Table 1 Detection accuracy and segmentation performance.

From: Handling missing MRI sequences in deep learning segmentation of brain metastases: a multicenter study

Voxel-based statistics Lesion-based statistics
Network AUC ROC DICE IoU Recall Precision FPR FP (no size limit) FP (10 mm3 size limit)
DeepLab V3 0.989 ± 0.023 0.774 ± 0.104 0.492 ± 0.186 0.631 ± 0.208 0.722 ± 0.206 0.001 ± 0.001 26.3 ± 17.2 7.0 ± 5.3
ILD-model 0.989 ± 0.029 0.795 ± 0.105 0.561 ± 0.225 0.671 ± 0.262 0.790 ± 0.158 0.001 ± 0.001 12.3 ± 10.2 3.6 ± 4.1
p-value 0.620 0.017 <0.001 0.167 0.095 0.065 <0.001 <0.001
  1. All metrics except AUC ROC were estimated using a probability threshold of 0.87 for the DeepLab V3 model, and 0.76 for the DropOut model.