Article
Contouring of tumour outlines in 2D-Ultrasonography – intrarater and interrater error
Konturierung von Tumorgrenzen im 2D-Ultraschall – intrarater und interrater Fehler
Search Medline for
Authors
Published: | May 4, 2005 |
---|
Outline
Text
Objective
Intraoperative Sonography (ioSO) is used for localisation and control of resection in brain tumour surgery. Definition of the surgical tumor-brain interface or border prior to resection has been reported to be difficult in ioSO and special experience has been requested for the use of ioSO. This study was performed to quantify the intra- and interrater error of ioSO in different tumour entities.
Methods
49 ioSO images from nine different tumours (2 meningeomas, 2 metastases, 3 low grade gliomas (LGG), 2 high grade gliomas (HGG)) were taken prior to dura opening and stored for evaluation (Siemens Elegra Advanced, Germany). The tumour borders were outlined by the surgeon after tumour resection (reference contour) and a reference pixel near the tumour centre was defined within each image. All images were mailed to 6 physicians (3 board certified neurosurgeons with ioSO experience; 1 paediatric radiologist; 2 neurosurgical residents) for digital outlining of the surgical tumour borders (test contours). The individual manual error was estimated from contouring a phantom image with distinct borders. Differences of test contours to reference were calculated in mm on radial segments (related to the reference pixel). Mean difference in mm and frequency of differences more than 3mm from reference (FOD 3mm) were calculated (U-Test and single-sample t-test; level of significance 0.05). Intrarater error was calculated from re-contouring of the reference images after 6 months.
Results
Mean difference (±SD) in test phantom contouring was 0.1±0,42mm without significant differences between physicians. Mean intrarater error ranged -0.17±1.09mm (meningeoma) to 0,44±1,48mm (metastasis). Extreme values ranged -8.4mm to 11.4mm in LGG but more than 80% of contour points were located within -1.9 to +1.6mm from reference. FOD 3mm was significantly different from zero in HGG and LGG (p<0.05). Mean interrater error was highest in LGG (1.37±4.22mm) with 80% of point located between -1.89 to 6.94mm from reference. FOD 3mm was significantly different from zero for LGG (radiologist) and for LGG and HGG (neurosurgeons). There was no significant difference between examiners for meningeomas.
Conclusions
Definition of surgical tumour borders by ioSO is difficult in gliomas with acceptable mean errors but a high variability. General knowledge of sonography images seems to improve tumour delineation on ioSO images.