Deep Learning MRI image analysis
Core codes of tensorflow implementation for end to end survival prediction in rectal cancer MRI
A csv file in the following formate, see also clini.csv:
| id | ori_path | mask_path | os | os_e | dfs | dfs_e |
|---|---|---|---|---|---|---|
| 10000 | /input/ori/A1 | ../input/mask/pre-A1.nii | 100 | 1 | 100 | 1 |
Training script is in train.py file, network structure based on ViTB/16, from ViT-kera repository, a Keras implementation of ViT.
Cox partial likelihood as the loss function for the network;
Cindex is the evaluation metric of model.
$$L(\theta)= -\sum_{i: E_{i}=1}\left(\hat{h}{\theta}\left(x{i}\right)-\log \sum_{j \in \Re\left(T_{i}\right)} e^{\hat{h}{\theta}\left(x{j}\right)}\right)$$