This project contains the work done for my master's thesis "Deep Learning in Medical Image Synthesis", at Utrecht University & Technical University of Eindhoven. This project mainly includes using different generation networks for the X-ray images.
For different classification models:
CheXNet: Use the train.py file in the Classification_Model/ChexPert/bin for training, main hyper-parameters are shown in the file.
DenseNet: Use the densenet.py file in the Classification_Model/DenseNet for training. Remember to set the path for every input and output file.
Inception-ResNet: Read the readme.md in the Classification_Model/Inception-ResNet-v2 and follow the instruction to train.
For different label imputation methods:
Change .fillna(0) and .replace(-1,1) in data preprocessing to .fillna(x) and .replace(-1,x).
For different image masking methods:
Use the mask.py file in the folder to perform masking operations on the dataset.
mask_1 function is the joint mask method.
mask_2 function is the inverted joint mask method.
mask_3 function is the square mask method.
mask_4 function is the separate mask method.
For different augmentation methods:
Traditional augmentation: According to the flag parameter in the create function, use the augmentation.py file in the Gereration_Model to generate image data sets of different traditional augmentation methods. If you want to generate a combined method, you can directly use multiple methods at the same time
DDPM model：use the training_DDPM.py file in the Generation_Model.
DDIM model：use the main.py file in the Generation_Model/ddim-main.
Stable Diffusion: use the huggingface.py file in the Generation_Model, the prompt can be changed in the file. Or can use the img2img_old.py file in the stable-diffusion-main/scripts to generate.
Stable Diffusion Finetune: I use the project from https://github.com/AUTOMATIC1111/stable-diffusion-webui
For different augmentation strategies:
For the unbalanced augmentation strategy, use the create function in the augmentation.py file in xxxx folder.
For the balanced augmentation strategy, use the create_2 function in the augmentation.py file in xxxx folder.
For other augmentation strategy, use the PCA.py file to visualise. The augmentation file is not uploaded yet.
For evaluation metrics:
All classification metrics such as accuracy, precision are in the eval_classification.py file.
All generation metrics such as SSIM, PSNR are in the eval_generation.py file.
Code for the mcnemar's test is in the mcnemar.py file.
Code for the heatmap is in the heatmap.py file.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent