Brain Tumor Segmentation with Deep Learning.

Physiology-informed brain tumor segmentation using DSC MRI hemodynamic properties and learning-based image analysis.

Project

— Deep Learning
— Medical Imaging
— Brain Tumors

Brain tumor segmentation workflow combining structural MRI, DSC MRI hemodynamic features, and deep learning.

Publication→

Notes

This project explores how DSC MRI hemodynamic information can add physiological context to brain tumor segmentation. By combining structural segmentation with density-based analysis of perfusion time courses, the method highlights tissue behavior that conventional morphology-only segmentation can miss.

Project Overview

Brain tumor segmentation is a critical task in medical imaging, especially when tumor tissue is heterogeneous and difficult to characterize from structural scans alone. This project addresses that problem by:

Integrating hemodynamic data with traditional imaging modalities
Combining U-Net segmentation with density-based analysis of DSC MRI time courses
Adding physiological context to segmented regions for more interpretable image analysis

Key Contributions

Incorporation of DSC MRI hemodynamic properties into a segmentation workflow
More physiologically informed characterization of heterogeneous tumor regions
A framework for comparing structural segmentation with perfusion-derived tissue behavior

Research Impact

This work contributes to AI-assisted medical imaging and has potential applications in:

Clinical research: Physiological context for tumor boundary interpretation
Treatment planning: Better visualization for surgical and radiation therapy
Research: Enhanced understanding of tumor hemodynamics

The research has been published in Quantitative Imaging in Medicine and Surgery and represents a step toward segmentation workflows that account for both image structure and tissue physiology.