Optimization of MRI Pulse Sequence by Reinforcement Learning.

This repository contains the code and report for my master thesis Optimization of MRI Pulse Sequence by Reinforcement Learning. This project focus on the optimization of gradient-echo sequences for 1-D objects using the Deep Deterministic Policy Gradient (DDPG) algorithm under constraints on gradient slew rate.

A 1D imaging protocol for a gradient echo sequence structure.

Schematic of the DDPG framework.

Actions and corresponding error in testing episodes for 1-D object.

Dependencies

pip install -r requirements.txt

DDPG model

1. Modify the target object in the env.py file.

self.density = np.zeros(len(self.x_axis)) # target object

2. Modify the arguments in the main.py file.
3. Run the main.py file.

python main.py

4. output of model and best action for each testing episode will be saved in the ./src/Training/{datetime} folder by default.

MR signal simulator

• run the `./src/simulator.ipynb' file to simulate the MR signal of the target object.