FDKM

This is the code for "Pseudo Kinetics-driven Federated Diffusion Hemodynamic Framework for Breast Tumor Segmentation in Pre-contrast MRI"

Installation and Setup

1. Hardware requirements: This project requires only a standard computer with enough RAM and a NVIDIA GPU to support operations. We ran the demo using the following specs:

   • CPU: 10 cores, 2.5 GHz/core
   • RAM: 40GB
   • GPU: NVIDIA Quadro RTX 8000 (48GB memory)
   • CUDA: 11.0

2. System requirements: This tool is supported for Linux. The tool has been tested on the following system:

   • Ubtuntu Linux release 18.04.6

3. Clone the Repository:

   git clone https://github.com/ttt553/FDKM
   cd FDKM

4. Install Required Packages: The basic environment requirements are:

   • Python: 3.10
   • CUDA: 11.0
   • Pytorch

5. Model pretraining:

    python pretrain.py

Please replace the data path using your specific path in the pretrain.py.

You can customize the training process using the following arguments: 

Argument	Description	Default
--ch_mult,	channel multiplier	[1,2,2,2]
--ch,	base channel of UNet	128
--beta_1,	start beta value	1e-4
--beta_T,	end beta value	0.02
--T,	total diffusion steps	1000
--lr,	target learning rate	2e-4
--img_size	image size	128
--save_step	frequency of saving checkpoints	5000
--ema_decay, -c	ema decay rate	0.9999

6. Model training and evaluaiton:

    python train.py

The training procedure can be customized via the following command-line arguments:

Argument	Description	Default
--epochs, -e	Total number of training epochs	500
--batch-size, -b	Number of samples per training batch	1
--learning-rate, -l	Learning rate for the optimizer	1e-5
--load, -f	Path to a checkpoint .pth file to resume training from	None
--scale, -s	Factor by which to downscale input images	1
--validation, -v	Percentage of the dataset reserved for validation (range: 0-100)	20.0
--amp	Enable Automatic Mixed Precision (AMP) during training	Disabled
--bilinear	Utilize bilinear upsampling in the network architecture	Disabled
--classes, -c	Number of output classes for segmentation	1


7. Model testing:

   python test.py