Table 2 Overview of the study designs for the assessment of clinical value in radiology AI (RADAR-1 through RADAR-5)

General description

Study design that analyzes data collected from a population, or a representative subset

Simulations of clinical trials using patient data

Study design where participants are randomly assigned to groups, typically an AI strategy and control group

Study design in which AI-exposed and non-exposed are followed over time for specific outcomes

Type of method/study design

Observational

Experimental

Experimental

Observational

Type of research question

Identification of disease (diagnostic)

Identification of disease (diagnostic)

Explanation (causation) of impact AI as opposed to standard of care

Explanation (causation) of impact AI as opposed to the standard of care or identification of disease (diagnostic)

Time frame

Instantaneous

Instantaneous to longitudinal (simulated over time)

Longitudinal

Longitudinal

Primary outcome

Efficacy of AI in diagnosing conditions (e.g., sensitivity, specificity)

Efficacy of AI in diagnosing conditions (e.g., sensitivity, specificity)

Differences in patient outcomes between treatment and control groups

Differences in patient outcomes between AI and non-AI groups

Example

Cross-sectional study of AI for predicting readmission or death after ICU discharge [45]

IST of digital breast tomosynthesis as a replacement for full-field digital mammography [46]

RCT of decision support algorithm for neonatal seizure recognition [47]

Cohort study of AI solution for referable thoracic abnormalities on chest radiography [48]