Table 3.
Sensitivity, specificity, accuracy, and receiver operating characteristics (ROC) area under the curve (AUC) for the performance of ΔDCE, ΔRSI3C, and ΔADC for prediction of non-pCR at each time point.
| ΔDCE | ΔRSI3C | ΔADC (↑ = response) | ||
|---|---|---|---|---|
| Early-Tx (3 weeks) |
AUC (95%CI) Sens. Spec. Accu. Thresh. |
0.64 (0.36-0.91) 0.55 1.00 0.71 -0.07 |
0.65 (0.38-0.92) 0.91 0.50 0.76 -0.58 |
0.45 (0.16-0.75) 0.82 0.33 0.65 0.47 |
| Mid-Tx (12 weeks) |
AUC (95%CI) Sens. Spec. Accu. Thresh. |
0.71 (0.45-0.96) 0.91 0.33 0.71 -0.73 |
0.60 (0.32-0.88) 1.00* 0.00* 0.65* <-1.00* |
0.35 (0.06-0.64) 0.91 0.17 0.65 1.06 |
| Post-Tx | AUC (95%CI) Sens. Spec. Accu. Thresh. |
0.80 (0.59-1.00) 0.73 0.83 0.76 -0.72 |
0.76 (0.52-0.99) 0.64 1.00 0.76 -0.81 |
0.36 (0.07-0.65) 0.91 0.17 0.65 0.99 |
Note that the post-Tx time point is after all neoadjuvant therapy prior to surgical intervention. As the data is normalized to pre-Tx time point, threshold values are unitless multiplication factors.
* The optimal accuracy is achieved by a threshold where all cases are classified as non-pCR (sensitivity = 100%).
pCR, pathological complete response; Tx, treatment; ΔDCE, change in size from pre-treatment time point for manual dynamic contrast-enhanced MRI; ΔRSI3C, change in size from pre-treatment time point for the three-component Restriction Spectrum Imaging model classifier; ΔADC, change in mean value from pre-treatment time point for apparent diffusion coefficient.