Evaluation of a Novel Approach for Automatic Volume Determination of Glioblastomas Based on Several Manual Expert Segmentations

TL;DR

This paper introduces an automatic method for glioblastoma volume measurement using a graph-based segmentation approach, validated against expert manual segmentations on MRI data, aiming to improve efficiency and consistency.

Contribution

The paper presents a novel graph-based segmentation technique for glioblastomas that automates volume determination using minimal s-t cut, validated against expert manual segmentations.

Findings

The automatic segmentation achieved high Dice Similarity Coefficient scores.

Manual and automatic segmentations showed strong correlation.

Inter- and intra-physician variability was quantified.

Abstract

The glioblastoma multiforme is the most common malignant primary brain tumor and is one of the highest malignant human neoplasms. During the course of disease, the evaluation of tumor volume is an essential part of the clinical follow-up. However, manual segmentation for acquisition of tumor volume is a time-consuming process. In this paper, a new approach for the automatic segmentation and volume determination of glioblastomas (glioblastoma multiforme) is presented and evaluated. The approach uses a user-defined seed point inside the glioma to set up a directed 3D graph. The nodes of the graph are obtained by sampling along rays that are sent through the surface points of a polyhedron. After the graph has been constructed, the minimal s-t cut is calculated to separate the glioblastoma from the background. For evaluation, 12 Magnetic Resonance Imaging (MRI) data sets were manually segmented slice by slice, by neurosurgeons with several years of experience in the resection of gliomas. Afterwards, the manual segmentations were compared with the results of the presented approach via the Dice Similarity Coefficient (DSC). For a better assessment of the DSC results, the manual segmentations of the experts were also compared with each other and evaluated via the DSC. In addition, the 12 data sets were segmented once again by one of the neurosurgeons after a period of two weeks, to also measure the intra-physician deviation of the DSC.