Table 4 Comparison of models among team A, B,C in the BMMR2 challenge and ours
| Â | Prediction Model | Sequence Used | Timepoint Used | Clinical Feature | Was Model Tested Using Other Data? | Model Performance |
|---|---|---|---|---|---|---|
Team A | Logistic regression | SER map and kinetic maps from DCE image | Pre-NAT and early-NAT | HR and HER2 | No | ACRIN 6698 test: 0.840 (95%CI: 0.748–0.932) |
Team B | XGBoost, Random forest, Logistic regression | DCE image, SER map, PE map, DW image | Pre-NAT, early-NAT, and mid-NAT | Age, Race, Lesion Type, SBR Tumor Grade, MRI Longest Diameter, HR and HER2 | No | ACRIN 6698 test: 0.838 (95%CI: 0.748–0,928) |
Team C | Logistic regression | DW image | early-NAT, | MRI Longest Diameter, HR and HER2 | No | ACRIN 6698 test: 0.803 (95%CI: 0.702–0.904) |
MESN | Neural networks | DCE image, DW image, and ADC image | Pre-NAT and early-NAT | None | Yes | ACRIN 6698 test: 0.860 (95%CI: 0.757–0.939) External test: 0.804 (95%CI: 0.752–0.864) |
MESN-C | Neural networks, Logistic regression | DCE image, DW image, and ADC image | Pre-NAT and early-NAT | HR and HER2 | Yes | ACRIN 6698 test: 0.903 (95%CI: 0.815– 0.965) External Test: 0.861 (95%CI: 0.811–0.906) |