U-NET

• Used Pydicom, OpenCV for DICOM file conversion processing; Utilized MATLAB to preprocess the dataset, applying image enhancement techniques and random rotation to improve model generalization
• Constructed a U-NET neural network using PyTorch, fine-tuning hyperparameters, precisely the number of U-NET layers and channels for the task of lung CT images
• Improved the accuracy of conventional image segmentation to 91% segmentation accuracy, contributed to the foundation of subsequent lung image classification tasks through precise lung image segmentation

ENVIRONMENT

pycharm+python3.6+pytorch1.3.1

HOW TO RUN:

Enter the dataset.py and correct the path of the datasets example:

python main.py --action train&test --arch UNet --epoch 21 --batch_size 21 

U-NET Model

RESULTS

after train and test,3 folders will be created,they are "result","saved_model","saved_predict".

saved_model folder:

After training,the saved model is in this folder.

result folder: