3D-Unet Segmentation of Lung lobes in CT volumes
This project aims at predicting segmented images based on CT scans of Lungs (3d volumes).
I uploaded the .py files and a Google Colab notebook where I trained the 3d-Unet model on the free GPU.
Files:
-
Data3D.py: Data Preprocessing
-
3D UNet.py: model design, training, testing and evaluation.
Dataset:
format = (slices x rows x columns)
scans = 51 .nrrd files of size 256x256x256.
masks = 51 .nrrd files of size 256x256x256.
Original size:
Patches of size 128x128x128:
Code:
- Importing files and data preprocessing
- Patch Extraction
- Model
- Training
- Cross validation
- Testing
- Loss and Loss Validation Plots
- Comparing predicted segmantation with scans and masks
- Future work for improving results
Keras/TensorFlow API