Fig. 5: Radiologists in a randomized study performed a test to differentiate real and AI-generated images. | Nature Communications

Fig. 5: Radiologists in a randomized study performed a test to differentiate real and AI-generated images.

From: Evaluating normative representation learning in generative AI for robust anomaly detection in brain imaging

Fig. 5

a Violin plots display average score distributions across metrics like Readiness, Quality, and Health. b Heatmap indicates inter-rater variability in scoring image realness among radiologists, including both experienced ([Exp]) and resident ([Res]). c Bar charts compare mean scores between residents (N = 13) and experienced radiologists (N = 3). The error bars represent the standard deviation (SD) across raters. Individual data points are overlaid to show the score distribution. d The matrix shows Pearson correlation coefficients between proposed metrics (RQI, AHI) and radiologists' judgments of image quality and health. Source Data are provided as a Source Data File.

Back to article page