Overcoming scarcity MRI data from the brain
Aymen Ayaz defended her thesis at the Department of Biomedical Engineering on April 18.
![](https://assets.w3.tue.nl/w/fileadmin/_processed_/5/b/csm_Ayaz_Aymen_BMT_AS_0260_PhD_ca4283f323.jpg)
The brain is one of the most complex organs in the human body with numerous intricate structures, each with its unique function and interconnected networks. Magnetic resonance imaging (MRI) of the brain has become an indispensable tool in both clinical practice and scientific research. Unfortunately, MRI data are scarce due to a lack of large, accurately annotated datasets, which complicates the development of effective deep learning algorithms for MRI analysis. With her PhD research, Aymen Ayaz focused on addressing the shortage of brain MRI data by exploring innovative methods to generate large, annotated datasets.
MRIs play a crucial role in diagnosing various neurological disorders, planning treatments and monitoring disease progression.
Analyzing brain MRIs is a challenging task and a time-consuming process for radiologists and neurologists. The emergence of deep learning (DL) technologies has introduced promising avenues for automating and streamlining MRI analyses. Yet the effectiveness of DL algorithms depends on access to large, precisely annotated MRI datasets. Such data are scarce due to a lack of good ground Truth (GT) annotations, restrictive data sharing policies and privacy concerns.
Scarce MRI data
Ayaz tackled the challenge of scarce MRI data by exploring innovative ways to artificially generate large, annotated datasets.
She did this by delving into two primary methods: physics-based simulations and DL-driven image synthesis.
With these, she developed a comprehensive MRI simulation framework that simulates realistic brain images by mimicking patient-specific anatomies. In addition, she investigated generative models for creating lifelike 3D MRI images of the brain, including images with specific pathologies such as tumors.
Moreover, the utility of synthetic MRI data in various DL applications was evaluated, from segmenting brain structures to improving image resolution. This will certainly highlight the effectiveness of both physics-based simulations and data-driven synthesis techniques in generating brain MRI data for medical image analysis tasks.
Title of PhD thesis: “Simulation and Synthesis for Brain Magnetic Resonance Image Analysis”
(Open access 18-10-2024)
Supervisors: Marcel Breeuwer and Josien Pluim
Latest news
![[Translate to English:] [Translate to English:]](https://assets.w3.tue.nl/w/fileadmin/_processed_/6/2/csm_universiteit%20jos%20keurentjes%20energie%202_d469fe3b2f.jpg)
![[Translate to English:] [Translate to English:]](https://assets.w3.tue.nl/w/fileadmin/_processed_/2/2/csm_Pre-EditLeviBaruch-0003_558369f155.jpg)
![[Translate to English:] [Translate to English:]](https://assets.w3.tue.nl/w/fileadmin/_processed_/7/d/csm_Boutin%20Banner%20image_165e239d2b.jpg)