Automatic segmentation of pelvic bones in Dixon MRI

Being able to analyze a 3D model of a patient’s pelvic bones allows doctors to plan surgeries and diagnosis diseases. With continued improvements in medical imaging and machine learning, it is now fast and accurate to produce these models. However, current techniques rely on CT scans that emit ionizing radiation. This type of radiation is usually safe but can not be recommended for all patients, for example, young adults with bony pelvic tumors. We are aiming to design a new machine learning solution for creating pelvic bone models by relying on MRI scans. These scans are traditionally used for viewing soft tissues like organs, but by using the Dixon MRI technique, we can observe bone structures. Besides providing accurate models for normally shaped pelvises, we hope to have better results than previous solutions when dealing with deformed pelvises because of disease or surgery.

Faculty Supervisor:

Thomas Fevens

Student:

Partner:

Ludwig-Maximilians-Universität München

Discipline:

Computer science

Sector:

Education

University:

Concordia University

Program: