Week #24 and #25

08 Apr 2019

Definition of the goals

We had a meeting on 11 April 2019 to define our final goal and each step goals. The main idea is to follow the steps of FreeSurfer and turns every step or every 2 steps to one neural network. We also discussed other possibilities such as using meshes rather than volumes. So the first step is to get a first result of Skull-strip.

Development of the Dataset and the training Experiments

Preprocess of Dataset

Using the same strategy as I used before, I slice the volumes into 3 planes. The brainmask size: torch.Size([182, 217, 182]). The slice numbers in each folders is as follows:

Axial	Coronal	Sagittal
182	182	217

Current process

All the T1 weighted images and the brain mask is resized to: 128x128. The original T1 weighted images value range is [0,172], so in order to normalize the data, if the maximum value is considered as 255, the data range after normalization is [-1,0.3412]. So in order to get data distributed in [-1,1], the original data could have a transform using data/172*255.