MRI Texture Analysis Reveals Brain Abnormalities in Medically Refractory Trigeminal Neuralgia

Abstract

Background: Several neuroimaging studies report structural alterations of the trigeminal nerve in trigeminal neuralgia (TN). Less attention has been paid to structural brain changes occurring in TN, even though such changes can influence the development and response to treatment of other headache and chronic pain conditions. The purpose of this study was to apply a novel neuroimaging technique-texture analysis-to identify structural brain differences between classical TN patients and healthy subjects. Methods: We prospectively recruited 14 medically refractory classical TN patients and 20 healthy subjects. 3-Tesla T1-weighted brain MRI scans were acquired in all participants. Three texture features (autocorrelation, contrast, energy) were calculated within four a priori brain regions of interest (anterior cingulate, insula, thalamus, brainstem). Voxel-wise analysis was used to identify clusters of texture difference between TN patients and healthy subjects within regions of interest (p < 0.001, cluster size >20 voxels). Median raw texture values within clusters were also compared between groups, and further used to differentiate TN patients from healthy subjects (receiver-operator characteristic curve analysis). Median raw texture values were correlated with pain severity (visual analog scale, 1-100) and illness duration. Results: Several clusters of texture difference were observed between TN patients and healthy subjects: right-sided TN patients showed reduced autocorrelation in the left brainstem, increased contrast in the left brainstem and right anterior insula, and reduced energy in right and left anterior cingulate, right midbrain, and left brainstem. Within-cluster median raw texture values also differed between TN patients and healthy subjects: TN patients could be segregated from healthy subjects using brainstem autocorrelation (p = 0.0040, AUC = 0.84, sensitivity = 89%, specificity = 70%), anterior insula contrast (p = 0.0002, AUC = 0.92, sensitivity = 78%, specificity = 100%), and anterior cingulate energy (p = 0.0004, AUC = 0.92, sensitivity = 78%, specificity = 100%). Additionally, anterior insula contrast and duration of TN were inversely correlated (p = 0.030, Spearman r = -0.73). Conclusions: Texture analysis reveals distinct brain abnormalities in TN, which relate to clinical features such as duration of illness. These findings further implicate structural brain changes in the development and maintenance of TN.

Keywords: anterior cingulate; chronic pain; insula; magnetic resonance imaging; neuroimaging; texture analysis; thalamus; trigeminal neuralgia.

Figure 1

Statistical parametric map showing voxel-wise differences in texture features between healthy controls (n = 20) and right-sided TN patients (n = 9). Autocorrelation was greater in healthy controls than TN patients in a left brainstem cluster (A). Contrast was greater in TN patients than healthy controls in three clusters: lower brainstem/cervical spinal cord (B), left brainstem at the level of the trigeminal nerve (C) and right anterior insula (D). Energy is greater in healthy controls than TN patients in four clusters located in the left brainstem (E), right mid-brain (F), and left (G) and right anterior cingulate (H).

Figure 2

Median raw texture values within region-of-difference clusters between healthy controls (n = 20) and right-sided TN patients (n = 9). Autocorrelation was greater in healthy controls than TN patients in a left brainstem cluster (A). Contrast is greater in TN patients than healthy controls in three clusters located in the: right anterior insula (B); left brainstem at the level of the trigeminal nerve (C); and lower left brainstem/spinal cord (D). Energy is greater in healthy controls than TN patients in four clusters located in the: left brainstem (E); left anterior cingulate (F); right midbrain (G); and right anterior cingulate (H). Numbers within bars represent median, error bars represent interquartile range.

Figure 3

Receiver-operator characteristic (ROC) curve analysis of texture features. The single ROC curve with the largest area under the curve (AUC) is displayed for each texture feature. (A) ROC curve for left brainstem autocorrelation has AUC 0.839 (p = 0.0040). The optimal operating threshold autocorrelation value of 71.99 is 89% sensitive and 70% specific for subject group. (B) Left brainstem autocorrelation values for each individual subject are displayed with optimal operating threshold autocorrelation value overlaid. (C) ROC curve for right anterior insula contrast has AUC 0.922 (p = 0.0003). The optimal operating threshold contrast value of 12.53 is 78% sensitive and 100% specific for subject group. (D) Right anterior insula contrast values for each individual subject are displayed with optimal operating threshold contrast value overlaid. (E) ROC curve for left ACC energy and subject group has AUC 0.917 (p = 0.0004). The optimal operating threshold energy value of 41.62 is 78% sensitive and 100% specific for subject group. (F) Left ACC energy values for each individual subject are displayed with optimal operating threshold energy value overlaid. TN, right trigeminal neuralgia patients; HC, healthy control subjects.

Figure 4

Spearman correlations of anterior insula cluster contrast with duration of TN (years) and pre-operative pain score (VAS). (A) There is a negative correlation (p = 0.030, r = −0.73) between contrast in the right anterior insula and duration of TN, (B) but no correlation with pre-operative VAS score.