Summer Scholarships

Image X prides itself on offering a summer research program filled with opportunities. Students can work onsite amongst our multidisciplinary team, learning about the different projects and careers within our group. We host social activities, tours of our research facilities, as well as an introductory seminar series about radiation therapy and cancer imaging, presented by our star researchers. Many summer students continue working with Image X after their projects, from taking on part time research positions through to commencing PhD projects.

Successful applicants will be invited to choose from the projects below (please note, more projects are being added every week). Scholarships will be awarded based on the student’s academic transcript, relevant academic background or any research experience.

The scholarships are offered on behalf of the Cancer Imaging and Targeted Radiotherapy Translational Program, which is a partnership between universities, local health districts and industry. The program aims to translate imaging and targeted radiotherapy medical devices from the discovery phase to broad clinical practice.

Deep-learning based tumour motion monitoring for liver SABR

Supervisors:
Dr Chandrima Sengupta, Prof Paul Keall, Adam Mylonas.

Project description:

During radiotherapy treatments, liver tumours move significantly from the planned treatment position. As a result, the radiation beam may miss the tumour and hit nearby healthy tissues, resulting in poor treatment outcomes. 

This project will develop and retrospectively evaluate a deep-learning based liver tumour tracking algorithm to detect gold markers in x-ray images acquired in the LARK clinical trial across multiple centres in Australia. The results will be compared against the triangulation method. The results will be compared among different breathing techniques, linear accelerators and different imaging parameters. 

Optimisation of deep learning networks for markerless tumour tracking during radiation therapy.

Supervisors:
Adam Mylonas, Mark Gardner, Emily Hewson

Project description:

Effective radiation therapy requires knowing the precise location of the tumour to maximise the radiation dose delivered to the tumour while minimising the radiation to the surrounding healthy tissue. At the Image X Institute, we are pioneering the use of deep learning to detect and track tumours in x-ray images acquired during radiation therapy to ensure we maximise the radiation dose to the tumour. The deep learning networks used are conditional Generative Adversarial Networks (cGANs) which use a unique training strategy for maximum effectiveness. Due to the unique cGAN training approach, there is limited understanding of the ideal training time, particularly when it comes to determining the optimal training stopping point. In this project, you will investigate and optimise training strategies for cGANs for the purpose of detecting tumours in x-ray images. This project does not require significant programming experience.

Optical Calibration for novel imaging trajectory development on C-arm CBCT systems

Supervisors:
Dr Tess Reynolds, Dr Youssef Ben Bouchta

Project description:

Cone Beam Computed Tomography (CBCT) imaging is increasingly being used during image-guided surgical interventions in disciplines ranging from cardiology, neurology, and radiology through to orthopaedics. The increased demand for intraoperative imaging drives innovations in the acquisition to reduce imaging dose, increasing image quality and expand the field of view, for example. One method to achieve this is to take advantage of the flexibility of modern robotic CBCT imaging systems and implement non-circular imaging trajectories. However, in order to produce a high-quality 3D image, precise geometry of the system during acquisition is required and is not natively available.

This project looks to evaluate the possibility of using commercially available optical tracking systems to provide geometric calibration for a state-of-the-art robotic CBCT imaging systems, facilitating the future development of novel imaging trajectories. This project will include experimental work with the Siemens ARTIS pheno robotic CBCT imaging system located at the University of Sydney’s Hybrid Theatre in the Charles Perkins Centre.

This project can be tailored to a range of students with broad backgrounds. Tasks might include software development, experimental validation, supervised use of imaging equipment, and data analysis.

The application of SAM on medical image segmentation

Supervisors:

Dr Yu Sun, Dr Erin Wang, Professor Annette Haworth

Project description:

SAM (Segment Anything Model) is an open source model published by Meta. It’s trained on Meta’s large natural image dataset (SA-1B, with 1.1 billion masks) and has a strong ability to comprehend edges and objects. It can be used as a zero/few-shot learning model, i.e. requiring very little fine tuning to adapt to a new environment. We can see SAM as a model to cut out objects based on boundaries, like humans do. Fig. 1 shows an example of SAM.

While SAM is not directly related to medical data, its output provides an important stepping stone to achieve segmentation efficiently (Fig 2).

This is because most segmentations in medical imaging data are based on boundaries such as the interfaces between normal and malignant tissues. Hence one can pick the segments from SAM’s output to match with the ground truth. The aim of this study is to investigate the additional value of SAM in segmenting subvolumes in medical images.

Measurement of interobserver variability in tumour delineation on longitudinal prostate MRI.

Supervisors:

Supervisors: Dr Yu Sun, Dr Erin Wang, Professor Annette Haworth, Dr Jonathan Sykes, Dr Joel Poder

Project description:

Historically, the entire prostate, rather than the tumour within the prostate, has been defined as the target volume for radiation therapy. Recent clinical trials have demonstrated a benefit from delivering higher doses to the tumour within the prostate as a method to improve tumour control without increasing toxicity. Interobserver variation in delineating the tumour on MRI has been previously documented. Uncertainty in delineating the tumour on MRI during and after treatment is less well understood. Tumour segmentation on longitudinal MRI is necessary to facilitate adaptive radiotherapy and post-RT response assessment.

This project will use data from 10 patients recruited to the SI-BiRT(2) study (ANZCTR UTN U1111-1221-9589). Variability in clinician defined tumour volumes will be assessed. This study will require co-registration of multiple data sets and assessment of variability in contouring using standard tools. Stretch goals include:

Validation or development of a DL tool to facilitate automated tumour delineation. The validated tool will be used to define tumour volumes on the 23 patient BiRT(1) cohort.

Impact of variability in tumour segmentation will be assessed by considering the impact on response assessment using a radiomics-based approach.

Predicting radiotherapy outcomes using machine learning, segmentation tools and radiobiology parameters

Supervisors:

Prof Lois Holloway and others depending on the focus of the project

Project description:

There is a growing amount of radiotherapy data available including clinical parameters such as blood tests and pathology, patient demographics and 3D imaging and radiation dose information and outcome data including cancer recurrence and treatment toxicity.

We are developing models to use the available data to predict outcomes and support treatment decisions by patients and clinicians. Development and validation of these models can be undertaken by machine learning. This also requires an understanding of the uncertainty of the factors/features that are included in the models. We are assessing the use of autosegmentation tools across different imaging and datasets for use in these models. We are also considering the impact of changes in fractionation and radiobiology models such as the linear quadratic model to account for these dose changes as techniques have changed over time.

This project can be tailored to a range of students with broad backgrounds. Tasks might include machine learning to validate or develop outcome models, data analysis, model validation and assessment of variation with changes in radiobiology parameters, validation and development of image segmentation tools.