Title:
Noise-aware neural networks for quantitative MRI
Project Leader:
Christopher Parker, christopher.parker@ucl.ac.uk
Project Description:
Neural networks are becoming an increasingly popular way of estimating parameters of biophysical models from MRI data. Yet, current approaches do not integrate knowledge of spatially-varying noise characteristics, which could influence the accuracy of parameter estimates. This project will explore different ways of estimating spatially-varying noise characteristics and of utilising them to build and test neural networks for biophysical modelling of MRI data.
Ideal Participant Characteristics:
Python programming experience (including PyTorch, Tensorflow or Keras) would be useful but is not essential
I will provide “realistic” simulated MRI datasets and other helpful code
Laptop required!
Tasks:
Some possible example tasks:
Task 1
Goal: Characterise spatially-varying noise
Steps: Read in MRI data and use statistical techniques to build a map of the noise hyper-parameters
Task 2
Goal: Evaluate different techniques
Steps: Compare the accuracy of different noise characterisation techniques and how they perform on different MRI acquisitions
Task 3
Goal: Build noise-aware neural networks
Steps: Integrate knowledge of the noise characteristics into neural network training and inference
Title: Noise-aware neural networks for quantitative MRI
Project Leader: Christopher Parker, christopher.parker@ucl.ac.uk
Project Description: Neural networks are becoming an increasingly popular way of estimating parameters of biophysical models from MRI data. Yet, current approaches do not integrate knowledge of spatially-varying noise characteristics, which could influence the accuracy of parameter estimates. This project will explore different ways of estimating spatially-varying noise characteristics and of utilising them to build and test neural networks for biophysical modelling of MRI data.
Ideal Participant Characteristics:
Resources:
Tasks: Some possible example tasks: Task 1 Goal: Characterise spatially-varying noise Steps: Read in MRI data and use statistical techniques to build a map of the noise hyper-parameters Task 2 Goal: Evaluate different techniques Steps: Compare the accuracy of different noise characterisation techniques and how they perform on different MRI acquisitions Task 3 Goal: Build noise-aware neural networks Steps: Integrate knowledge of the noise characteristics into neural network training and inference