Table of Contents CHAPTER 7: ORGANIZING OBSERVATIONS: DATA MODELS 299 Data Models for Representing Medical Data Spatial Data Models.… 300 Spatial Relationships and Reasoning 01 Anatomical and Imaging-based Models.. Temporal Data Models Temporal Relationships and Reasoning Some Open Issues in Temporal Modeling Clinically-oriented Views 16 Alternative Views and Application Domains 318 Discussion and Applications. 319 A Phenomenon-centric View: Supporting Investigation What is a Mass? An Exercise in Separating Observations from Inferences. PCDM Core Entities Implementing the PCDM. PART IV TOWARD MEDICAL DECISION MAKING 333 CHAPTER 8: DISEASE MODELS. PART I: GRAPHICAL MODELS ....................4..335 Uncertainty and Probability .. Why Probabilities? 335 Laws of Probability: A Brief Review 337 Probability and Change Graphical Models 340 Graph Theory…… Graphs and Probabilities Representing Time… Graphs and Causation Bayesian Belief Networks in Medicine Belief Network Construction: Building a Disease Model Causal Infered Causal Models, Interventions and counterfactuals. 351 Latent Projections and their Causal Interpretation. Discussion and Applications 359 Building Belief and Causal Networks: Practical Considerations Accruing Sufficient Patient Data Handling Uncertainty in Data.... andling Selection Bias References 365 CHAPTER 9: DISEASE MODELS, PART II: QUERYING APPLICATIONS 371 Exploring the Network: Queries and Evaluation.. 371 Inference: Answering Queries 371 Belief Updating 372 Abductive Reasoning
Table of Contents xxi CHAPTER 7: ORGANIZING OBSERVATIONS: DATA MODELS ....................................299 Data Models for Representing Medical Data.................................................... 299 Spatial Data Models........................................................................................... 3 Spatial Representations .............................................................................................. 300 Spatial Relationships and Reasoning........................................................................... 301 Anatomical and Imaging-based Models...................................................................... 304 Temporal Data Models ...................................................................................... 308 Representing Time ...................................................................................................... 308 Temporal Relationships and Reasoning ...................................................................... 313 Some Open Issues in Temporal Modeling................................................................... 315 Clinically-oriented Views ................................................................................... 316 Alternative Views and Application Domains ............................................................... 318 Discussion and Applications............................................................................. 319 A Phenomenon-centric View: Supporting Investigation ................................... 319 What is a Mass? An Exercise in Separating Observations from Inferences................. 320 PCDM Core Entities..................................................................................................... 323 Implementing the PCDM............................................................................................. 325 References....................................................................................................... 326 PART IV TOWARD MEDICAL DECISION MAKING.................................................333 CHAPTER 8: DISEASE MODELS, PART I: GRAPHICAL MODELS .................................335 Uncertainty and Probability............................................................................. 335 Why Probabilities?............................................................................................. 335 Laws of Probability: A Brief Review............................................................................. 337 Probability and Change ............................................................................................... 339 Graphical Models............................................................................................... 340 Graph Theory .............................................................................................................. 340 Graphs and Probabilities............................................................................................. 341 Representing Time ...................................................................................................... 343 Graphs and Causation ................................................................................................. 345 Bayesian Belief Networks in Medicine .............................................................. 346 Belief Network Construction: Building a Disease Model............................................. 347 Causal Inference ................................................................................................ 351 Causal Models, Interventions, and Counterfactuals ................................................... 351 Latent Projections and their Causal Interpretation..................................................... 354 Identification............................................................................................................... 355 Discussion and Applications............................................................................. 359 Building Belief and Causal Networks: Practical Considerations ........................ 360 Accruing Sufficient Patient Data ................................................................................. 361 Handling Uncertainty in Data...................................................................................... 363 Handling Selection Bias............................................................................................... 364 References....................................................................................................... 365 CHAPTER 9: DISEASE MODELS, PART II: QUERYING & APPLICATIONS ......................371 Exploring the Network: Queries and Evaluation............................................... 371 Inference: Answering Queries ........................................................................... 371 Belief Updating ........................................................................................................... 372 Abductive Reasoning................................................................................................... 377 00
Table of Contents Inference on Relational Models Diagnostic, Prognostic, and Therapeutic Questions.. Evaluating BBNs Predictive Power 384 Interacting with Medical BBNs/Disease Models Defining and Exploring Structure Expressing Queries and Viewing Results Discussion and Applications 389 aive Bayes… Imaging Applications Querying and Problem-centric BBN Visualization Visual Query Interface.. AneurysmDB References CHAPTER 10: EVALUATION Biostatistics and Study Design: A Primer. Statistical Concepts 403 Confidence Intervals Significance and Hypothesis Testing Assessing Errors and Performance udy Design Types of Study Designs 410 Study Variable Selection and Population Definition 412 Population Size: Sample Size and Power Calculations 413 Study Bias and Error 41 Decision Making 418 Regression Analysis 418 Informatics Evaluation 420 Evaluating Information Retrieval Systems Information nee 421 Relevance Evaluation metrics… 424 Medical Content-based Image Retrieval Evaluation Assessing Usabilit Evaluation Techniques............. Discussion.… 433 References 434 NDEX
xxii Table of Contents Inference on Relational Models .................................................................................. 380 Diagnostic, Prognostic, and Therapeutic Questions.................................................... 381 Evaluating BBNs................................................................................................. 383 Predictive Power ......................................................................................................... 383 Sensitivity Analysis ...................................................................................................... 384 Interacting with Medical BBNs/Disease Models ............................................... 386 Defining and Exploring Structure ................................................................................ 386 Expressing Queries and Viewing Results..................................................................... 387 Discussion and Applications............................................................................. 389 Naïve Bayes ....................................................................................................... 390 Imaging Applications ......................................................................................... 391 Querying and Problem-centric BBN Visualization ............................................. 392 Visual Query Interface ................................................................................................ 392 AneurysmDB ............................................................................................................... 396 References....................................................................................................... 398 CHAPTER 10: EVALUATION.............................................................................403 Biostatistics and Study Design: A Primer .......................................................... 403 Statistical Concepts ........................................................................................... 403 Confidence Intervals ................................................................................................... 403 Significance and Hypothesis Testing ........................................................................... 404 Assessing Errors and Performance.............................................................................. 407 Study Design ...................................................................................................... 409 Types of Study Designs ............................................................................................... 410 Study Variable Selection and Population Definition ................................................... 412 Population Size: Sample Size and Power Calculations ................................................ 413 Study Bias and Error.................................................................................................... 416 Meta-analysis.............................................................................................................. 417 Decision Making ................................................................................................ 418 Regression Analysis..................................................................................................... 418 Decision Trees............................................................................................................. 419 Informatics Evaluation..................................................................................... 420 Evaluating Information Retrieval Systems......................................................... 421 Information Needs ...................................................................................................... 421 Relevance.................................................................................................................... 423 Evaluation Metrics ...................................................................................................... 424 Medical Content-based Image Retrieval Evaluation ................................................... 426 Assessing Usability............................................................................................. 428 Evaluation Techniques ................................................................................................ 429 Discussion........................................................................................................ 433 References....................................................................................................... 434 INDEX ........................................................................................................439
PART I Performing the Imaging Exam Wherein an introduction to medical imaging informatics(MIl)is provided; as is a review of the current state of clinical medical imaging and its use in understanding the human condition and disease. For new students and the informatician with a minimal background in medical imaging and clinical applications, these chapters help provide a basis for understanding the role of mll, the present needs of physicians and researchers dealing with images, and the future directions of this discipline Chapter 1-Introduction Chapter 2-A Primer on Imaging Anatomy and Physiology
PART I Performing the Imaging Exam Wherein an introduction to medical imaging informatics (MII) is provided; as is a review of the current state of clinical medical imaging and its use in understanding the human condition and disease. For new students and the informatician with a minimal background in medical imaging and clinical applications, these chapters help provide a basis for understanding the role of MII, the present needs of physicians and researchers dealing with images, and the future directions of this discipline. Chapter 1 – Introduction Chapter 2 – A Primer on Imaging Anatomy and Physiology
Chapter 1 Introduction ALEX A.T. BUL RICKY K. TAIRA, AND HOOSHANG KANGARLOO M edical imaging informatics is the rapidly evolving field that combines biomedical informatics and imaging, developing and adapting core meth ods in informatics to improve the usage and application of imaging in healthcare; and to derive new knowledge from imaging studies. This chapter intro duces the ideas and motivation behind medical imaging informatics. Starting with an illustration of the importance of imaging in today's patient care, we demonstrate imaging informatics' potential in enhancing clinical care and biomedical research From this perspective, we provide an example of how different aspects of medical imaging informatics can impact the process of selecting an imaging protocol. To help readers appreciate this growing discipline, a brief history is given of different efforts that have contributed to its development over several decades, leading to its current challenges What is Medical Imaging Informatics? Two revolutions have changed the nature of medicine and research: medical imaging and biomedical informatics. First, medical imaging has become an invaluable tool in modern healthcare, often providing the only in vivo means of studying disease and the human condition. Through the advances made across different imaging modalities majors insights into a range of medical conditions have come about, elucidating mat ters of structure and function. Second, the study of biomedical informatics concerns itself with the development and adaptation of techniques from engineering, computer science,and other fields to the creation and management of medical data and knowl edge. Biomedical informatics is transforming the manner by which we deal and think with (large amounts of) electronic clinical data. Medical imaging informatics is the discipline that stands at the intersection of biomedical informatics and imaging, bridg ing the two areas to further our comprehension of disease processes through the unique lens of imaging and from this understanding, improve clinical care Beyond the obvious differences between images and other forms of medical data, the very nature of medical imaging set profound challenges in automated understanding and management. While humans can learn to perceive patterns in an image -much as a radiologist is trained -the nuances of deriving knowledge from an image still defy the best algorithms, even with the significant strides made in image processing and computer vision. Imaging informatics research concerns itself with the full spectrum of low-level concepts(e.., image standardization; signal and image processing) to higher-level abstractions(e.., associating semantic meaning to a region in an image visualization and fusion of images)and ultimately, applications and the derivation of new knowledge from imaging. Notably, medical imaging informatics addresses not only the images themselves, but encompasses the associated data to understand the context of the imaging study: to document observations; and to correlate and reach new conclusions about a disease and the course of a medical problem. AA.T d R.K. Taira(eds ) Medical Imaging Informatics, DOI 10.1007/978-1-4419-0385-3_1. Springer Science Business Media, LLC 2010
A.A.T. Bui and R.K. Taira (eds.), Medical Imaging Informatics, 3 DOI 10.1007/978-1-4419-0385-3_1, © Springer Science + Business Media, LLC 2010 Chapter 1 Introduction ALEX A.T. BUI, RICKY K. TAIRA, AND HOOSHANG KANGARLOO edical imaging informatics is the rapidly evolving field that combines biomedical informatics and imaging, developing and adapting core methods in informatics to improve the usage and application of imaging in healthcare; and to derive new knowledge from imaging studies. This chapter introduces the ideas and motivation behind medical imaging informatics. Starting with an illustration of the importance of imaging in today’s patient care, we demonstrate imaging informatics’ potential in enhancing clinical care and biomedical research. From this perspective, we provide an example of how different aspects of medical imaging informatics can impact the process of selecting an imaging protocol. To help readers appreciate this growing discipline, a brief history is given of different efforts that have contributed to its development over several decades, leading to its current challenges. What is Medical Imaging Informatics? Two revolutions have changed the nature of medicine and research: medical imaging and biomedical informatics. First, medical imaging has become an invaluable tool in modern healthcare, often providing the only in vivo means of studying disease and the human condition. Through the advances made across different imaging modalities, majors insights into a range of medical conditions have come about, elucidating matters of structure and function. Second, the study of biomedical informatics concerns itself with the development and adaptation of techniques from engineering, computer science, and other fields to the creation and management of medical data and knowledge. Biomedical informatics is transforming the manner by which we deal and think with (large amounts of) electronic clinical data. Medical imaging informatics is the discipline that stands at the intersection of biomedical informatics and imaging, bridging the two areas to further our comprehension of disease processes through the unique lens of imaging; and from this understanding, improve clinical care. Beyond the obvious differences between images and other forms of medical data, the very nature of medical imaging set profound challenges in automated understanding and management. While humans can learn to perceive patterns in an image – much as a radiologist is trained – the nuances of deriving knowledge from an image still defy the best algorithms, even with the significant strides made in image processing and computer vision. Imaging informatics research concerns itself with the full spectrum of low-level concepts (e.g., image standardization; signal and image processing) to higher-level abstractions (e.g., associating semantic meaning to a region in an image; visualization and fusion of images) and ultimately, applications and the derivation of new knowledge from imaging. Notably, medical imaging informatics addresses not only the images themselves, but encompasses the associated data to understand the context of the imaging study; to document observations; and to correlate and reach new conclusions about a disease and the course of a medical problem. M
A.A.T. Bui et al The Process of Care and the role of Imaging From a high-level perspective, the healthcare process can be seen in of thr clinical questions(Fig. 1. 1), each related to aspects of the scientific od For a given patient, a physician has to: 1 )ascertain what is wrong with the t (identify the problem, develop a hypothesis); 2 )determine the seriousness of a patient,'s condi tion by performing diagnostic procedures(experiment); and 3)after obtaining all needed information, interpret the results from tests to reach a final diagnosis and initiate therapy (analyze and conclude). At each point, medical imaging takes on a critical role What is wrong? Patient presentation, for the most part, is relatively subjective For example, the significance of a headache is usually not clear from a patient,'s description (e.g, my head throbs). Imaging plays a major role in objectifying clinical presentations (e.., is the headache secondary to a brain tumor, intra cranial aneurysm, or sinusitis? and is an optimal diagnostic test in many cases to relate symptoms to etiology. In addition, when appropriately recorded, imaging serves as the basis for shared communication between healthcare providers detailing evidence of current and past medical findings 2. How serious is it? For many conditions, the physical extent of disease is visually apparent through imaging, allowing us to determine how far spread a problem has become(e.g, is it confined to a local environment or is it systemic? ) Moreover, imaging is progressively moving from qualitative to quantitative assessment. Already, we use imaging to document physical state and the severity of disease tumor size in cancer patients; dual energy x-ray absorptiometry(DXA) scores in osteoporosis; cardiothoracic ratios; arterial blood flow assessment based on Doppler ultrasound; and coronary artery calcification scoring are all rudimentary quantitative imaging techniques that further characterize biophysical phenomena. 3. What to do? Treatment is contingent on an individuals response: if a given drug or intervention fails to have the desired effect, a new approach must be taken to resolve the problem. For many diseases, response assessment is done through imaging: baseline, past, and present studies are compared to deduce overall behavior. By way of illustration, many of today's surgical procedures are assessed on a follow-up imaging study, and the effects of chemotherapy are tracked over time(eg, is the tumor getting smaller?). Additionally, contemporary image-guided interventional techniques are opening new avenues of treatment. As the ubiquity and sophistication of imaging grows, methods are needed to fully real- ize its potential in daily practice and in the full milieu of patient care and medical research. The study of medical imaging informatics serves this function. analyze results how serious is it Figure 1. 1: The process of care can be roughly summarized in three stages: 1) what is wrong, which entails identifying the problem and establishing a differential diagnosis 2) how serious is it, which involves testing the differential diagnosis and determining the extent of the problem; and 3) what to do, which based on analysis of test results concludes with a treatment decision
4 A.A.T. Bui et al. The Process of Care and the Role of Imaging 1. What is wrong? Patient presentation, for the most part, is relatively subjective. For example, the significance of a headache is usually not clear from a patient’s description (e.g., my head throbs). Imaging plays a major role in objectifying clinical presentations (e.g., is the headache secondary to a brain tumor, intracranial aneurysm, or sinusitis?) and is an optimal diagnostic test in many cases to relate symptoms to etiology. In addition, when appropriately recorded, imaging serves as the basis for shared communication between healthcare providers, detailing evidence of current and past medical findings. 2. How serious is it? For many conditions, the physical extent of disease is visually apparent through imaging, allowing us to determine how far spread a problem has become (e.g., is it confined to a local environment or is it systemic?). Moreover, imaging is progressively moving from qualitative to quantitative assessment. Already, we use imaging to document physical state and the severity of disease: tumor size in cancer patients; dual energy x-ray absorptiometry (DXA) scores in osteoporosis; cardiothoracic ratios; arterial blood flow assessment based on Doppler ultrasound; and coronary artery calcification scoring are all rudimentary metrics that quantify disease burden. On the horizon are more sophisticated quantitative imaging techniques that further characterize biophysical phenomena. 3. What to do? Treatment is contingent on an individual’s response: if a given drug or intervention fails to have the desired effect, a new approach must be taken to resolve the problem. For many diseases, response assessment is done through imaging: baseline, past, and present studies are compared to deduce overall behavior. By way of illustration, many of today’s surgical procedures are assessed on a follow-up imaging study; and the effects of chemotherapy are tracked over time (e.g., is the tumor getting smaller?). Additionally, contemporary image-guided interventional techniques are opening new avenues of treatment. As the ubiquity and sophistication of imaging grows, methods are needed to fully realize its potential in daily practice and in the full milieu of patient care and medical research. The study of medical imaging informatics serves this function. Figure 1.1: The process of care can be roughly summarized in three stages: 1) what is wrong, which entails identifying the problem and establishing a differential diagnosis; 2) how serious is it, which involves testing the differential diagnosis and determining the extent of the problem; and 3) what to do, which based on analysis of test results, concludes with a treatment decision. From a high-level perspective, the healthcare process can be seen in terms of three clinical questions (Fig. 1.1), each related to aspects of the scientific method. For a given patient, a physician has to: 1) ascertain what is wrong with the patient (identify the problem, develop a hypothesis); 2) determine the seriousness of a patient’s condition by performing diagnostic procedures (experiment); and 3) after obtaining all needed information, interpret the results from tests to reach a final diagnosis and initiate therapy (analyze and conclude). At each point, medical imaging takes on a critical role: