2026 IASLC CT Screening and Early Detection Symposium-WS02.02 - Talk 2: Radiomics as an Imaging Biomarker

Pdf Summary

Harriet Louise Lancaster’s presentation, <strong>“Radiomics as an imaging biomarker,”</strong> explains how radiomics could improve lung cancer detection and management, but also why clinical adoption remains limited.

A key point is that <strong>longitudinal imaging and nodule growth measurement are already powerful biomarkers</strong> in lung cancer screening. Changes in volume and volume doubling time can distinguish malignant from benign nodules over time. Radiomics aims to add value by extracting more information from CT scans than visual assessment alone, including <strong>size, growth, density, morphology, intensity, texture, heterogeneity, and high-order features</strong>.

The presentation describes a radiomics workflow in which a segmented CT nodule is analyzed to generate features and predict a <strong>malignancy probability</strong>. However, Lancaster emphasizes that <strong>external validation is essential</strong>, because model performance can vary depending on scanner type, reconstruction settings, voxel size, segmentation method, and other technical factors.

She also highlights the <strong>translation gap</strong> between research and clinical practice. Although radiomics models may achieve high <strong>AUCs</strong>, that does not necessarily mean they improve patient care. AUC alone can ignore class imbalance, the consequences of false results, and whether outputs are comparable across settings.

The presentation argues that the most useful clinical applications may be where there is an <strong>unmet need</strong>, such as:
- identifying malignancy at the <strong>baseline low-dose CT</strong> scan,
- reducing unnecessary surveillance of indeterminate nodules,
- adding information <strong>before or beyond volume doubling time</strong>,
- and potentially applying radiomics outside standard screening pathways, such as in <strong>incidental pulmonary nodules</strong> with more variable imaging conditions.

The overall message is that <strong>radiomics is promising, but not yet ready for routine use without stronger validation, better clinical targeting, and implementation science</strong>. Lancaster concludes that future progress will likely come from <strong>multimodal AI</strong>, integrating radiomics with molecular and longitudinal imaging biomarkers.

Asset Subtitle

Harriet Lancaster

Meta Tag

Concept Longitudinal Imaging

Concept Nodule Growth Measurement

Concept Radiomics

Concept Imaging Biomarker

Concept CT Scan

Speaker Harriet Lancaster

Keywords

radiomics

imaging biomarker

lung cancer detection

lung cancer screening

CT scans

nodule growth

volume doubling time

malignancy probability

external validation

multimodal AI

Longitudinal Imaging

Nodule Growth Measurement

Radiomics

Imaging Biomarker

CT Scan

Harriet Lancaster