A MATLAB-Based Subject-Specific Framework for Vessel Quantification and Localization in Longitudinal Photoacoustic Neuroimaging

Photoacoustic imaging (PAI) enables visualization of cerebral blood vessels and the measurement of hemodynamic responses, at single vessel resolution, to neural stimulation. However, existing approaches do not provide a unified and systematic framework for consistently localizing vessels and quantifying vessel related functional responses across multiple sessions from the same subject in a longitudinal setting.This thesis presents a MATLAB-based subject-specific framework for vessel quantification and localization in longitudinal photoacoustic brain images, developed and evaluated on data from a single Non-Human Primate (NHP) under 3mA and 8mA peripheral electrical stimulation. The framework integrates frame registration, interactive vessel selection, signal extraction, and amplitude quantification into a single reproducible protocol, implemented as PAI VascuTrack, a standalone MATLAB application that integrates key framework stages into a single graphical interface for longitudinal photoacoustic brain image analysis. A U-Net deep learning model was trained on manually annotated frames for vessel localization, evaluated under two conditions: intra-session and cross-session. Detection accuracy was 0.99 under both evaluation conditions. Under intra-session evaluation, localization accuracy was 0.99 with a mean error of 3.01 pixels. Under cross-session evaluation, localization accuracy dropped to 0.56 with a mean error of 129.57 pixels, reflecting the limited training variability in a small single-subject dataset. For amplitude quantification, which represents the brain functional change, moving maximum followed by moving average was applied to suppress cardiac-frequency oscillations while preserving the underlying hemodynamic response. The framework was validated on an independent dataset from a second NHP subject, confirming correct detection of stimulus-evoked responses on an independent dataset. When applied to the primary dataset, the 3mA condition produced more consistent responses across sessions, while 8mA produced stronger but more variable responses. This work provides a practical and reusable tool for hemodynamic analysis in longitudinal photoacoustic neuroimaging, with future application in the study of neurovascular changes following brain injury.

Name thesis:degree_name

M.S.

Discipline thesis:degree_discipline

Bioengineering

Grantor dc:publisher

University of Kansas

Year dc:date.issued

2026

Author dc:creator

• Teklehaimanot, Eleny Mulugeta

Advisor dc:contributor.advisor

• Yang, Xinmai

• Hemodynamic response
• Longitudinal neuroimaging
• Photoacoustic imaging
• Subject-specific study
• Vessel localization
• Vessel responce quantification

Language dc:language.iso

en

Dc Identifier Other

https://www.proquest.com/LegacyDocView/DISSNUM/32702689

OAI identifier oai:identifier

oai:kuscholarworks.ku.edu:1808/39523