Student Name

Date of Submission

Spring 2026

Supervisor

Dr. Tahir Qasim Syed, Assistant Professor, Department of Computer Science

Co-Supervisor

Dr. Behraj Khan

Committee Member 1

Dr. Tariq Mahmood

Committee Member 2

Dr. Muhammad Atif Tahir

Degree

Master of Science in Data Science

Department

Department of Computer Science

Faculty/ School

School of Mathematics and Computer Science (SMCS)

Keywords

Magnetic Resonance Imaging, Infant Brain, Volumetric Hippocampus Segmentation, Foundation Models, 3D Window-Based Disassembly-Reassembly

Abstract

Precise volumetric delineation of hippocampal structures is essential for quantifying neurodevelopmental trajectories in pre-term and term infants, where subtle morphological variations may carry prognostic significance. While foundation encoders trained on large-scale visual data offer discriminative representations, their 2D formulation is a limitation with respect to the 3D organization of brain anatomy. This thesis proposes a volumetric segmentation strategy that adapts frozen DINOv3 features to infant MRI through a 3D decoding framework. This is done through a structured window-based disassembly-reassembly mechanism: the global MRI volume is decomposed into nonoverlapping 3D windows or sub-cubes, each processed via a separate decoding arm built upon frozen high-fidelity features, and subsequently reassembled prior to a ground-truth correspondence using a dense-prediction head. This design preserves a constant decoder memory footprint while forcing predictions to lie within an anatomically consistent geometry. Evaluated on the ALBERT and LISA datasets for hippocampal segmentation, the proposed approach achieves a Dice score of 0.65 and 0.51 for a single 3D window, respectively. The method demonstrates that volumetric anatomical structure could be recovered from frozen 2D foundation representations through structured 3D compositional decoding, offering a preliminary but principled direction for adaption vision foundation models to 3D medical image segmentation.

Document Type

Restricted Access

Submission Type

Thesis

Available for download on Sunday, August 31, 2031

Share

COinS