Inter-Method Performance Study of Tumor Volumetry Assessment on Computed Tomography Test-Retest Data

Area of Science:

• Medical Imaging Analysis
• Quantitative Imaging Biomarkers
• Computational Pathology

Background:

• Tumor volume change is a critical biomarker for cancer diagnosis, treatment planning, and response assessment.
• Semiautomated algorithms for lung tumor volume measurement using computed tomography (CT) data are increasingly utilized.
• Standardization and precision are essential for reliable biomarker application, necessitating evaluation against established profiles like the Quantitative Imaging Biomarker Alliance (QIBA).

Purpose of the Study:

• To evaluate and compare the precision of semiautomated lung tumor volume measurement algorithms using clinical thoracic CT datasets.
• To assess intra-algorithm repeatability and inter-algorithm reproducibility across different algorithms and tumor sizes.
• To inform development approaches and testing requirements for algorithms seeking conformance with the QIBA Computed Tomography Volumetry Profile.

Main Methods:

• A challenge study involving industry and academic participants evaluated multiple semiautomated lung tumor segmentation algorithms.
• Intra-algorithm repeatability and inter-algorithm reproducibility were quantified.
• Linear mixed-effects modeling estimated sources of variability, and segmentation boundaries were compared for optimization.

Main Results:

• Intra-algorithm repeatability varied widely (13% to 100%), generally improving with larger tumor size.
• Inter-algorithm reproducibility was 58% for the top four groups, 70% for groups meeting repeatability, and 84% for all but the lowest performer.
• Segmentation boundary differences were substantial, impacting overall volume and detail; larger tumors benefited from human editing, unlike smaller ones.

Conclusions:

• Nine of 12 algorithms met precision requirements similar to the QIBA Profile, enabling tumor volume change measurement within ±14% for sizes >10 mm.
• No algorithm partition met QIBA interchangeability requirements down to 10 mm, but the best performers met this for tumors >40 mm.
• Significant variability originated from sources independent of the algorithms, highlighting the need for comprehensive quality control in quantitative imaging.