Skip to main content

Table 3 Performance of different models for the prediction of patient-specific quality assurance with different percent dose difference/distance to agreement criteria

From: Integrating plan complexity and dosiomics features with deep learning in patient-specific quality assurance for volumetric modulated arc therapy

Criteria

Models/parameters

AUC (95% CI)

Sensitivity

Specificity

Accuracy

3%/3 mm

PC model

0.914 (0.809–1.000)

0.845

1.000

0.869

D model

0.914 (0.841–0.987)

0.914

1.000

0.918

DL model

0.943 (0.840–1.000)

0.845

1.000

0.853

PC + D model

0.940 (0.844–1.000)

0.845

1.000

0.853

Overall model

0.948 (0.880–1.000)

0.914

1.000

0.918

3%/2 mm

PC model

0.762 (0.569–0.955)

0.875

0.717

0.738

D model

0.783 (0.612–0.955)

0.875

0.679

0.705

DL model

0.849 (0.724–0.974)

1.000

0.566

0.623

PC + D model

0.872 (0.764–0.979)

0.875

0.755

0.771

Overall model

0.890 (0.801–0.980)

1.000

0.755

0.787

2%/2 mm

PC model

0.789 (0.530–1.000)

0.833

0.836

0.836

D model

0.842 (0.709–0.975)

0.833

0.782

0.787

DL model

0.841 (0.702–0.979)

1.000

0.600

0.639

PC + D model

0.915 (0.825–1.000)

0.833

0.927

0.918

Overall model

0.942 (0.856–1.000)

1.000

0.818

0.836

  1. PC plan complexity, D dosiomics, DL deep learning, AUC area under curve