Neural Network Architecture Design for Deformable Image Registration in Female Pelvic Anatomy

The Radiation Medicine Program at the Princess Margaret Cancer Centre treats cancer patients with Radiation Therapy (RT). RT relies on the principle that the cancer can be killed by delivering a high radiation dose to the tumor while sparing healthy tissues. Deformable image registration (DIR) is a cornerstone of modern RT, enabling the alignment of images across multiple treatment courses and imaging modalities. It plays an important role in clinical workflows such as image fusion, tumor tracking, and dose accumulation. However, in female pelvic radiotherapy, DIR methods face substantial challenges in managing complex anatomical changes. These include large deformations caused by different organ fillings such in the bladder, rectum or bowel, organ sliding, and tumor shrinkage or swelling, which can compromise the accuracy of registration. To address these challenges, there is a critical need for an automated DIR pipeline capable of handling such complexities in real time, facilitating more precise adaptive radiotherapy and improving clinical outcomes.

Faculty Supervisor:

Aviad Levis

Student:

Partner:

Princess Margaret Cancer Centre

Discipline:

Computer science

Sector:

Health and Related Sciences & Technology

University:

University of Toronto

Program: