Related Experiment Videos
Statistical issues in analysis of diagnostic imaging experiments with multiple observations per patient
M Gönen1, K S Panageas, S M Larson
1Department of Epidemiology, Memorial Sloan-Kettering Cancer Center, 1275 York Ave, Box 44, New York, NY 10021, USA. gonenm@mskcc.org
Radiology
|November 24, 2001
Summary
Standard statistical methods are inadequate for clustered imaging data. This review introduces simple statistical methods accounting for within-subject correlation in diagnostic imaging studies, preventing incorrect conclusions.
Area of Science:
- Medical Imaging
- Biostatistics
- Radiology
Background:
- Diagnostic imaging studies often involve multiple observations per patient.
- These clustered data exhibit within-subject correlations, violating assumptions of standard statistical methods.
- Examples include cancer metastasis evaluation and cardiac artery segment analysis.
Purpose of the Study:
- To review statistical methods for analyzing clustered data in diagnostic imaging.
- To introduce simple methods that account for within-subject correlation.
- To highlight the risks of using inappropriate statistical techniques.
Main Methods:
- Positron emission tomographic (PET) studies serve as a framework for discussion.
- Introduction of simple statistical methods adaptable to standard tests (e.g., chi-squared, t-tests).
- Illustration with a PET study data analysis.
Main Results:
- Standard statistical methods can lead to incorrect conclusions when applied to clustered imaging data.
- Appropriate methods are necessary to accurately analyze dependent observations within subjects.
- The study demonstrates potential pitfalls with an illustrative PET data analysis.
Conclusions:
- Accurate statistical analysis of clustered diagnostic imaging data requires specialized methods.
- Ignoring within-subject correlation can lead to erroneous study findings.
- Alternative methods are essential for reliable interpretation of multi-observation imaging studies.