Table 3 The feature number and performance of mixed radiomics features models and Radscore models base on consecutive vertebrae for predicting low BMD.

From: Radiomics models based on thoracic and upper lumbar spine in chest LDCT to predict low bone mineral density

Consecutive vertebrae

Feature number

Training set

Validation set

(Significant/Lasso)

AUC (95%CI)

Sensitivity (%)

Specificity (%)

AUC (95%CI)

Sensitivity (%)

Specificity (%)

Mixed radiomics features models

T1-T3

232/31

0.865

(0.837–0.894)

0.779

0.792

0.795

(0.743–0.846)

0.669

0.784

T4-T6

235/37

0.871

(0.843–0.898)

0.764

0.828

0.820

(0.771–0.869)

0.719

0.759

T7-T9

231/41

0.881

(0.854–0.907)

0.783

0.839

0.792

(0.740–0.844)

0.653

0.735

T10-T12

208/45

0.904

(0.880–0.927)

0.843

0.809

0.831

(0.784–0.878)

0.793

0.728

L1-L2

159/36

0.913

(0.890–0.935)

0.775

0.893

0.883

(0.843–0.922)

0.736

0.864

Radscore models

T1-T3

232/31

0.876

(0.849–0.903)

0.730

0.893

0.809

(0.759–0.859)

0.620

0.833

T4-T6

235/37

0.870

(0.842–0.898)

0.719

0.879

0.855

(0.812–0.899)

0.645

0.870

T7-T9

231/41

0.930

(0.910–0.950)

0.843

0.893

0.845

(0.802–0.890)

0.711

0.765

T10-T12

208/45

0.915

(0.892–0.937)

0.802

0.887

0.871

(0.831–0.912)

0.760

0.815

L1-L2

159/36

0.917

(0.893–0.938)

0.813

0.851

0.906

(0.871–0.940)

0.818

0.840

  1. AUC, area under the curve of the receiver operating characteristic (ROC) curve. BMD, bone mineral density. CI: confidence interval. LASSO, least absolute shrinkage and selection operator.