Fully Convolutional DenseNet with Multiscale Context for Automated Breast Tumor Segmentation

Table 2

The quantitative comparisons of ASPP-FC-DenseNet model with different loss functions.

Methods	DI	IOU	PA
No weighted loss	0.7151	0.5974	0.8015
Dice loss	0.7108	0.5920	0.7988
Ours	0.7697	0.6041	0.7983