Available

Project reference

2024_P24_Sparks_Vercauteren

Start date

June or October 2025
4

Back to projects

Developing automated contrast agnostic brain parcellation for neurosurgical planning and guidance

T1-weighted MRI with contrast agent is the most commonly acquired scan for patients undergoing neurosurgical intervention. Due to time and cost constraints, this is often the only scan acquired, e.g. patient undergoing emergency brain tumour biopsy. Currently there are a lack of computational tools capable of performing accurate parcellation on these images. In this project the student will focus on developing computational methods to perform brain parcellation in the presence of contrast agent for a range of images from radiologically normal scans to those with anatomy-distorting pathology. The student will first adapt a state-of-the-art convolutional neural network for this task. They will then develop semi-supervised learning techniques to help train a model that is robust and accurate. Finally, the student may consider quantifying segmentation quality to train the model to predict its own accuracy. Upon successful completion the final algorithm will be incorporated into an existing neurosurgical planning software.