Skip to main content
Fig. 1 | Insights into Imaging

Fig. 1

From: Deep learning enables automated MRI-based estimation of uterine volume also in patients with uterine fibroids undergoing high-intensity focused ultrasound therapy

Fig. 1

Schematic illustration of the used architectures: One layer in the default nnU-Net configuration consists of a convolution followed by an instance normalization and a leaky ReLU activation (blue). The modified architecture investigated in this work contains layers with additional CBAMs in the encoder (yellow), consisting of a convolution and an instance normalization followed by a channel and a spatial attention module, each connected by element-wise multiplication. The output of the first convolutional block and the output of the CBAM are merged by element-wise summation and again activated by leaky ReLU activation function

Back to article page