Introduction
One-third of the focal epilepsy cases are medically refractory, necessitating surgical intervention. However, only two third of them demonstrate MRI lesions. Further investigation with invasive stereoelectroencephalography (SEEG) is required in absence of epileptogenic lesion identification in MRI. Voxel-based analysis, a non-invasive method, when merged with non-parametric combination, a multivariate analysis technique, allows joint inference of different MRI modalities collected from the same subjects.
Objective
To evaluate the efficacy of non-parametric combination in pre-surgical evaluation of MRI-negative refractory focal epilepsy.
Patients and Methods
69 medically refractory focal epilepsy undergoing pre-surgical evaluation along with 62 healthy controls were included in the study. The patients with focal epilepsy were divided into:
- Discrete lesions on MRI scans (n=42)
- No visible findings on MRI scans (n=27)
Assessment
- MRI Imaging
- Standard T1-weighted imaging
- Fluid-attenuated inversion recovery (FLAIR)
- Fractional anisotropy (FA) and Mean diffusivity (MD) from diffusion tensor imaging
- Neurite density index (NDI) from neurite orientation dispersion and density imaging
Results
Univariate analysis
- 26 out of 42 patients had visually concordant findings in NDI, providing a sensitivity of 62%.
- All other modalities had sensitivities below 50%.
- GMC yielded the least optimal sensitivity measures at 29% and 26% for increased and decreased GMC, respectively.
- Decreased NDI provided the highest dice score of 0.14, while all other univariate analyses resulted in dice scores ≤ 0.10.
- Decreased NDI provided true-positive results in average 20% of lesional voxels in MRI-positive patients.
- 7–8% true-positive rates were observed in increased FLAIR, increased MD and decreased FA.
- 2–3% of lesional voxels were detected by Increased and decreased GMC.
- All univariate analyses had voxel-wise false-positive rates ≤ 2%.
- In patients from the MRI-negative cohort with conclusive ground truth, 33% of patients showed abnormalities similar to seizure onset zone from decreased NDI.
- Increased FLAIR and MD both yielded sensitivity measures of 28%.
- Decreased FA and increased GMC demonstrated 22% MRI-negative group sensitivity.
- Decreased GMC had the poorest sensitivity at 17%.
- In MRI-negative subjects with inconclusive ground truth, significant findings were seen in 1 out of 9 patients via decreased FA and increased FLAIR analyses while in 2 out of 9 patients via decreased NDI.
- No significant findings across all MRI negative, SEEG inconclusive subjects were observed using increased GMC, decreased GMC and increased MD analyses.
- The leave-one-out cross validation of controls from the unimodal analyses had high specificity measures.
- Decreased FA, decreased NDI and decreased GMC showed a specificity of 97%.
- Increased FLAIR and GMC had a specificity of 98%, while increased MD had a specificity of 100%.
Non-parametric combination
- In the MRI-positive group, 34 patients’ analysed using non-parametric combinations were visually concordant with ground truth lesion masks, having a sensitivity of 81%.
- In MRI-positive patients, non-parametric combination had a dice score of 0.19, a voxel-wise true-positive rate of 41% and a false-positive rate of 6%.
- In MRI-negative cohort with conclusive ground truth, non-parametric combination identified abnormalities visually similar to seizure onset zone with a sensitivity of 50%.
- In those with inconclusive SEEG, significant findings were observed in 2 patients.
- An average specificity from the non-parametric combination analysis was 97%.
Conclusion
The use of a multimodal Voxel-based analysis with Non-parametric combination for detecting epileptogenic lesions in MRI-negative focal epilepsy was found to be superior to univariate analyses with 50% sensitivity.