Learning Day 69: Image segmentation for biomedical applications — U-Net

U-Net

• It is a special type of FCN which contains only convolution layers
• It gradually reduces the volume of feature maps (reduce size, increase depth; overall volume is decreasing) to a small feature map, then gradually increase the volume of feature maps again
• At the volume increasing stage (2nd half of U-Net), It employs upsampling techniques such as deconvolution mentioned in Day 67.
• It also has shortcuts from the left to right feature maps without going through the entire U-Net
• In each convolution block, it employs the structure of ResNet block

An example of U-Net architecture (ref)

Loss function — Dice

• It is a measurement of overlap between two samples
• 0≤Dice≤1
• Developed for binary data

Dice coefficient (ref)

• It is very similar to IoU mentioned in Day 62. IoU is also called Jaccard index/similarity coefficient:

Jaccard index (ref)

• The difference between Dice and Jaccard coefficients was explained here

Data pre-processing in U-Net

• These techniques are suitable for biomedical image segmentation or other similar applications

Curvature driven flow for image denoising

• Biomedical images tend to contain noises
• Gaussian blur is not fantastic in this application as it usually blurs the boundary
• Curvature driven flow is able to preserve sharp boundaries with smoothing occurring only within a region (ref)
• It is based on a theory that the brightness of objects should be smooth and the iso-brightness contours should have small curvature. If noise exists, irregular iso-brightness contours emerge resulting in high curvature at those regions

Elastic deformation for data augmentation (Besides the conventional ones)

• It generates a form of distorted data as illustrated below

Elastic deformation example from original paper (ref)

Before (left) and after (right) elastic deformation (ref)

• This method was used to suit the context of biomedical image: since the imaging of internal organs may distort at different measurement postures

Reference

All presented contents without specific references are summarised from the course link1