathematical techniques to analyze a reliable clinical or image database to generate an accurate interpretation; however, Gorry30 has pointed out that purely statistical programs have three failings that are impediments to the physicians’ acceptance: (1) the programs have no real “understanding” of their problem area, (2) the programs have no mechanism for “discussing” their knowledge with the user, and (3) the programs have no means for “explaining” (justifying their findings) to physicians. AI approaches such as knowledge-based systems, with its emphasis on knowledge representation, offer a natural environment for implementing the tools needed to address these limitations.31 Over the past several years, AI methods have been investigated in nuclear cardiology as a way to develop such tools. Examples include neural networks32, 33, 34, 35, 36, 37, 38, 39, 40 including the approach by Johansson et al in this issue,14 case-based reasoning41 techniques, and knowledge-based expert systems described below.

Section snippets

Decision Support Systems

A clinical decision support system (CDSS) is an interactive computer software designed to assist physicians and other health professionals with decision making tasks such as determining a diagnosis based on patient data.42 Dr. Robert Hayward of the Centre for Health Evidence has observed that; “Clinical Decision Support systems link health observations with health knowledge to influence health choices by clinicians for improved health care.”42 A clinical decision support system has been coined

Artificial Neural Network (ANN)

ANNs have been developed as an attempt to simulate the highly connected biological system found in the brain through the use of computer hardware and/or software. In the brain, a neuron receives input from many different sources. The neuron integrates all these inputs and if the result is greater than a set threshold, the neuron “fires” (sending a pulse down the nerve to other connected neurons). In the same way, a neural network has nodes (the equivalent of a neuron) that are interconnected

Need for Continuous Knowledge Update

One of the main limitations of CDSS is that once these systems are implemented, validated, and distributed, they start to become obsolete since they have no means to stay current with advancing scientific/clinical knowledge. One example of this limitation is the present article by Johansson et al14 where they determined that the newly developed EXINI CDSS ANN system outperformed the diagnostic accuracy of the 1997 version of PERFEX. Thus, a mechanism must be developed and executed on a

Need for Structured Reporting

A natural approach to display the cardiac imaging clinical results and conclusions of a CDSS is the use of structured reporting. In this context, structured reporting is the process of organizing and communicating data by abstracting and integrating all the evidence during the imaging procedure56 resulting in a clinical report with standardized content and definitions in a clinically relevant, predictable, and coherent format. The use of structured reporting by the CDSS, assures that the

Gaps in Current Knowledge

It should be convincing just from the volume of references in this article on all the various aspects of decisions support systems applied to nuclear cardiology imaging that our field will continue to be one of the first in all of healthcare to benefit from the widespread applications of CDSS. Several important issues related to how to implement these techniques into mainstream nuclear cardiology have been reviewed including types and attributes of techniques available, how to combine

Acknowledgments

EVG and JLK receive royalties from the sale of the Emory Cardiac Toolbox and PERFEX described in this article. The terms of this arrangement have been reviewed and approved by Emory University in accordance with its conflict-of-interest practice. This work was supported in part by NHLBI Grant 5R42HL 106818-03 from the National Institutes of Health.

References (60)

  • AllisonJS et al.

    Artificial neural network modeling of stress single-photon emission computed tomographic imaging for detecting extensive coronary artery disease

    Am J Cardiol

    (2005)
  • HaddadM et al.

    Feasibility analysis of a case-based reasoning system for automated detection of coronary heart disease from myocardial scintigrams

    Artif Intell Med

    (1997)
  • KurganLA et al.

    Knowledge discovery approach to automated cardiac SPECT diagnosis

    Artif Intell Med

    (2001)
  • SachaJP et al.

    Bayesian learning for cardiac SPECT image interpretation

    Artif Intell Med

    (2002)
  • DouglasPS et al.

    ACCF/ACR/AHA/ASE/ASNC/HRS/NASCI/RSNA/SAIP/SCAI/SCCT/SCMR 2008 health policy statement on structured reporting in cardiovascular imaging

    JACC

    (2009)
  • PetrettaM et al.

    Assessment of coronary flow reserve using single photon emission computed tomography with technetium 99 m labeled tracers

    J Nucl Cardiol

    (2008)
  • KochanekKD et al.

    Deaths: Final data for 2009

    Natl Vital Stat Rep

    (2011)
  • IMV Medical Information Division, Inc. 2008 Nuclear medicine market summary report. 2008....
  • KnickmanJR et al.

    The 2030 problem: Caring for aging baby boomers

    Health Serv Res

    (2002)
  • FyeWB

    Cardiology’s workforce shortage: Implication for patient care and research

    Circulation

    (2004)
  • NelsonN

    Principles of artificial intelligence

    (1980)
  • MatsumotoN et al.

    Quantitative assessment of motion artifacts and validation of a new motion-correction program for myocardial perfusion SPECT

    J Nucl Med

    (2001)
  • GermanoG et al.

    Automatic reorientation of 3-dimensional transaxial myocardial perfusion SPECT images

    J Nucl Med

    (1995)
  • MullickR et al.

    Automatic determination of left ventricular orientation from SPECT data

    IEEE Trans Med Imaging

    (1995)
  • Johansson L, Edenbrandt L, Nakajima K, Lomsky M, Svensson S-E, Trägårdh E, et al. Computer-aided diagnosis system...
  • GarciaEV et al.

    Quantification of rotational thallium-201 myocardial tomography

    J Nucl Med

    (1985)
  • Van TrainKF et al.

    Quantitative analysis of tomographic stress thallium-201 myocardial scintigrams: A multicenter trial

    J Nucl Med

    (1990)
  • SharirT et al.

    A new algorithm for the quantitation of myocardial perfusion SPECT. II: Validation and diagnostic yield

    J Nucl Med

    (2000)
  • SlomkaPJ et al.

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

    J Nucl Med

    (1995)
  • DePasqualeE et al.

    Quantitative rotational thallium-201 tomography for identifying and localizing coronary artery disease

    Circulation

    (1988)
  • Cited by (14)

    • Evaluating machine learning models for sepsis prediction: A systematic review of methodologies

      2022, iScience
      Citation Excerpt :

      For example, several features that were common in the intensive care units (ICU) or emergency departments (systolic blood pressure, diastolic blood pressure, heart rate, respiratory rate, and body temperature) were selected in studies by Barton et al. and Mao et al. (Barton et al.,2019; Mao et al.,2018). However, only relying on clinical expertise could lead to strong subjectivity and may overlook some key features (Garcia et al.,2014). Even though the models performed adequately, the outcomes were difficult to be validated by external data; therefore, the applicability of these models is limited.

    • Classification models for SPECT myocardial perfusion imaging

      2020, Computers in Biology and Medicine
      Citation Excerpt :

      It also improves the diagnostic accuracy and reduces the diagnostic time and healthcare costs [3,9,11,12]. It can be used as a second opinion to assist in the analysis of complex cases and to educate inexperienced readers as well [13]. In SPECT MPI, analysis software typically quantifies the disease as the difference between the stress and rest defect sizes obtained by separate comparisons of the stress and rest normal limits, which are obtained from groups of patients retrospectively identified as normal and are specific to the gender of the patients and the camera used [14].

    • Artificial Intelligence in Cardiology: Present and Future

      2020, Mayo Clinic Proceedings
      Citation Excerpt :

      Unlike other disciplines in cardiology, AI techniques have already been incorporated into several routines in nuclear cardiology. Artificial intelligence algorithms have been applied to image processing, allowing the ability to perform completely automatic single-photon emission computed tomography (SPECT) myocardial perfusion imaging (MPI) motion correction, reconstruction, tomographic oblique reorientation, quantification, and high-level analysis of the results.34 Commercialized and US Food and Drug Administration–approved image viewing software programs have incorporated automated comparison to databases of normal myocardial perfusion distributions, providing computer-aided adjunctive diagnostic tools used by expert readers to identify hypoperfused myocardium.

    • Nuclear imaging and artificial intelligence

      2020, Machine Learning in Cardiovascular Medicine
    • Decision Support Tools, Systems, and Artificial Intelligence in Cardiac Imaging

      2018, Canadian Journal of Cardiology
      Citation Excerpt :

      Weighting factors of each variable are dependent on the importance of the category. This is a continuous iterative process of changing the connection weights with different combinations until an optimal model has been achieved and no significant improvement can be made (Fig. 3).9-11 In unsupervised ANN training, only input data are provided and the system is free to find or “assign” class or group associations.

    View all citing articles on Scopus
    View full text