Synopsis

This half-day Educational Course will provide an introduction to statistical methods for imaging studies. The course will focus on concepts of study design and data analysis that are frequently used in imaging studies. Topics to be covered include prospective and retrospective studies, error rates and hypothesis testing, ROC curves, basic sample size calculations, methods for evaluating quantitative imaging biomarkers, and statistical concepts in radiomics analysis.

Part 1: Foundations of Imaging Study Design

This is the first lecture on the foundations of imaging studies. We will present an overview of key concepts of study design, focusing on issues specific to imaging studies. Four building blocks of study design will be described. Some considerations and examples for each building block will be provided. Prospective and retrospective studies will be discussed and the advantages and disadvantages of both presented. We will also define paired and unpaired designs and contrast the two.

• Building blocks of study design
  • Study objectives
    • Examples of how to write study objectives
  • Performing and interpreting imaging tests
  • Reference information
    • Examples
    • What if you do not have a reference standard?
  • Endpoints
    • Definition
    • Examples
• Prospective and retrospective studies
  • Differences
  • Examples
• Paired vs. unpaired study designs
  • Definitions
  • Advantages and disadvantages

Part 3: Statistical Methods for Evaluating the Accuracy of Imaging Tests

This is the third lecture on statistical methods used in imaging studies. We will define standard metrics for imaging tests measured on a binary scale, sensitivity and specificity and predictive values, and explain differences between them. We will focus on Receiver Operating Characteristic (ROC) curve analysis. Using a motivating example of a test measured on a continuous scale, we will show where and when ROC curve analysis is needed. The area under the ROC curve (AUC), commonly used to summarize ROC curves, will be presented. Approaches for comparing ROC curves will be presented.

• Sensitivity, specificity
• Predictive values
• ROC curves
  • Motivating example
  • Sensitivity and specificity shown together with all possible cutpoints
• Construction of an ROC curve
• Area under ROC curve (AUC)
  • Definition
  • Interpretation
  • Example
• Estimation
• Comparing ROC curves

Part 5: Concepts in Quantitative Imaging Biomarker Evaluation

This lecture will provide an overview on how to evaluate quantitative imaging biomarkers. We will discuss the different types of evidence that need to be demonstrated before quantitive imaging biomarkers should be adopted. Different endpoints and metrics will be described. The importance of evaluating precision will be highlighted, and there will be an emphasis on studies and methods for evaluating the precision of quantitative imaging biomarkers. The application of these methods will be illustrated using examples from the literature.

• Definition of a quantitative imaging biomarker
• What are the requirements for a quantitative imaging biomarker to be useful?
  • Analytic validity
  • Clinical validity
  • Clinical usefulness
• Endpoints and metrics used when evaluating quantitative imaging biomarkers
  • For analytic validity
    • Analytic ROC curves, analytic sensitivity and specificity, predictive values, precision
  • For clinical validity
    • Clinical ROC curves, clinical sensitivity and specificity, predictive values, measures of association between the quantitative imaging biomarker and an outcome
  • For clinical usefulness
    • Measures of association between the quantitative imaging biomarker and patient- or society-outcome, cost-effectiveness
• Precision
  • Difference between precision and bias
  • Importance of having a precise biomarker
  • Repeatability vs. reproducibility
  • Methods for evaluating precision
  • Examples of studies

Acknowledgements

References

Obuchowski NA and Gazelle GS (Eds). (2016) Handbook for Clinical Trials of Imaging and Image-Guided Interventions. Hoboken, NJ. Wiley.

Pepe MS. (2003) The Statistical Evaluation of Medical Tests for Classification and Prediction. New York, NY. Oxford University Press.