ETRO VUB
About ETRO  |  News  |  Events  |  Vacancies  |  Contact  
Home Research Education Industry Publications About ETRO

Master theses

Current and past ideas and concepts for Master Theses.

Unbiased automatic quantification of PET amyloid brain imaging

Subject

Positron emission tomography (PET) imaging is a well-established technique to quantify abnormal loads of ?-amyloid (A?) plaques depositions in the brain, one of the hallmarks of Alzheimer’s disease (AD), proving high diagnostic accuracy at the early stages of the disease. Standardized uptake value ratios (SUVR), the most common normalization technique, is performed by computing the ratio of the PET image intensity in the whole brain to that in cerebellar gray matter. These brain regions are determined from a structural Magnetic Resonance Imaging (MRI) scan from the brain of the same patient. This process involves many computationally complex steps, including image alignment, normalization and segmentation that, depending on the specific pipeline followed, may produce different results [1]. To overcome this issue, some authors have proposed automatic processing tools based on Deep Learning (DL) that estimate the average SUVR directly from the PET image alone [2].
One drawback of these methods is that rich spatial information is lost, because a single value of the average SUVR is estimated. In addition, DL methods are at high risk of overfitting to training dataset, especially in medical imaging, where the number of subjects is much less than the number of dimensions. This produces biased models and overconfident performance estimations. Fitting to confounders occurs when extraneous variables are correlated with some aspect of the image and also with the disease prevalence, and therefore the SUVR score. If we are not aware of the presence of confounders the model could wrongly learn to predict a SUVR score from an image feature that is not related to amyloid burden, but with the confounder.
This problem is solved in classical statistical analysis by including the confounder as a term in the regression, but it is much harder to fix in DL. Recently, some DL frameworks are being proposed to produce low dimensional features that are highly predictive of an outcome of interest (dementia diagnosis in our case) but do not include any information on potential confounders (PET compound, scanner model, country, etc.), reducing the bias and improving the model generalizability [3]. This task is not trivial, as in this process of eliminating the confounder information, one should take care to not eliminate also useful discriminative information.

Kind of work

We will explore the potential of confounder-free deep learning models to learn low dimensional PET imaging features with high predictive discrimination power to detect future AD development. In addition, we will aim to identify different patterns of brain amyloid load that characterize different subtypes of AD.

Framework of the Thesis

This thesis is part of an ongoing project at ETRO together with the UZ Brussel hospital that is developing tools for the early detection of Alzheimer's disease.

Resources:
[1] Kolinger GD, et al. (2021) Amyloid burden quantification depends on PET and MR image processing methodology. PLoS ONE 16(3): e0248122. https://doi.org/10.1371/journal.pone.0248122
[2] Kim JY, et al. (2019) Alzheimer’s Disease Neuroimaging Initiative. Amyloid PET Quantification Via End-to-End Training of a Deep Learning. Nucl Med Mol Imaging. 53(5): 340-348. https://doi.org/10.1007/s13139-019-00610-0
[3] Zhao, Q., et al. (2020) Training confounder-free deep learning models for medical applications. Nat Commun 11, 6010. https://doi.org/10.1038/s41467-020-19784-9

Number of Students

1

Expected Student Profile

- Following an MSc in a field related to one or more of the following: electrical engineering, computer science, applied mathematics, Biomedical Engineering
- Experience with signal processing.
- Experience with machine learning and statistics.
- Strong programming skills (python, matlab, R, …).
- Ability to write scientific reports and communicate research results at conferences in English.

Promotor

Prof. Hichem Sahli

+32 (0)2 629 2916

hsahli@etrovub.be

more info

Supervisor

Dr. Matías Bossa

+32 (0)2 629 1529

mnbossa@etrovub.be

more info

Image

- Contact person

- IRIS

- AVSP

- LAMI

- Contact person

- Thesis proposals

- ETRO Courses

- Contact person

- Spin-offs

- Know How

- Journals

- Conferences

- Books

- Vacancies

- News

- Events

- Press

Contact

ETRO Department

Tel: +32 2 629 29 30

©2022 • Vrije Universiteit Brussel • ETRO Dept. • Pleinlaan 2 • 1050 Brussels • Tel: +32 2 629 2930 (secretariat) • Fax: +32 2 629 2883 • WebmasterDisclaimer