Robust Deep Learning for Medical Image Reconstruction

About the Project

The Department of Biomedical Engineering at the University of Dundee is inviting applications for a self-funded PhD research project under the supervision of Dr Alessandro Perelli. This position is an opportunity to conduct cutting-edge research within computational medical imaging, machine learning and medicine, with the possibility of collaboration with clinicians at the Ninewells Hospital in Dundee.

This research project focuses on the development of robust deep learning algorithms for image reconstruction in medical imaging with applications to X-ray Computed Tomography (CT) and Magnetic Resonance Imaging (MRI).

Image reconstruction is an inverse problem that deals with the estimation of an image from measurement data obtained from physics-based acquisition processes. As instance, modern techniques used in oncology such as CT and MRI require fast and accurate image reconstruction algorithms to deal with incomplete data to speed-up the acquisition, noise and artefacts [1-3].

Deep learning methods, that use for training past datasets of successfully reconstructed images together with the measurement that produced them, have been shown to produce some impressive results in image reconstruction. However, the problem with such data-driven imaging is currently the reliability of the results and robustness to model acquisition mismatch, i.e. different scanners used in training and testing which hinders the use for clinical assessment. Indeed, even small deviations in the data can result in large differences in the outcome which has devastating implications for many applications [4].

Furthermore, the collection of training ground truth clinical data is sometimes impossible to obtain under indirect acquisition of the specimen through medical scanners.

Therefore, an important question is to understand whether it is possible to deploy deep learning algorithms which do not require accurate ground truth data rather noisy ones or can learn from the testing image or data itself without training dataset.

In this project you will answer these questions by developing new algorithms, based on deep learning and optimization, and analysis techniques towards robust computational image reconstruction for CT and MRI. You are expected to further advance current developments of supervised (or self-supervised) model-based deep learning with convolutional network and deep generative prior [5-7]. The expected project’s outcomes will involve writing research papers with the developed methods together with the creation of the software to reproduce the results and clinical validation.

There will be opportunities for the student to collaborate with clinicians at NHS Tayside and to exploit the facilities in the Division for Imaging and Technology, University of Dundee.

Diversity statement

Our research community thrives on the diversity of students and staff which helps to make the University of Dundee a UK university of choice for postgraduate research. We welcome applications from all talented individuals and are committed to widening access to those who have the ability and potential to benefit from higher education.