Table 7 Race prediction performance across different models

From: Generalisable deep learning method for mammographic density prediction across imaging techniques and self-reported race

Model

Accuracy (95% CI)

AUROC (95% CI)

Race prediction model

86.7% (86.0–87.4)

95.1% (94.7–95.5)

Fine-tuned Model A

74.2% (73.3–75.3)

80.8% (80.0–81.9)

Fine-tuned Model B

75.8% (74.8–76.8)

82.8% (81.9–83.6)

Fine-tuned Model C

73.2% (72.2–74.3)

79.2% (78.4–80.3)

  1. First row shows the performance of a model trained specifically for race prediction. Rows two to four show the performance for race prediction of models that were built by fine-tuning the model backbones of breast density prediction models. Density models A and B were originally trained for density prediction using data from a single race group only.
Back to article page