A Flexible Semi-Automatic Approach for Glioblastoma multiforme Segmentation

TL;DR

This paper introduces a flexible semi-automatic segmentation method for grade IV gliomas in MRI scans, combining graph-based algorithms with user input to improve accuracy and efficiency in tumor volume assessment.

Contribution

The novel segmentation scheme uses a directed 3D graph and minimal cost s-t cut, allowing user-guided refinement for glioma segmentation in MRI images.

Findings

Achieved an average Dice Similarity Coefficient of 77.72% with the proposed method.

Compared favorably against a one-click segmentation approach, with DSC of 83.91%.

Tested on 12 MRI datasets with manual ground truth annotations.

Abstract

Gliomas are the most common primary brain tumors, 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 segmentation methods. In this paper, a flexible semi-automatic approach for grade IV glioma segmentation is presented. The approach uses a novel segmentation scheme for spherical objects that creates a directed 3D graph. Thereafter, the minimal cost closed set on the graph is computed via a polynomial time s-t cut, creating an optimal segmentation of the tumor. The user can improve the results by specifying an arbitrary number of additional seed points to support the algorithm with grey value information and geometrical constraints. The presented method is tested on 12 magnetic resonance imaging datasets. The ground truth of the tumor boundaries are manually extracted by neurosurgeons. The segmented gliomas are compared with a one click method, and the semi-automatic approach yields an average Dice Similarity Coefficient (DSC) of 77.72% and 83.91%, respectively.