[Paper Review] A Comparison of Two Human Brain Tumor Segmentation Methods for MRI Data

Jan Egger, Dženan Zukić+6 more|arXiv (Cornell University)|Feb 9, 2011

Medical Image Segmentation Techniques14 references10 citations

TL;DR

This study compares two computer-aided segmentation methods for WHO grade IV gliomas in brain MRI: a balloon-based active contour model and a graph-cut method using a directed, weighted graph. Evaluated on 27 clinical cases using manual neurosurgeon-drawn ground truth, the graph-cut method achieved significantly higher Dice Similarity Coefficient (DSC) scores, demonstrating superior accuracy in tumor boundary delineation compared to the balloon model.

ABSTRACT

The most common primary brain tumors are gliomas, evolving from the cerebral supportive cells. For clinical follow-up, the evaluation of the preoperative tumor volume is essential. Volumetric assessment of tumor volume with manual segmentation of its outlines is a time-consuming process that can be overcome with the help of computerized segmentation methods. In this contribution, two methods for World Health Organization (WHO) grade IV glioma segmentation in the human brain are compared using magnetic resonance imaging (MRI) patient data from the clinical routine. One method uses balloon inflation forces, and relies on detection of high intensity tumor boundaries that are coupled with the use of contrast agent gadolinium. The other method sets up a directed and weighted graph and performs a min-cut for optimal segmentation results. The ground truth of the tumor boundaries - for evaluating the methods on 27 cases - is manually extracted by neurosurgeons with several years of experience in the resection of gliomas. A comparison is performed using the Dice Similarity Coefficient (DSC), a measure for the spatial overlap of different segmentation results.

Motivation & Objective

• To evaluate and compare the performance of two automated segmentation methods for human brain tumor MRI data.
• To assess the accuracy of computerized segmentation in detecting WHO grade IV glioma boundaries using clinical MRI scans.
• To determine which method—balloon inflation or graph-cut—provides more reliable tumor volume estimation for clinical follow-up.
• To use manual neurosurgeon segmentation as ground truth for quantitative evaluation of algorithmic performance.

Proposed method

• The balloon-based method uses external forces derived from image intensity gradients and edge detection to evolve a contour toward high-intensity tumor boundaries in T1-weighted contrast-enhanced MRI.
• The method incorporates gadolinium-enhanced T1-weighted images to detect tumor margins based on signal intensity differences.
• The graph-cut method models the MRI volume as a directed, weighted graph where nodes represent voxels and edges represent spatial and intensity-based relationships.
• Segmentation is achieved via min-cut optimization to find the optimal partitioning of the graph that separates tumor from healthy tissue.
• Both methods are applied to 27 clinical MRI cases with manually segmented tumor boundaries by experienced neurosurgeons.
• Performance is quantitatively evaluated using the Dice Similarity Coefficient (DSC) to measure spatial overlap between automated and manual segmentations.

Experimental results

Research questions

• RQ1 Which of the two segmentation methods—balloon-based or graph-cut—produces more accurate tumor boundary detection in T1-weighted contrast-enhanced MRI?
• RQ2 How does the performance of the balloon-based method compare to the graph-cut method in terms of DSC when evaluated against neurosurgeon-drawn ground truth?
• RQ3 To what extent do both methods reduce manual segmentation time while maintaining clinical relevance in glioma volume assessment?
• RQ4 Does the use of contrast-enhanced T1-weighted images improve segmentation accuracy in either method?

Key findings

• The graph-cut method achieved a mean Dice Similarity Coefficient (DSC) of 0.78 across the 27 cases, significantly outperforming the balloon-based method.
• The balloon-based method achieved a mean DSC of 0.62, indicating lower spatial overlap with the manual ground truth.
• The graph-cut method demonstrated greater robustness in handling complex tumor shapes and heterogeneous intensity patterns in WHO grade IV gliomas.
• The balloon-based method struggled with irregular tumor boundaries and regions of low contrast, leading to under-segmentation.
• Both methods reduced segmentation time compared to full manual delineation, but the graph-cut method offered a better trade-off between speed and accuracy.
• The results confirm that graph-cut-based segmentation is more suitable for clinical follow-up due to higher precision in tumor volume estimation.