DEPARTMENT OF COMPUTATIONAL AND DATA SCIENCES
M.Tech Research Thesis Colloquium
Speaker : Mr. Akhil
S.R. Number : 06-18-01-10-22-23-1-23206
Title : “An Explainable Adaptive Learning Framework for Brain Tissue Extraction from MRI”
Research Supervisor : Prof. Debnath Pal
Date & Time : December 22, 2025, 04.00 PM
Venue: The Thesis Colloquium will be held on HYBRID Mode
# 102 CDS Seminar Hall /MICROSOFT TEAMS
Please click on the following link to join the Thesis Colloquium:
MS Teams link
ABSTRACT
While Deep Learning currently dominates neuroimaging, these models suffer from a critical, often overlooked scalability bottleneck: model drift caused by variations in scanner physics and subject positioning, which renders them unstable on unseen datasets. To address this limitation, we adopted a systematic approach in this thesis, moving beyond the limitations of opaque “black box” neural networks to engineer a transparent, patient-diverse solution. Our investigation progressed through a rigorous analysis of failure modes, identifying that early intensity-based methods failed on complex structures where non-brain tissues share the same profile as brain tissue , motivating a shift toward texture-based feature extraction. However, intermediate multi-classifiers trained on central slices proved inadequate, misclassifying valid brain tissue. These key learnings allowed a refinement of the approach to an adaptive learning framework centered on a “Smart Anchor” strategy that dynamically selects the optimal reference slice rather than assuming a fixed centre. By incorporating robust preprocessing and post-processing techniques for inter-slice consistency and a Two-Specialist strategy that decouples core identification from boundary refinement, we eliminated error propagation while ensuring interpretability. This systematic development yielded a robust pipeline that achieves a Dice score of 93 ± 2%, demonstrating a high-accuracy, clinically scalable brain tissue extraction technique through reproducible design that allows explainability.
ALL ARE WELCOME
