SPS22-112UL

Skull Stripping MRI images with 3D CNN

By: Sebin Yoon and Anneke Moeller         

Department: Computer Science

Faculty Advisors: Dr. Ilmi Yoon, Dr. Julio Ramirez, and Julia Cluceru (Genentech)

Data preprocessing is an important step to prepare data for imaging analysis and machine learning. In medical imaging analysis data may often be inconsistent due to the inconsistencies in the way that imaging is acquired and as well as patient movement which affects image orientation and resolution. One aspect of data preprocessing for medical imaging analysis is a process called skull stripping. Currently skull stripping is done with a mix of computers and humans. An algorithm can do a part of the work, but the cleanup has to be done manually. This can be very time consuming. A machine learning algorithm that can perform both the manual tasks and the algorithmic tasks would reduce much of the time that goes into data preprocessing. To create this algorithm a dataset with 125 images was used to train a machine learning model to create a brain mask for a MRI scan of a patient.