Predicting time-to-symptom-onset in genetic frontotemporal dementia using machine learning based on multimodal MRI

Master Thesis (2025)
Author(s)

G.G. Bregman (TU Delft - Mechanical Engineering)

Contributor(s)

F.M. Vos – Graduation committee member (TU Delft - ImPhys/Computational Imaging)

E. E. Bron – Graduation committee member (Biomedical Imaging Group Rotterdam)

Julia Neitzel – Graduation committee member (Erasmus MC)

M.F. van Haaften – Graduation committee member (Erasmus MC)

Faculty
Mechanical Engineering
More Info
expand_more
Publication Year
2025
Language
English
Coordinates
51.9109, 4.4681
Graduation Date
21-05-2025
Awarding Institution
Delft University of Technology
Programme
['Biomedical Engineering | Neuromusculoskeletal Biomechanics']
Sponsors
Erasmus MC
Faculty
Mechanical Engineering
Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Abstract

Genetic frontotemporal dementia (FTD) is a heterogeneous neurodegenerative disease primarily caused by pathogenic mutuations in one of three genes: \textit{C9orf72}, \textit{MAPT}, and \textit{GRN}. Accurately predicting time-to-symptom-onset could improve clinical care and patient stratification in clinical trials. This study aimed to develop a machine learning framework for individualized prediction of symptom onset in genetic FTD using multimodal MRI data under limited sample size conditions. We explored two strategies: (i) a support vector machine (SVM) classifier distinguishing symptomatic FTD from non-FTD scans, using the distance to the decision boundary (DDB) as a proxy for time-to-symptom-onset. While DDB values increased as conversion approached, they lacked precision for individual-level prediction. (ii) Five binary classifiers, each trained to predict conversion within a different time window (1–5 years) before symptom onset. Combining outputs from these classifiers yielded personalized onset predictions with a mean absolute error (MAE) of 0.71 years, outperforming a linear elastic-net regression baseline (MAE = 2.97 years). This approach successfully predicted symptoms five years prior to conversion, which represents an important step toward personalized medicine in genetic FTD.

Files

Thesis_5_.pdf
(pdf | 0 Mb)
License info not available
warning

File under embargo until 31-12-2026