Table 1.
Comparison of Characteristics Between the Training and Test Sets.
Table 2.
Prediction Performance of Various Radiomic Signatures for LVI of Invasive Breast Cancer in Test set.
Fig 1.
Prediction performance of RSpre, RSpost, RST2, RSDelta in LVI.
ROC of Test set (A); Delong test of Test set (B). These showed that RSDelta had the best performance for detecting LVI with AUC of 0.764, which had statistical differences with RSpre(p = 0.025) and RSpost (p = 0.036). But there was no statistical difference with RST2 (p = 0.239).
Fig 2.
The confusion matrices of the four radiomics signatures in the test set.
Fig 3.
Feature importance of the RSDelta model.
Fig 4.
Decision curve analysis(DCA) of four Radiomic Signatures (RSpre, RSpost, RSDelta, RST2) indicating that the RSDelta achieve higher net benefit across most thresholds.
(A) DCA of Training set, (B) DCA of Test set, (C) Calibration curves of train set, (D) Calibration curves of test set.