Abstracts

To determine agreement levels between two independent radiologists interpreting MR images of intractable seizure patients. As part of a multicenter study of epilepsy surgery, 2 radiologists independently evaluated 512 preop MR scans. A standard data form was used. Quantitative measurements were not performed.
Hippocampal size, signal, and hippocampal sclerosis (HS) diagnosis were assessed on a 5-point scale (definitely normal=1, probably normal=2, equivocal=3, probably abnormal=4, definitely abnormal=5). Presence of periventricular heterotopia, cortical thickening, sulcal morphologic changes, focal volume loss, [amp] temporal lobe encephalocele were assessed on a 3-point scale (no, equivocal, yes), along with lesion location and diagnosis. Weighted Kappa assessed interrater agreement. MRI lesions occurred in 70%: HS 44%, dual path 11%, atrophy 5%, development 4%, tumor 4%, vascular 3%. In the STUDY group, number of abnormalities both raters agreed were present: hippocampal changes=222; volume loss=34; cortical thickening=9; sulcal changes=5; heterotopia=4; encephalocele=0.
Interrater agreements were excellent for hippocampal volume, hippocampal signal changes and MR diagnosis of HS (wKappas 0.80-0.83); but rater bias was present with rater 2 scoring higher than rater 1 on the 5-point scale in 76.5% of the disagreed cases (p[lt].0001). For all 5-point assessments, categories of Definitely Normal or Abnormal produced excellent wKappas, while Probable categories produced good-fair wKappas, and Equivocal category produced poor wKappas. Good agreement occurred for cortical thickening and location (wKappas 0.70-0.72). Agreement was fair for volume loss and heterotopia and poor for sulcal changes and encephaloceles.[table1][table2] There was excellent inter-rater agreement for hippocampus abnormalities and rater bias did not preclude this robust agreement. Kappas for other assessments ranged from good to poor, in part due to the relatively rare occurrence of these abnormalities, lack of a focused region of brain (unlike the hippocampus) and the subjective nature of these observations. With a 5-point scale, ratings at the ends of the scale were more robust than in the scale[apos]s center and changing to a 3-point scale would mask this phenomenon. In summary, qualitative assessment of MR images of epilepsy patients using a systematic approach can reliably detect and localize hippocampal and other brain lesions. This is crucial for surgical planning, which demands robust preoperative measures. (Supported by NIH grant IR01 NS32375-01)