Sequential glioblastoma segmentation via topological data analysis and spatial adjacency.

Area of Science:

• Medical Imaging
• Computational Biology
• Radiology

Background:

• Glioblastoma segmentation in medical imaging is difficult due to irregular shapes and unclear boundaries.
• Conventional methods often struggle with accuracy, necessitating advanced techniques.

Purpose of the Study:

• To introduce a novel segmentation framework for glioblastoma using Topological Data Analysis (TDA).
• To improve segmentation accuracy and structural fidelity while reducing the need for extensive annotated data.

Main Methods:

• The framework utilizes TDA, including filtrations and persistent homology, combined with spatial adjacency information.
• It segments Whole Tumor (WT), Enhancing Tumor (ET), and Tumor Core (TC) with Edema (ED) sequentially.
• A novel fuzzy Edge-Dice score was developed for performance evaluation.

Main Results:

• The TDA-based framework demonstrated robust and accurate glioblastoma segmentation on BRATS2021 and BRATS2022-Reg datasets.
• The method effectively captures intrinsic topological features for improved delineation.
• Achieved competitive or superior results compared to conventional deep learning methods.

Conclusions:

• TDA presents a promising approach for accurate and reliable glioblastoma segmentation in medical imaging.
• This framework offers a valuable alternative or complement to existing deep learning techniques.
• The potential of TDA methods in clinical applications warrants further investigation.