The splenial angle: a novel radiological index for idiopathic normal pressure hydrocephalus

Idiopathic normal pressure hydrocephalus (NPH) is characterized by the classic triad of gait apraxia, dementia, and urinary incontinence but confounding factors commonly exist [1, 2]. Neuroimaging provides supplementary information to the comprehensive clinical assessment of NPH [3‐5]. Ventriculomegaly in the absence of obstructive hydrocephalus or raised intracranial pressure is typical. Specifically, disproportionately enlarged subarachnoid space hydrocephalus, characterized by ventriculomegaly with sulcal crowding near the vertex and enlargement of the Sylvian fissures, is well described [4]. Evans’ index (EI), a quantitative radiological index included in the international and Japanese guidelines for the diagnosis and management of NPH [3‐5], is a well-accepted gross measure of ventriculomegaly. However, it is non-specific, performs poorly at distinguishing between differing causes of ventriculomegaly, and is an unreliable predictor for shunt responsiveness [6].

The callosal angle (CA) is more specific and useful in discriminating NPH from other neurodegenerative disorders with overlapping clinico-radiological presentations of ventriculomegaly from brain atrophy, cognitive impairment, and/or Parkinsonism gait disorder [4, 5]. A narrow CA in NPH is believed to result from mechanical stretching of the corpus callosum by enlarged lateral ventricles and compression by enlarged Sylvian fissures. A small CA is also associated with better shunt responses [4, 7, 8]. Yet, given its advantages, the CA measurement is limited by its notable variability in practice [7, 9].

Diffusion tensor imaging (DTI) fractional anisotropy (FA) maps are increasingly incorporated into clinical practice [10, 11] due to their rapid acquisition and availability on commercial DTI software packages. The unique anatomical information afforded by its color encoding for directional molecular water diffusion allows for consistent identification of major fiber tracts. In NPH, gross deformation of the red-encoded callosal commissural fibers is well depicted on color FA maps (Fig. 1). We propose the splenial angle (SA) as a novel alternative axial angular index of lateral ventricular dilatation for the clinical workup of NPH. We hypothesize that the SA could quantitate lateral ventricular distension in NPH as the CA does. We evaluated its (i) values, reproducibility, and ease of training; (ii) temporal changes in NPH patients after shunt surgery; and (iii) performance in differentiating NPH from Alzheimer’s disease (AD), Parkinson’s disease (PD), and healthy control (HC), compared to EI and CA.

Approval from the institutional ethics committee for each NPH, AD, and PD study was obtained. All participants gave informed consent.

The demographics of the 76 participants in the four groups and the mean and standard deviation values of their brain MRI measurements are shown in Table 1, with boxplots of the group measurements graphically depicted in Fig. 4. The cohorts were well matched for age and sex. While the NPH patients met the EI criterion of ≥ 0.3, the HC group had EI well within the normal cutoff of < 0.3, and the EI of the PD and AD patients straddled across 0.3. Similar to the CA, the SA range was narrower (mostly < 35°) and distinct from the non-NPH groups (mostly > 45°) in general (see Supplementary Materials, Table 1 EI/CA/SA ranges and CA-SA scatterplot figure).

The inter-rater reliability was excellent for all measurements and detailed for each measurement and group in Supplementary Table 2. The ICCs and 95% confidence intervals for EI were 0.97 (0.93, 0.99), CA 0.99 (0.99, 1.00), and SA 0.99 (0.98, 0.99). Intra-rater reliability was also excellent with ICCs > 0.90 for EI, CA, and SA measurements repeated for a subset of 38 subjects by the research assistant after a time-lapse of 1 month.

Both the EI and SA measurements were easy to train in. The research assistant readily acquired proficiency in effecting reproducible EI and SA measurements after a morning of training. In contrast, training to effect reproducible CA measurements extended over a month. Figures 1 and 2 demonstrate representative SA measurements for HC, PD, AD, and NPH groups. The red-encoded posterior commissural fibers at the body of the corpus callosum and forceps major were more severely deformed by the gross ventricular distension than the forceps minor and anterior callosal fibers in NPH patients. The SA (Figs. 1 and 2c) is well placed to document these morphological changes, and severely narrowed in NPH patients, in stark contrast to non-NPH subjects. Figure 3 shows the multiple potential pitfalls to positioning a “true” coronal plane and its impact on CA variability. Much effort was expended learning to reformat and position a “true” coronal MPRAGE image at the PC. Specifically, learning anatomical landmarks on the high-resolution 3D MPRAGE sequence (using the internal auditory canal and inner ear structures) to place the coronal image without right-left rotations protracted training. There was greater variability in the CA with erroneous acute/obtuse angulation (Fig. 3a) than with lateral right-left rotations of the coronal plane (Fig. 3b).

Pairwise comparisons between groups for EI, CA, and SA are summarized in the box plots in Fig. 4. As expected, only the EI (Fig. 4) differentiated HC from patients, given the established international NPH diagnostic guidelines. There were significant differences (p ≤ 0.001) between NPH and non-NPH groups. The SA best segregated NPH from non-NPH groups compared to CA and EI (Fig. 4).

The ROC analysis curves for the performance of EI, CA, and SA in differentiating NPH from each non-NPH group individually are detailed in Fig. 5, including cutoff, sensitivity, specificity, accuracy, and positive and negative predictive values. Using cutoff thresholds chosen by the closest to the top-left corner of the ROC plot, SA (AUC > 0.98) performed well in predicting NPH from HC, PD, and AD groups, compared to CA (AUC > 0.91). The cutoffs to identify and distinguish NPH from non-NPH groups of HC, PD, and AD were 43.9°, 33.7°, and 40.5° respectively, with narrower angles favoring NPH. The sensitivities and specificities for these cutoffs were 94.7% and 100%, 94.7% and 94.7%, and 94.7% and 100% respectively. SA was the only measure left in the final model of the multivariable analysis. The addition of more predictors (EI and CA) did not improve the ability to differentiate NPH from the other groups (Table 2).

Follow-up brain MRI at 1 year was available in 11 of the 19 NPH patients for further image analysis. Of these, six NPH patients had undergone ventriculoperitoneal shunting shortly after the baseline MRI scan, but five did not due to contraindications to surgery or refusal by patient or caregiver. Table 3 detailed the temporal changes in EI, CA, and SA measurements in these eleven NPH patients. At follow-up MRI, all measures were largely unchanged in non-shunted patients; EI was also stable in shunted patients. Both CA and SA showed similar mean angular widening by 13° at 1-year follow-up. However, the CA variability (± 14.3°) is twice that of SA (± 6.7°), and the angular change more significant for SA than CA measurement given that the baseline SA value was less than half that of CA.

In this study, we present the SA, a novel radiological index for the imaging workup of NPH measured on the ubiquitous and conventional axial plane that is standardized in most MRI centers, compared to the less routine coronal plane for CA measurement where the placement of the coronal plane is far more variable between clinical radiological practices and requiring more training to achieve uniformity across sites. The SA is directly determined on the axial DTI color FA maps and easily reproduced with minimal training and without the need for reformatting of the already predefined axial acquisition. The SA is measured using standard radiological tools readily available on the clinical reporting workstation.

DTI sequences are single-shot echo planar acquisitions that are rapidly acquired and very time efficient as an added sequence in the clinical protocol given its unique trajectory, sampling multiple Cartesian lines in k-space in a single readout [10, 11]. The color FA maps indicate the orientation of the major anisotropic white matter fiber tracts and allow easy discrimination of red-encoded commissural tracts traversing the corpus callosum from right to left and green-encoded anterior-posterior cingulate association fibers. With increasing clinical applications of DTI, it may be apt to integrate simple DTI sequences into the routine neuroradiological assessment of NPH. The SA is unique in its sensitivity to ventricular morphological changes and capacity to bypass traditional concerns of imaging markers in NPH, including variability of the CA, complex image post-processing requisites for 3D volumetry or validity of DTI parametric comparisons across sites and scanner makes [2, 4, 7, 9, 13, 15, 19‐22]. Our excellent ICC scores for the SA measurement after effortless training attest to its reproducibility.

The CA is variable in untrained hands in routine clinical practice [7, 9, 22]. The CA narrows as the choice of the coronal plane for its measurement moves posteriorly across the corpus callosum [4, 7, 9, 22]. Even when the coronal plane for CA measurement is standardized and fixed at the PC, we found reproducible CA challenging if due training and caution were not undertaken to correct for non-orthogonal or right-left tilting of the said coronal plane (Fig. 3) [9]. These challenges to CA measurements on coronal planes truly orthogonal to the bi-commissural plane contribute to its variable cutoffs reported in the literature. Learning to correct for the erroneous right-left rotation of the coronal plane was time-consuming, albeit the CA variability on these right-left mal-rotated planes were less than those arising from acute-obtuse mal-angulation of the coronal plane. Hence, 3D volumetric data is excellent for fine adjustments to achieve a “true” coronal plane, but this may be difficult to readily translate into busy clinical reporting.

SA outperformed the EI and CA in predicting NPH from each non-NPH group in our ROC analysis (Fig. 5) and predicted NPH from all non-NPH groups with high performance, without improvement after the addition of EI and CA (Table 2). Stark narrowing of the SA in NPH patients reflected severe medial compression and stretching of the posterior callosal fibers and forceps major against the unyielding inferior free edge of the interhemispheric falx secondary to the gross superior expansion and towering of the lateral ventriculomegaly. This is in sharp contrast to the normal fanning of the forceps major and posterior callosal fibers in non-NPH subjects (Figs. 1 and 2). Interestingly, Yamada et al [23] also found linear indices (derived from semi-automated segmentation of ventricular and subarachnoid spaces) defining compression of the brain just above the ventricle useful in differentiating NPH from AD. However, these sophisticated indices require complex image post-processing and analytical tools in 3D quantitative volumetric analysis [23, 24] to enable detailed characterization of ventriculomegaly and distribution of the CSF compartments. Despite their limitations, simple quantitative tools, such as the EI and CA, have been shown to be effective as diagnostic screening tools to differentiate between NPH and non-NPH subjects [4, 7, 22, 25, 26]. The relationship between SA and automated volumetry remains to be determined.

Radiographically, the posterior callosal fibers appear more at risk of deformation by the progressive lateral ventriculomegaly in NPH than the anterior callosal fibers. This may be secondary to blunt pressure from enlarged CSF volumes directed posteriorly against the medial walls of the lateral ventricles as the CSF gush out from the foramen of Monro during diastole. Follow-up MRI in our small NPH subset who underwent shunting showed that the SA was sensitive to temporal morphological changes in post-shunted NPH patients. Perceptible angular changes in both the CA and SA were registered following shunting, suggesting some relief of pressure on the posterior callosal fibers. MRI features in NPH are known to correct partially following shunt intervention [4, 27]. However, similar angular improvement (13°) (Table 3) given the narrower pre-shunting SA (25°) compared to the wider pre-shunting CA (58.3°) (Table 1) suggested that the vertically directed CA measurement might be less effective in capturing the angular change in posterior-medial ventricular wall compression. Of note, the computed angular change was also less variable for the SA than the CA on the follow-up MRI scans, likely reflecting the robustness of the SA measurement. The EI, which documents frontal ventricular pressures and distension, was inadequate in registering the ventricular morphological changes post-shunting, remaining unaffected between shunted and non-shunted NPH patients (Table 3). Meier et al also found the EI to be a poor indicator of clinical response to shunt placement in NPH [6]. The utility of SA as a predictor of shunt responsiveness or indirect index of shunt function warrants further study.

Our study had its limitations. First, the NPH group was compared to AD, PD, and HC datasets that were acquired from non-identical imaging protocols optimized for different studies. Nevertheless, the differences in in-plane resolution for the sequences (DTI/MPRAGE/FLAIR) where SA, CA, and EI were measured were generally < 1 mm². Second, the DTI acquisition times were longer than conventional clinical MR sequences for the purpose of rigorous DTI parametric quantitative analysis in the respective studies. However, the proposed SA is an angular measure that is independent of quantitative DTI parameters and protocol variations [13, 19‐21]. A simple DTI sequence with a minimum of six diffusion sensitization non-colinear directions and acquisition time of < 2 min would suffice for generation of useful color-encoded FA maps [10] for SA measurement. Third, we did not compare the efficiency of the SA measurement across other routine axial MRI sequences. In our experience, the body of the corpus callosum in NPH patients is notably compressed and narrowed and we found the 2-2.5-mm slice thickness in the DTI protocol ideal for selection of the axial cut containing the complete body of the corpus callosum for SA placement in NPH patients. Slice selection for SA measurement given partial volume averaging of the thinned-out corpus callosum on 4-5-mm thicker routine conventional T1- and T2-weighted axial imaging would require further assessment.

Future work needs to be done to assess the reproducibility of SA by different raters across sites, its inclusion in radiological scales summarizing imaging features into structured scores for diagnosis or prognosis [28, 29], clinical triaging for further sophisticated 3D volumetric analysis [23, 24] or CSF tap tests for diagnostic confirmation of NPH, and utility as a predictor of shunt responsiveness in a larger cohort of NPH patients undergoing surgical intervention [27]. In addition, while the bi-commissural plane is a common reference standard for axial imaging in many clinical radiological practices and indeed, even for defining the true coronal plane for CA measurements, future work could be done to assess the impact of variations to the central inter-commissural AC-PC line (e.g., the canthomeatal or corpus callosum lines [30] which may be used in other institutions) on the SA values.

The SA measured on axial DTI color FA maps is a simple radiological index reflecting a horizontal angular perspective of lateral ventricular dilatation in NPH. It shows promise in differentiating NPH from the ex vacuo ventriculomegaly of aging and neurodegenerative disorders and sensitivity to morphological changes before or after shunt surgery in NPH. Further work is needed to evaluate feasibility in translating its efficiency from DTI FA maps to other routine axial CT or MRI sequences.