Abstracts

Rationale: Temporal lobectomy (TL) is an effective treatment for medication refractory temporal epilepsy. But not all benefit, 20-40% of patients still have seizures. Pre-surgical neuroimaging, typically MRI and FDG-PET, are qualitatively read and hence subjective. It is unclear how this subjectivity affects surgical outcomes. Here the variability in imaging interpretation is assessed by comparing the radiology reports to the epilepsy surgery conference notes and then to outcomes/pathology. Methods: The epilepsy surgery database at MGH was queried for patients that underwent TL. Inclusion criteria were, age>18, epilepsy surgery conference note, MRI and PET imaging at MGH. Patients were excluded if they had a structural lesion other than hippocampal sclerosis (HS), prior epilepsy surgery, or surgery beyond TL. The imaging reports were categorized on a scale of 0 for no findings of HS or temporal lobe hypometabolism (HM), 1 for mild/possible/subtle, and 2 for definite/probable. 2X2 contingency tables were used for comparison of sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV). Seizure-free at 1 year follow-up was used to define a good outcome (ILAE 1). The relationship between interpretation and outcomes was assessed using logistic regression. Results: There were 104 subjects in the study, (incomplete data 44 subjects). Mean age at surgery 40 yrs, mean follow-up 36 mnths, 70% of the patients were ILAE class 1 at 1year, 66% had HS on pathology. The radiology report and the conference consensus were different in 33% of FDG-PET and 45% of MRIs. The FDG-PET radiology (ILAE 1 as standard) had sensitivity 0.98, specificity 0.20, PPV 0.73, NPV 0.83; FDG-PET conference had sensitivity 1.0, specificity 0.14, PPV 0.70, NPV 1.0. The MRI radiology (HS on pathology as standard) had sensitivity 0.76, specificity 0.52, PPV 0.77, NPV 0.5; MRI conference had sensitivity 0.91, specificity 0.33, PPV 0.73, NPV 0.64. Overall the trend was for conference to have a higher sensitivity and lower specificity. Univariate/multivariate logistic regression results are in Table 1. PET was a significant predictor of outcome and MRI was significant predictor of HS. Comparison of proportions showed a significant difference in outcomes based on PET interpretation, but not MRI. On multivariate regression PET remained significant predictor of outcome (adjusted for MRI) in radiology group. MRI remained a significant predictor (adjusted for PET) for presence of HS. Conference results in general had a higher R^2. Conclusions: The interpretation of imaging findings of mesial temporal lobe epilepsy on MRI and FDG-PET varied between the radiology report and the epilepsy conference in 33% of PETs and 45% of MRIs. Overall radiology was more specific and the epilepsy conference was more sensitive. These differences may stem from the influence of clinical/electrographic data and a different threshold for abnormal. The major limitation is selection bias secondary to only including surgical patients. Quantitative imaging techniques could help standardize image interpretation.