Systematic Assessment of Multispectral Voxel-Based Morphometry in Previously MRI-Negative Focal Epilepsy

R. Kotikalapudi; P. Martin; J. Marquetand; T. Lindig; B. Bender; N.K. Focke

doi:10.3174/ajnr.A5809

Article Figures & Data

Figures

Tables

Download figure
Open in new tab
Download powerpoint
Fig 1.
Diagnostic performance of different VBM models. A and B, The area under the curve for different smoothing kernels for gray matter concentration and volume. C, An example of an intersection plot for T1 VBM for the concordant rate and specificity against statistical cutoffs. D, An example of a receiver operating characteristic curve for T1 VBM at 12-mm smoothing for GMC. The Euclidean distance is calculated from the optimized T-threshold (3.7 in GMC analysis), where C_R, S_P = 38.5, 33.9 to C_R, S_P = 100, 100. False positive rate = 100-S_P.
Download figure
Open in new tab
Download powerpoint
Fig 2.
Sample case (patient 4). A and C, An overview in the native space T1, T2, and FLAIR images. B and D, Zoomed-in images focusing on a possible epileptogenic lesion. A subtle disruption of cortical morphology is visible in this figure, more prominent on FLAIR images. This finding is concordant with the clinical hypothesis supported by noninvasive and intracranial EEG, indicating seizure onset in the right frontal lobe. The possible lesion was detected as increased GM by all VBM models, but the extent and effect size were clearly better for the VBM models, including FLAIR (E).
Download figure
Open in new tab
Download powerpoint
Fig 3.
Sample case (patient 13). A, A VBM finding detected only by T1+FLAIR in the left temporoinsular region overlaid on the native space FLAIR image. B, Magnified images of T1, T2, and FLAIR suggesting a blurred gray-white matter junction (arrow). This finding is concordant with the clinical hypothesis of seizure onset.

Tables

Figures

View popup

Table 1:

GMC analysis for VBM models at 12 mm and a T-threshold of 3.7^{^a}

Model	Concordant Rate (%)	Specificity (%)	Discordant Rate (%)	Euclidean Distance (Range, 0–141.14)	Concordant Rate/Discordant Rate
T1	38.5	33.9	84.6	90.29	0.46
T1+T2	46.2	40.3	53.8	80.37	0.86
T1+FLAIR	46.2	37.1	61.5	82.77	0.75
T1+T2+FLAIR	46.2	35.5	76.9	84.00	0.60

↵a For VBM GMC automated analysis (prior to visual interpretation), values of concordant rate, specificity, discordant rate, Euclidean distance from C_R = S_P = 100, and concordant/discordant ratio are provided for all models: namely, T1, T1+T2, T1+FLAIR, and T1+T2+FLAIR. The smoothing level and T-threshold are 12 mm and 3.7, respectively.

View popup

Table 2:

GMV analysis for VBM models at 12 mm and a T-threshold of 3.0^{^a}

Model	Concordant Rate (%)	Specificity (%)	Discordant Rate (%)	Euclidean Distance (Range, 0–141.14)	Concordant Rate/Discordant Rate
T1	23.1	21.0	84.6	110.25	0.27
T1+T2	7.7	17.7	76.9	123.66	0.10
T1+FLAIR	38.5	9.7	84.6	109.25	0.46
T1+T2+FLAIR	38.5	14.5	76.9	105.32	0.50

↵a For VBM GMV automated analysis (prior to visual interpretation), values of concordant rate, specificity, discordant rate, Euclidean distance from C_R = S_P = 100, and concordant/discordant ratio are provided for all models: namely, T1, T1+T2, T1+FLAIR, and T1+T2+FLAIR. The smoothing level and T-threshold are 12 mm and 3.0, respectively.

View popup

Table 3:

Patients—visual analysis results for GMC^{^a}

Model	Potentially Epileptogenic (and Visible)	Potentially Epileptogenic (and Not Visible)	Potentially Epileptogenic (Combined)	Nonepileptogenic	Unclear	Artifacts
Concordant lobe (%)
T1	15.4	15.4	30.8	7.7	15.4	0
T1+T2	15.4	0	15.4	7.7	15.4	0
T1+FLAIR	15.4	30.8	46.2	7.7	23.1	0
T1+T2+FLAIR	15.4	15.4	30.8	7.7	23.1	0
Discordant lobe (%)
T1	7.7	38.5	46.2	7.7	30.8	15.4
T1+T2	7.7	23.1	30.8	0	23.1	0
T1+FLAIR	7.7	30.8	38.5	7.7	30.8	7.7
T1+T2+FLAIR	7.7	23.1	30.8	0	30.8	7.7

↵a Visual interpretation results of GMC analysis of patients are provided. The results contain percentages of patients scored by the reviewer as potentially epileptogenic and visible, potentially epileptogenic and not visible, potentially epileptogenic (number of patients with potentially epileptogenic and visible/not visible or both), nonepileptogenic, unclear, and artifacts. The results are reported for both concordant and discordant lobes.

View popup

Table 4:

Controls—visual analysis results for GMC^{^a}

Models	All Lobes (%)
Models	Nonepileptogenic	Unclear	Artifacts	Corrected Specificity (Excluding Artifacts and Nonepileptogenic)
T1	14.5	53.2	22.6	46.8
T1+T2	21.0	45.2	17.7	54.8
T1+FLAIR	16.1	50.0	19.4	50.0
T1+T2+FLAIR	16.1	48.4	25.8	51.6

↵a Visual interpretation results for GMC analysis for controls are presented. Results contain the percentage of controls scored by the reviewer as nonepileptogenic, unclear, and artifacts. Finally, corrected specificity is reported as the percentage of controls that did not have unclear findings—that is, all findings identified as artifacts/nonepileptogenic lesions and patients with no findings (VBM specificity prior to visual analysis).

View popup
Table 5:
Nonvisible findings—visual analysis results for GMC^{^a}
Models All Lobes (%)
Controls Patients
T1 53.2 61.5
T1+T2 45.2 30.8
T1+FLAIR 50.0 61.5
T1+T2+FLAIR 48.4 46.2
↵a Rates of findings not visible—that is, unclear findings in controls and unclear/potentially epileptogenic and not visible findings in patients across all models for all lobes (%) are reported.