Reliability of Semi-Automated Segmentations in Glioblastoma.
Clin Neuroradiol
; 27(2): 153-161, 2017 Jun.
Article
em En
| MEDLINE
| ID: mdl-26490369
ABSTRACT
PURPOSE:
In glioblastoma, quantitative volumetric measurements of contrast-enhancing or fluid-attenuated inversion recovery (FLAIR) hyperintense tumor compartments are needed for an objective assessment of therapy response. The aim of this study was to evaluate the reliability of a semi-automated, region-growing segmentation tool for determining tumor volume in patients with glioblastoma among different users of the software.METHODS:
A total of 320 segmentations of tumor-associated FLAIR changes and contrast-enhancing tumor tissue were performed by different raters (neuroradiologists, medical students, and volunteers). All patients underwent high-resolution magnetic resonance imaging including a 3D-FLAIR and a 3D-MPRage sequence. Segmentations were done using a semi-automated, region-growing segmentation tool. Intra- and inter-rater-reliability were addressed by intra-class-correlation (ICC). Root-mean-square error (RMSE) was used to determine the precision error. Dice score was calculated to measure the overlap between segmentations.RESULTS:
Semi-automated segmentation showed a high ICC (> 0.985) for all groups indicating an excellent intra- and inter-rater-reliability. Significant smaller precision errors and higher Dice scores were observed for FLAIR segmentations compared with segmentations of contrast-enhancement. Single rater segmentations showed the lowest RMSE for FLAIR of 3.3 % (MPRage 8.2 %). Both, single raters and neuroradiologists had the lowest precision error for longitudinal evaluation of FLAIR changes.CONCLUSIONS:
Semi-automated volumetry of glioblastoma was reliably performed by all groups of raters, even without neuroradiologic expertise. Interestingly, segmentations of tumor-associated FLAIR changes were more reliable than segmentations of contrast enhancement. In longitudinal evaluations, an experienced rater can detect progressive FLAIR changes of less than 15 % reliably in a quantitative way which could help to detect progressive disease earlier.Palavras-chave
Texto completo:
1
Coleções:
01-internacional
Base de dados:
MEDLINE
Assunto principal:
Neoplasias Encefálicas
/
Reconhecimento Automatizado de Padrão
/
Interpretação de Imagem Assistida por Computador
/
Glioblastoma
/
Imageamento Tridimensional
/
Imagem de Difusão por Ressonância Magnética
Tipo de estudo:
Diagnostic_studies
Limite:
Female
/
Humans
/
Male
/
Middle aged
Idioma:
En
Revista:
Clin Neuroradiol
Ano de publicação:
2017
Tipo de documento:
Article