Table 2.
Comparison between the parameters of conventional PET and MRI parameters and convolutional neural network methods for predicting pathological response to neoadjuvant chemotherapy.
| Sensitivity (%) | Specificity (%) | Accuracy (%) | AUC, median | |
|---|---|---|---|---|
| SUV0a | 50 | 88 | 84 | 0.652 |
| PET0b | 79 | 94 | 97 | 0.886 |
| SUV1 | 67 | 70 | 70 | 0.687 |
| PET1 | 72 | 96 | 95 | 0.980 |
| ADC0c | 100 | 56 | 61 | 0.703 |
| MRI0d | 18 | 90 | 85 | 0.602 |
| ADC1 | 100 | 38 | 45 | 0.537 |
| MRI1 | 14 | 90 | 88 | 0.701 |
AUC area under the curve.
aSUV0 maximum standardized uptake value at baseline, SUV1 maximum standardized uptake value on interim images.
bPET0 baseline PET image data for deep learning, PET1 interim PET image data for deep learning.
cADC0 apparent diffusion coefficient at baseline, ADC1 apparent diffusion coefficient on interim images.
dMRI0 baseline MR image data for deep learning, MRI1 interim MR image data for deep learning, PET positron emission tomography, MRI magnetic resonance imaging.