Table 1.

Descriptions of how study aims, subjects, and metrics vary for different phases of imaging biomarker assessment

Study phase	General study aim	Specific research question	Study subjects	Metrics measured
Preclinical	Radiomic biomarker discovery	Is the biomarker associated with the target pathology?	Patients with severe disease and with no disease. Phantoms	Technical validation (precision, repeatability, reproducibility, etc.)
Translational	Can the biomarker identify/predict disease?	Can the biomarker distinguish/predict diseased from normal patients?	Patients with severe disease and with no disease	Technical validation: (precision, repeatability, reproducibility, ROC AUC, etc.)
Early clinical: single-centre setting	Is the biomarker clinically useful?	Can the biomarker distinguish/predict all stages of the target disease and differentiate from patients without the disease (but who may have alternative diagnoses)?	Patients with all stages of the target disease. Patients seen in clinic but without the target disease	Diagnostic/predictive accuracy (sensitivity, specificity, detection rates, PPV, NPV, etc.)
				Diagnostic test impact (does the result impact on patient management?)
Late clinical: multi-centre setting	Is the biomarker generalisable and affordable?	Is the biomarker clinically useful and cost-effective in different centres and healthcare settings?	Representative patients of all who would receive biomarker test, with and without disease	Diagnostic/predictive accuracy
				Diagnostic test impact
				Cost-effectiveness