Spring into STEM: AI to Improve Proton Beam Therapy Delivery in Children with Cancer

Proton Beam Therapy (PBT) is a form of cancer treatment that recently has become available in the UK. Compared to the conventional radiotherapy, PBT is highly targeted, so reduces the risk of radiation damage. That makes it particularly favourable to children, who are very radiosensitive and commonly develop radiation-induced side effects of treatment later in their life. Abdominal structures in children are affected by significant gas filling fluctuation. That might result in treatment plan deviations. The treatment plan accuracy is important and requires monitoring to ensure safe and successful treatment. That can be achieved by improving quality of Cone Beam Computed Tomography (CBCT) images acquired on-the-day of treatment, generating synthetic Computed Tomography (synCT) images. These synCT images can be used for the treatment plan evaluation. In our methodology we adopted an AI cycle consistency Generative Adversarial Network. We tailored the original framework to address challenges in paediatric application. The proposed method resulted in improved quality on consistency of the synCT. Visual improvements were confirmed by numerical evaluation conducted in a 5-fold cross validation manner measuring differences in Hounsfield Units. The proposed method also achieved better results than its counterparts on an unseen testing dataset. Our AI-based method improves the quality of CBCTs, while preserving their anatomical structures. That potentially allows for Proton Beam Therapy plan verification and its immediate adaptation, to ensure successful treatment.

About the speaker

Adam is a postdoctoral researcher at CMIC, developing machine learning methods for image analysis in radiotherapy. He completed his DPhil at the  Institute  of  Biomedical  Engineering  at  the University of Oxford,  where  he worked on novel methods for multimodal  lung  image  registration for  CT,  MRI  and HPXe-MRI, and CT-based ventilation estimation methods. He worked to develop an automatic framework for identifying and segmenting different classes of parenchymal lung damage. Recently, he worked on AI-methods for paediatric patients, including segmenting organs at risk and synthesizing CT images from lower quality Cone Beam CTs.

Our taster lectures will give you a chance to find out about the subjects we teach and research, you'll be able to meet our lecturers and researchers and you can put your questions to the team. We look forward to welcoming you.

• See all the lectures in the Spring into STEM series at UCL

UCL uses a third party (Zoom) to administer our webinar/virtual open days and manage your personal information on our behalf. If you are happy for us to process your data solely for this purpose, please continue by entering your details in the registration form.
Our Prospective Student Privacy notice is available here