This poster is published under an
open license. Please read the
disclaimer for further details.
Keywords:
Neuroradiology brain, CT, Observer performance, Trauma
Authors:
T. Vande Vyvere1, L. Claes2, L. van den Hauwe1, G. Wilms2, D. Smeets2, P. M. Parizel1; 1Antwerp/BE, 2Leuven/BE
DOI:
10.1594/ecr2016/B-0205
Results
Almost perfect overall agreement was found between the non-radiologist raters and the standard (NS1 and NR1,
κ=0.958,
NS2 and NR1,
κ=0.915; 95%,
see fig.
6).
Similar agreement was found between the non-radiologist raters seperately (NS1 and NS2,
κ=0.958; 95% CI,
see fig.
6).
Agreement for individual TBI pathoanatomic lesion types was good to almost perfect for all lesion types (range from κ=0.638 to κ=1; see fig.
6).
The lowest agreement was found for patients with diffuse axonal injuries and cerebral herniation.
Perfect agreement was found for midline shift between all raters.
Agreement for the classification scores was almost perfect between all raters as well (Marshall scale: NS1 and NR1,
κ= 0.938,
NS2 and NR1,
κ=0.957 and ; Rotterdam scale: NS1 and NR1,
κ=0.949,
NS2 to NR1,
κ=0.912,
see fig.
6).
Diagnostic accuracy for reporting abnormalities was high (NS1: 94.67%; NS2: 95.00%),
as was diagnostic performance (NS1: Sensitivity: 92.81,
CI [87.50,96.36],
Specificity: 95.46 CI [93.15,97.17]; NS2: Sensitivity: 93.38 CI [87.81,96.93],
Specificity: 95.47 CI [93.16,97.18],
see fig.
7 and 8).