Table 3 Performances of the predictive models in the training and validation cohorts

From: Sub-regional radiomics combining multichannel 2-dimensional or 3-dimensional deep learning for predicting neoadjuvant chemo-immunotherapy response in esophageal squamous cell carcinoma: a multicenter study

Model and metric

AUC (95% CI)

Accuracy

Sensitivity

Specificity

P-value

Training cohort (n = 161)

         

DLRad1

0.910 (0.867–0.954)

0.857 (138/161)

0.891 (41/46)

0.843 (97/115)

Reference

0.043

0.894

0.042

0.036

DLRad2

0.834 (0.769–0.900)

0.819 (132/161)

0.652 (30/46)

0.887 (102/115)

0.043

Reference

0.136

0.885

0.808

2D DL

0.906 (0.855–0.956)

0.720 (116/161)

0.956 (44/46)

0.626 (72/115)

0.894

0.136

Reference

0.143

0.112

3D DL

0.831 (0.763–0.899)

0.789 (127/161)

0.652 (30/46)

0.843 (97/115)

0.042

0.885

0.143

Reference

0.870

Rad_H1

0.823 (0.750–0.896)

0.826 (133/161)

0.782 (36/46)

0.843 (97/115)

0.036

0.808

0.112

0.870

Reference

Internal validation cohort (n = 69)

         

DLRad1

0.829 (0.712–0.945)

0.811 (56/69)

0.800 (16/20)

0.816 (40/49)

Reference

0.383

0.810

0.331

0.376

DLRad2

0.756 (0.643–0.869)

0.696 (48/69)

0.850 (17/20)

0.632 (31/49)

0.383

Reference

0.574

0.875

0.962

2D DL

0.807 (0.672–0.942)

0.797 (55/69)

0.800 (16/20)

0.795 (39/49)

0.810

0.574

Reference

0.500

0.558

3D DL

0.742 (0.614–0.871)

0.637 (44/69)

0.950 (19/20)

0.510 (25/49)

0.331

0.875

0.500

Reference

0.915

Rad_H1

0.752 (0.630–0.874)

0.782 (54/69)

0.650 (13/20)

0.836 (41/49)

0.376

0.962

0.558

0.915

Reference

External validation cohort (n = 41)

         

DLRad1

0.793 (0.652–0.935)

0.804 (33/41)

0.833 (10/12)

0.793 (23/29)

Reference

0.494

0.815

0.471

0.501

DLRad2

0.718 (0.559–0.877)

0.658 (27/41)

0.916 (11/12)

0.551 (16/29)

0.494

Reference

0.649

0.897

0.980

2D DL

0.769 (0.622–0.915)

0.731 (30/41)

0.833 (10/12)

0.689 (20/29)

0.815

0.649

Reference

0.600

0.662

3D DL

0.701 (0.497–0.906)

0.781 (32/41)

0.666 (8/12)

0.827 (24/29)

0.471

0.897

0.600

Reference

0.879

Rad_H1

0.721 (0.568–0.874)

0.756 (31/41)

0.833 (10/12)

0.724 (21/29)

0.501

0.980

0.662

0.879

Reference

  1. AUC area under the receiver operating characteristic curve, CI confidence interval, DL deep learning, 3D three-dimensional, 2D two-dimensional, H1 habitat, Rad radiomic.
Back to article page