Evaluation of diagnostic systems

Evaluation of Diagnostic Systems: Methods from Signal Detection Theory addresses the many issues that arise in evaluating the performance of a diagnostic system, across the wide range of settings in which such systems are used. These settings include clinical medicine, industrial quality control, environmental monitoring and investigation, machine and metals inspection, military monitoring, information retrieval, and crime investigation. The book is divided into three parts encompassing 11 chapters that emphasize the interpretation of diagnostic visual images by human observers. The first part of the book describes quantitative methods for measuring the accuracy of a system and the statistical techniques for drawing inferences from performance tests. The subsequent part covers study design and includes a detailed description of the form and conduct of an image-interpretation test. The concluding part examines the case study of a medical imaging system that serves as an example of both simple and complex applications. In this part, three mammographic modalities are used: industrial film radiography, low-dose film radiography, and xeroradiography. The case study focuses on the overall reliability of accuracy indices made by its main components, that is, the variabilities across cases, across readers, and within individual readers. The supplementary texts provide study protocols, a computer program for processing test results, and an extensive list of references that will assist the reader in applying those evaluative methods to diagnostic systems in any setting. This book is of value to scientists and engineers, as well as to applied, quantitative, or experimental psychologists who are engaged in the study of the human processes of discrimination and decision making in either perceptual or cognitive tasks.

Preface

Acknowledgments

Introduction

The General Setting

The General Problem

The Proposed Solution

Four Types of Evaluation

General Issues in Test Design

Some Application Vignettes

Organization of This Book

Part I Quantitative Methods

Chapter 1 Fundamentals of Accuracy Analysis

1.1 Two Aspects of Performance

1.2 Analysis of Performance via the ROC Curve

1.3 Often Used, but Inadequate, Indices of Accuracy

1.4 Recommended Indices of Accuracy

1.5 A Computational Example

1.6 A Suitable Accuracy Index Based on One ROC Point

1.7 Indices of the Decision Criterion

1.8 Summary

Chapter 2 Extensions of Accuracy Analysis

2.1 Multiple Alternatives

2.2 Other Interpretations of ROC Data

2.3 Combining Individual Results into a Group or Average ROC

2.4 Summary

Chapter 3 Statistical Design of a Performance Test

3.1 Questions to Be Answered by the Reading Test

3.2 Logistic and Cost Constraints

3.3 General Approach to Statistical Analysis of Accuracy Data

3.4 The Degree of Statistical Power Required

3.5 Some Practical Rules for Study Design

3.6 Summary

Chapter 4 Statistical Treatment of Accuracy Results

4.1 Standard Error of the Accuracy Index

4.2 Confidence Interval of the Accuracy Index

4.3 Statistical Significance of a Difference between Modalities

4.4 Statistical Analysis of True-Positive Probability at a Fixed False-Positive Probability

4.5 Assumptions Underlying the Recommended Statistical Procedure

4.6 Tests Comparing More Than Two Modalities

4.7 Summary

Chapter 5 Forms of Efficacy Analysis

5.1 Cost-Benefit Analysis: A Conceptual Starting Point

5.2 Impediments to Cost-Benefit Analysis

5.3 Tailoring the Analysis to Suit the Situation

5.4 Setting the Efficacy Analysis at an Appropriate Level

5.5 Summary

Part II Experimental Methods

Chapter 6 Elements of Study Design

6.1 Developing Study Goals

6.2 Outlining the End Product

6.3 Planning the Collection of Data

6.4 Summary

Chapter 7 Procedure of a Performance Test

7.1 Case and Image Presentation

7.2 Response Format

7.3 Background Information Given Readers on Each Case

7.4 Timing of Trials and Sessions

7.5 Viewing Environment

7.6 Role of Test Administrator

7.7 Reader Orientation

7.8 Feedback to and from Readers

7.9 Summary

Chapter 8 Drawing the Original Case Sample

8.1 Functions of the Original Case Sample

8.2 Types of Data Collected on Each Case

8.3 Selecting a Representative Case Sample

8.4 Bias in Case Data

8.5 Required Size of the Original Case Sample

8.6 Summary

Chapter 9 Selection of Test Cases

9.1 Number of Cases for Adequate Representation

9.2 Proportional Representation of Case Types

9.3 Screening Out Inadequate Cases

9.4 Checking Representativeness of the Reading-Test Sample

9.5 Summary

Chapter 10 Selection of Test Readers

10.1 Number of Readers for Adequate Representation

10.2 Type of Reader

10.3 Summary

Part III A Methodological Case Study

Chapter 11 Illustrative Evaluation of Imaging Systems

11.1 Case Selection

11.2 Response Format

11.3 Experimental Design

11.4 Test Procedures

11.5 Reader Ratings of Image Quality

11.6 Accuracy Results

11.7 Correlational Analyses

11.8 Obtaining Estimates of Pure Variance and Correlation Terms

11.9 Discussion: Sizing a Final Study from a Pilot Study

Appendixes

Appendix A Sequence of Steps in Evaluation

Appendix B Representativeness of a Performance-Test Sample

Appendix C Illustrative Posttest Questionnaire

Appendix D Computer Program for Accuracy Analysis

Appendix E References to ROC Applications

References

Index