Thank you for the introduction, Amma.

Dear professors, dear colleagues, and dear friends, thank you for coming to my PhD defense.

First of all, let me introduce what image angle tomography is.

Combing CT is a widely used medical imaging technology.

It reconstructs a 3D volume, which provides cross-sectional information of an imaged

patient.

In clinical applications, typically an angiographic C-arm system is used to acquire 3D images

for planning, guiding, and also monitoring of interventional therapies.

In order to do 3D reconstruction, the source and the detector need to rotate at least 180

degrees plus a cone angle as a short scan, as displayed in this video.

Here the source and the detector rotate like a propeller.

So this rotation mode is called the propeller mode.

And these are the acquired projections at different angular positions.

Note that there is no limited angle problem here in this video.

However, in practice, the rotation might be restricted by some external obstacles, for

example, an extra surgical device near the patient bed.

Then the problem of limited angle tomography arises.

There are two rotation modes in C-arm devices.

The other one is called the orbital mode, where the source and the detector rotate along

the arc of the C-arm.

For some systems, the arc length is decided to be small so that there is more opening

space for surgical operations.

For example, this Siemens multipurpose system, it can rotate 150 degrees only.

If we want to use such a system for 3D image reconstruction, then the problem of limited

angle tomography also arises.

So image reconstruction, for data acquired in an insufficient angular range is called

the limited angle tomography.

In CT, filtered back projection is the standard image reconstruction technique.

However, in limited angle tomography, due to missing data, artifacts will occur.

For example, this is a custom reference image.

And this is 160 degree FBP reconstruction, with the trajectory from 10 degrees to 170

degrees.

So here we can see the boundary is distorted and many edges are blurred.

And also there are many horizontal large streaks.

The orientations of these streaks are highly dependent on the scan trajectory.

For example, if we switch to this trajectory from 100 degrees to 260 degrees, then the

streak orientations are almost vertical instead of horizontal.

So this is called the trajectory dependency property of limited angle tomography.

So here the task of my PhD project is to improve image quality for limited angle tomography

using various approaches.

The most intuitive one is to extrapolate missing data using, for example, some data consistency

conditions.

The next method is called a compressed sensing, which can achieve better image quality for

image reconstruction from insufficient data by using some sparsity regularization.

And nowadays, machine learning has achieved overwhelming success in various fields, including

CT reconstruction.

For limited angle tomography, there are so many state of the art methods out there.

For example, the U-net method, GaNS, and variational neural networks.

And with this trend, so all of my contributions to limited angle tomography also lie in these

aspects.