Joint Estimation of Neural and Hemodynamic Responses in Functional Ultrasound Using Variational Inference

Master Thesis (2025)
Author(s)

T. Fujioka (TU Delft - Mechanical Engineering)

Contributor(s)

K. Batselier – Mentor (TU Delft - Team Kim Batselier)

Sofia Eirini Kotti – Graduation committee member (TU Delft - Signal Processing Systems)

Borbala Hunyadi – Graduation committee member (TU Delft - Signal Processing Systems)

Faculty
Mechanical Engineering
More Info
expand_more
Publication Year
2025
Language
English
Graduation Date
09-07-2025
Awarding Institution
Delft University of Technology
Programme
['Mechanical Engineering | Systems and Control']
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

Functional Ultrasound (fUS) is an emerging neuroimaging technique capable of capturing brain activity similarly to functional magnetic resonance imaging (fMRI) but with higher spatiotemporal resolution and lower operational cost. This thesis investigates the extension of the joint detection-estimation (JDE) framework to jointly estimate both the hemodynamic response function (HRF) and neural response function (NRF) from fUS imaging data. In the proposed model, the two response functions are represented as a cascade linear time-invariant (LTI) convolution systems, enabling indirect estimation of neural activity signals from fUS measurements. Since direct recordings of neural activity are often unavailable, this Bayesian approach offers a data-driven means of probing the brain’s functional organization.
Inference is performed within a coordinate ascent variational inference (CAVI) framework. The proposed algorithm was applied to fUS datasets and validated against simultaneous recordings of neural firing rates. Results demonstrate that the model successfully captures neural activity, achieving a Pearson correlation coefficient (PCC) of approximately 0.24-0.30, and provides a modest improvement over the conventional boxcar input stimulus model. Additionally, the jointly estimated HRFs were consistent with existing literature, and regional HRF estimates revealed differences in response dynamics between the visual cortex and hippocampus, highlighting region-specific hemodynamic properties.

Files

MSc_Thesis_TFujioka.pdf
(pdf | 13.9 Mb)
License info not available