Target lesion selection: an important factor causing variability of response classification in the Response
Sebastian Keil1, Alexandra Barabasch, Timm Dirrichs
1From the Departments of *Diagnostic and Interventional Radiology and †Haematology, Oncology, and Stem Cell Transplantation, University of Aachen, Aachen, Germany.
Investigative Radiology
|March 22, 2014
Summary
Reader variability in cancer response assessment stems from inconsistent target lesion selection, not measurement methods. Standardizing lesion choice is crucial for reliable tumor response evaluation using Response Evaluation Criteria in Solid Tumors 1.1.
Area of Science:
- Radiology and Medical Imaging
- Oncology
- Biostatistics
Background:
- Accurate tumor response assessment is critical for guiding cancer treatment decisions.
- The Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 is widely used but can be subject to interpretation variability.
- Factors influencing RECIST 1.1 categorization require systematic analysis to improve reliability.
Purpose of the Study:
- To systematically analyze factors contributing to variability in RECIST 1.1 response categorization.
- To compare manual versus automated and unidimensional versus 3-dimensional size assessments.
- To evaluate the impact of different target lesion selections on response classification.
Main Methods:
- Analysis of contrast-enhanced thoracoabdominal CT scans from 41 female metastatic breast cancer patients.
- Independent interpretation of scans by 3 experienced radiologists.
- Evaluation of response using a computer-assisted diagnosis system for automated size assessment (unidimensional and 3D).
Main Results:
- Moderate overall between-reader agreement (κ = 0.53), with 46% of patients showing diverging response classifications.
- Selection of different target lesions by readers significantly increased disagreement rates (81% disagreement vs. 24% agreement when same lesions chosen).
- Variability due to manual vs. automated or unidimensional vs. volumetric measurements was less impactful than target lesion selection.
Conclusions:
- The primary source of variability in RECIST 1.1 response categorization is the inconsistent selection of target lesions among readers.
- Computer-assisted diagnosis and tumor volumetry can reduce variability from measurement techniques but do not resolve the issue of target lesion selection.
- Standardizing target lesion selection protocols is essential for improving the reliability of tumor response assessment.
