Ablation of Apparent Diffusion Coefficient Hyperintensity Clusters and Mesial Temporal Lobe Epilepsy Improves Seizure Outcomes after Laser Interstitial Thermal Therapy

Introduction

Mesial temporal lobe epilepsy (mTLE) is one of the most prevalent types of seizures in the world. Up to a third of the mTLE patients become drug resistant and require surgical intervention to control their seizures. MRI-guided Laser Interstitial Thermal Therapy (LiTT) is a minimally invasive surgical procedure that is rapidly becoming an alternate first-line surgical option in drug-resistant mTLE.² LiTT allows for more selective targeting and has been associated with shorter hospital stay^{2, 3} and reduced postoperative neuropsychological complications compared to anterior temporal lobectomy, which remains the gold standard.^{4, 5}

While LiTT is promising, post-operative seizure outcome is highly variable among patients and centers. The inconsistent outcome may be partly due to variation in mesial temporal love anatomy as well as in the size and location of the epileptogenic foci that current practice does not account for. Past studies have proposed several predictors of seizure outcome after LiTT, including target structures,^{6, 7} ablation volumes,⁷ and surgical approaches.^{6, 8, 9} However, these factors remain controversial in the literature with conflicting results across centers. Since LiTT is designed to target a smaller, localized tissue with real-time MR guidance, it is imperative to identify pre-operative variables that can help surgeons identify precise location of target tissues that will lead to meaningful improvement in long-term seizure outcome.

One promising radiographic predictor of seizure outcome in mTLE is Apparent Diffusion Coefficient (ADC). ADC is an MRI sequence that is derived from Diffusion Weight Imaging (DWI), and it represents the mean diffusivity of water molecules in tissues. ADC is a marker of tissue degeneration, demyelination, ischemic cell death, and deterioration of tissue integrity in white and grey matter.^10-12. High intralesional ADC intensity has been observed in TLE patients,¹³ especially in pathological tissues representing as Mesial Temporal Sclerosis (MTS) ^14-16 and Focal Cortical Dysplasia (FCD)^{17, 18}. Thus, it is conceivable that selectively targeting tissues with abnormally elevated ADC signals in the mesial temporal lobe with LiTT may lead to improved seizure control by selectively targeting the most epileptogenic foci.

Past investigations that have demonstrated ADC’s ability to delineate pathological tissues in mTLE, mainly used voxel-wise evaluation. The conventional voxel-wise approach generally does not have the resolution to identify exact spatial positions of high ADC intensity clusters. Moreover, because ADC can be distributed as a gradient across multiple voxels, this approach cannot produce discrete ‘ablation targets’ with clear spatial boundary and morphometry that can be readily incorporated into existing clinical workflow. A method that can identify spatially distinctive targets within the mesial temporal lobe can help optimize trajectory planning and ablation. In this study, we hypothesized that ADC cluster analysis can identify foci of pathological tissues within the mesial temporal lobes of mTLE patients undergoing LiTT. and that the extent of high intensity ADC clusters is associated with seizure outcome.

Methods

Cluster Extraction

After co-registration of mTL volumes, a custom-developed unsupervised clustering pipeline was implemented to identify spatially distinct, hyperintensive clusters in preoperative ADC sequence. Because all subjects had epileptogenic zones localized to mTL areas prior to LiTT, and most ablation volumes were constrained to hippocampus and amygdala, this pipeline was performed exclusively on mTL structures ipsilateral to surgical side. The pipeline identified discrete regions with elevated ADC signals and demarcated them as individual clusters. The spatial boundaries of each cluster were made by iteratively comparing between intra-cluster ADC and extra-cluster ADC signal intensity. The final boundary of each cluster was determined when the intra-cluster tissues had relatively higher ADCs compared to surrounding tissues. The major advantage of this approach is that the ADC threshold to determine the boundary of clusters was dynamic and not predetermined arbitrarily. The pipeline does not require a prior training set or manual intervention because the pipeline produces cluster morphology based on local ADC distributions near each hyperintensive zones,.

The outputs of the pipeline are proposed cluster numbers, volumes of each cluster generated, cluster voxel coordinates, and individual cluster segmentations in native ADC MR space as shown in Figure 1A and 1B. Clusters of sample subjects with good (ILAE Class I) and poor outcome (ILAE Class II-VI) are visualized in Figure 1C and 1D.

• Download figure
• Open in new tab

Figure 1. Cluster Extraction and Visualization.

(A) Visualization of hippocampus (purple), (B) The clustering algorithm detects individual, spatially distinct clusters (green) composing the hippocampus. Ablation volume (red) is overlayed onto clusters (green) of sample subject with (C) good outcome (ILAE Class I) and (D) poor outcome (ILAE Class II – VI).

Results

5. Cluster Properties and Prediction

Cluster Properties

The association between cluster properties and seizure outcomes was characterized by (1) preoperative numbers of proposed clusters and (2) proportion of cluster ablated.

The mean number of preoperative hippocampal clusters across all subjects was 6.2 ± 0.8 (SEM). For subjects with good outcomes, the mean number of proposed hippocampal clusters was 8.3 ± 1.7, and 4.8 ± 0.7 for subjects with poor outcomes as shown in Figure 2A. The difference in preoperative hippocampal clusters was not significant (Mann-Whitney U=62, p = 0.067). For amygdala clusters, the mean number of clusters was 2.7 ± 0.3 across all subjects, 3.5 ± 0.7 for good outcome subjects, and 2.3 ± 0.4 for poor outcome subjects as shown in Figure 2B. The differences in number were not significant (Mann-Whitney U=67.5, p = 0.090).

• Download figure
• Open in new tab

Figure 2. Preoperative Cluster Numbers and Seizure Outcome.

The association of seizure outcome is between hippocampal (A) and amygdala (B) clusters.

For hippocampal ablation, the mean proportion of hippocampal cluster number ablated across all subjects was 12.6 ± 3.5 %. Subjects with the good outcome had higher proportion ablated than subjects with poor outcome (27.1 % vs, 4.2 %, Mann-Whitney U=40, p = 0.0007). For amygdala ablation, the mean proportion ablated across all subjects was 5.2 ± 2.7 %, and subjects with the good outcome also had a significantly higher proportion of amygdala ablated (13.2 % vs. 0 %, Mann-Whitney U=59.5, p = 0.016). Results of proportion of hippocampal and amygdala cluster numbers ablated are reported in Figure 3.

• Download figure
• Open in new tab

Figure 3. Proportion of Fully Ablated Clusters and Seizure Outcome.

Mean proportion of cluster numbers ablated for hippocampus (A) and amygdala (B).

Hippocampal Cluster Ablation and Seizure Freedom Prediction

The predictive power of hippocampal cluster ablation on seizure outcome was assessed. Using univariable logistic regression, the proportion of cluster ablated was fitted with binary seizure outcomes (ILAE Class I = 1, ILAE Class II – VI = 0). The predictive power of the cluster ablated was independently analyzed based on two different metrics: (1) by net cluster volume and (2) by cluster numbers ablated.

Based on univariate logistic regression between proportion of cluster ablated and seizure outcome, seizure outcomes was significantly associated with proportion of cluster ablated based on cluster volume (OR = 1.063, 95% CI [1.016 – 1.126], p = 0.017), and by cluster number (OR = 1.084, 95% CI [1.031 – 1.616], p = 0.0063). Based on the AUC of the ROC curve, the predictive capacity of both of cluster ablation metrics also proved to be significant. The AUC of ROC between cluster volume ablated and seizure outcome was 0.7679 (95% CI [0.58 – 0.96], p = 0.0160). AUC of ROC for cluster number ablated was 0.8086 (95% CI [0.6258 – 0.9914], p = 0.0055). ROC curves based on proportion of cluster ablated by volume and by number are shown in Figure 4A and 4B, respectively.

• Download figure
• Open in new tab

Figure 4. ROC Analysis for Seizure Freedom from Hippocampal Cluster Ablation.

ROC curves were generated from univariable logistic regression between seizure outcome and degree of (A) cluster volume and (B) cluster number ablated.

Discussion

LiTT is a highly promising surgical alternative to anterior temporal lobectomy. However, reported seizure outcomes after LiTT are highly variable despite efforts to standardize the procedure. One possible reason is that the current LiTT techniques do not adequately address patient variability; Hence, there is a need for identification and validation of novel patient specific factors that are predictive of seizure outcome. ADC is a radiographic marker of disease that is easy to acquire and incorporate into the clinical workflow, can enable surgeons to fine tune the laser trajectories and intra-operative post-ablation assessment to ultimately enhance seizure outcome.

Our study explores the utility of (1) identifying hyperintense ADC clusters within mesial temporal structures and (2) ablation of these clusters improves mTLE seizure outcome following MRg-LiTT. Employing a machine-learning based approach, spatially distinct, discrete clusters were identified in the hippocampus and amygdala. The degree of hippocampal clusters ablated, both by volume and number, was significantly associated with complete seizure freedom (ILAE Class I) and was able to reliably predict seizure outcome following LiTT. To the best of our knowledge, this is one of the first studies to evaluate the relationship between ADC clusters in the mesial temporal structures and seizure outcome in mTLE patients following LiTT.

In mTLE, pathology is generally localized in the mesial temporal structures, and primarily in the amygdala, hippocampus, subiculum, and parahippocampal complex, that are strongly associated with seizure generation and initiation.^26-31 We chose ADC as a surrogate marker of pathological tissues distributed across the mesial temporal structures. ADC measures have been closely connected to mTLE pathology, as well as in comorbid tissue abnormalities as MTS and FCD. Elevated ADCs have been reported in ipsilateral mesial temporal structures in the ictal side,^{32, 33} and tissues composed of MTS,³⁴ and FCD.³⁵ While it is unclear whether the tissue clusters consisting of distinct ADC hyperintensity represent the highly epileptogenic zones, mounting evidence suggests that formation of pathological tissues observed in MTS and FCD may be a consequence of chronic seizure generation.^36-39. These local pathological tissues may be detected based on radiographic evaluation, such as in MTS ^{40, 41} and FCD,^{42, 43} or based on abnormal electrophysiological signals detected with EEG or SEEG.⁴⁴

In a previous study, we demonstrated that mTLE patients with higher ADC values within the ablation zone had significantly higher chance of good seizure outcome.⁴⁵ In this study, our results further suggest that the clusters likely demarcate patient specific pathology that are epileptogenic and/or is part of the main seizure network. Also, the fact that we failed to find association between proportion of hippocampus or amygdala ablated and seizure outcome also suggests that currently used ablation volumetrics may be inadequate because it fails to account for the heterogeneity of epileptogenic zone volume and locations among patients.⁴⁶

According to our findings, inter-cluster ADC values of the amygdala and hippocampus were not associated with seizure outcomes. Furthermore, no difference in the proposed number and volumes of clusters was observed between the seizure outcomes. These collectively support the consistency of the clustering approach to producing distinctive clusters. Through this approach, the proposed numbers, size, and ADC intensities forming the clusters have been controlled for, and seizure outcome became dependent only on the proportion of clusters ablated.

For the case of volumetrics, the total amygdala and hippocampal cluster volumes proposed were approximately 54.5 mm³ and 98.8 mm³, respectively. In meta-analysis across 13 studies, the pooled mean total ablation volume was 5376 mm^{3 47} with a range of 2900 – 7110 mm ³.^{48, 49} Relatively smaller cluster volumes allow for more focused ablation targets, which may allow for a reduction in potential postoperative cognitive complications. Compared to traditional open resection procedures, it is reported that cognitive measures, such as object naming and recognition,⁵ are preserved after laser ablation. It is presumed that a more favorable outcome is expected due to more focal tissue damage without affecting other regions involved in cognitive processing,⁵⁰ including extra-mesial temporal regions. Additionally, a decrease in verbal memory and visual memory was observed with the extent of left and right temporal lobe resection, respectively,⁵¹ in temporal lobectomy. Through our approach, it is may thus conceivable to further preserve cognitive functions with more precise means for targeting.

From our results, the proportion of (1) cluster volumes and (2) cluster numbers ablated were significantly associated with seizure outcome. For the hippocampus, the increasing degree of both cluster volumes and numbers was correlated with complete seizure freedom. For the amygdala, only a reduction of cluster numbers was associated with seizure freedom. Acknowledging the topographic variability between the amygdala and hippocampus, aiming to ablate the maximum number of spatially distinct clusters detected may pose a substantial therapeutic effect. In the case of hippocampal clusters, we further established a strong predictive power of cluster ablation on seizure outcomes through ROC analysis on both number (AUC = 0.8086) and volume (AUC = 0.7679).

With an increasing understanding and growing works of literature on potential predictors of LiTT outcome, we here propose that significant seizure freedom may be expected when clusters of hyperintense ADC signals are targeted. This information may be crucial for presurgical planning, such that trajectory and laser power can be adjusted to produce ablation volume maximally composed of pre-operatively detected clusters. Further benefits may also extend to the postoperative reduction of neuropsychological complications due to relatively small, proposed ablation volumes.

Limitations

This study has several limitations. First, it is a retrospective study with a relatively small cohort size. Our 6-month postoperative seizure freedom rate was 36.7%, which was lower than that previously reported rates, which suggest difference in patient population or selection criteria for LiTT among centers.⁵² Prospective multicenter investigation with a larger population and longer follow-up is needed to validate our results.

Second, variability in acquisition parameters of the DWI / ADC map as all images were extracted from a retrospective clinical database. Because voxel intensities of ADC sequence are known to be sensitive to parameters such as b-value, TR, and TE, we acknowledge the potential confounding effects when analyzing ADC signals. We attempted to minimize these effects and extend the generalizability of our findings by undergoing a pre-processing step and normalizing the voxel intensities (Z-score) of each ADC map. Standardization of the voxel intensities for each sequence partly allowed us to compare them across the two clinical cohorts and use them to produce more reliable clusters. Furthermore, we further attempted to control for acquisition parameters by including them as beta coefficients in multivariate analysis when comparing ADC signals between cohorts.

Third, we performed ablation volume segmented based on intraoperative post-ablation sequences. Because past studies have suggested that ablation volumes tend to decrease following initial ablation,⁵³ it is possible that our approach may have overestimated the ablation volumes. However, we have had restricted access to delayed post-operative MRI because our institutional clinical protocol only captures intra-operative T1 MRI sequence. Furthermore, only a few subsets of subjects had undergone post-operative MRI because it is generally performed for those with post-operative recurrent seizures. Therefore, we used intraoperative sequence for this study because it was the only sequence available in every patient in our cohort. The use of intra-operative post-ablation sequence to segment ablation volumes is also consistent with methods employed in previous literature on large multicenter study.⁵² Future study should evaluate mTL volumetry with ablation volumes using delayed post-operative imaging.

Conclusion

Greater extent of ablation of ADC hyperintensity cluster in the hippocampus, and the amygdala to some degree, was significantly associated with complete seizure freedom in mTLE patients who underwent LiTT. Our results suggest that ADC may potentially identify patient-specific pathological and epileptogenic areas in the mesial temporal lobe that are high yield for ablation. ADC cluster-based approach can enable clinicians to optimize patient selection process, pre-operative planning, and intra-operative post-ablation assessment to further improve seizure outcomes.

Disclosure of Conflicts of Interest

W.S. A sits on Advisory Boards for Longeviti Neuro Solutions, and NeuroLogic. He is also a paid consultant for Globus Medical. Other authors report no disclosures