Table 6 Results of cross-dataset evaluation compared to other approaches.

From: Investigation of an efficient multi-modal convolutional neural network for multiple sclerosis lesion detection

Approach

Consensus mask

DSC

PPV

TPR

LFPR

LTPR

Multi-branch U-Net (proposed)

0.68 [0.65, 0.71]

0.77 [0.73, 0.81]

0.63 [0.59, 0.67]

0.63 [0.58, 0.68]

0.64 [0.60, 0.68]

DeepLesionBrain21

0.639

0.768

0.608

0.319

0.700

2.5D Tiramisu17

0.664

0.741

0.658

0.284

0.695

  1. This table compares the results for the cross-dataset performance from our approach to other state-of-the-art approaches. All networks were initially trained on the ISBI set16 and tested on the MSSEG dataset28. The numbers are averaged over all subjects. Our approach is using three modalities (FLAIR, T1w, T2w), whereas the other two were trained with two modalities (FLAIR, T1w). The numbers in square brackets indicate the 95%-confidence intervals. The performance metrics for the other approaches were taken from R. A. Kamraoui et. al.21. The best and second-best results are written in bold and italic, respectively.
Back to article page