Skip to main content
. 2019 Dec 30;9:20311. doi: 10.1038/s41598-019-56767-3

Figure 1.

Figure 1

Feature extraction by AlexNet. The lesion image was cut out using a VOI and resized to 243 × 243 pixels. The image was cropped to 227 × 227 pixels with a shift of ±8 pixels and rotated/flipped for data augmentation. The augmented image was input to the pretrained AlexNet to acquire the texture features for classification.