Skip to main content
Fig. 1 | European Radiology Experimental

Fig. 1

From: Evaluation of a CTA-based convolutional neural network for infarct volume prediction in anterior cerebral circulation ischaemic stroke

Fig. 1

The neural network architecture consisted of 40 three-dimensional convolutional layers (conv) with valid padding and exponential linear unit activation. Skip-connections (curved arrows) with concatenations (+) were used with appropriate cropping. These were followed by two fully connected layers (1 × 1 × 1 convolutions, FC) with 50 and 2 output neurons, and rectified linear unit and softmax activation, respectively. In the end, the predicted patches were stitched together to produce a lesion segmentation of the same size as the input

Back to article page