Statistical issues in the comparison of quantitative imaging biomarker algorithms using pulmonary nodule volume as an...

Area of Science:

• Medical Imaging
• Biostatistics
• Radiomics

Background:

• Quantitative imaging biomarkers are crucial for disease diagnosis and monitoring.
• Computer algorithms used for these measurements exhibit varying technical performance characteristics.

Purpose of the Study:

• To illustrate appropriate statistical methods for assessing and comparing bias, precision, and agreement of computer algorithms.
• To compare and contrast study designs and performance metrics for quantitative imaging biomarker studies.

Main Methods:

• Utilized data from three studies involving pulmonary nodules: a small phantom study for repeatability, a large phantom study for bias and reproducibility of tumor volume measurement, and a small clinical study for tumor change assessment.
• Assessed six algorithms' performance in measuring tumor change in the clinical study.
• Compared and contrasted various statistical methods for quantitative imaging biomarker studies.

Main Results:

• Demonstrated metrics for assessing repeatability using a small phantom study.
• Evaluated bias and reproducibility of four algorithms for tumor volume and change measurement in a large phantom study.
• Provided a direct assessment of six algorithms' performance for measuring tumor change in a clinical study.

Conclusions:

• Highlighted the advantages and limitations of common statistical methods for quantitative imaging biomarker studies.
• Emphasized the importance of appropriate statistical evaluation for reliable biomarker measurement.
• Provided practical examples for assessing algorithm performance in medical imaging.