Cluster Analysis of Glaucoma Patients Using the Retinal Nerve Fiber Layer Thickness of the Optic Nerve and DTI Parameters of the Optic Radiation

Glaucoma is the leading cause of irreversible blindness in industrialized countries. Open angle glaucoma (OAG) accounts for about 50% of glaucoma blindness [1]. The pathophysiological mechanisms are not completely elucidated. Both experimental [2,3] and human post-mortem studies [4] have shown that in addition to the damage of retinal ganglion cells also the post-retinal magnoand parvocellular layers of the lateral geniculate nucleus and the primary visual cortex were impaired in glaucoma. Some of the 4th neurons in the lateral geniculate nucleus have died and surviving neurons were atrophic. A concomitant damage of the 3rd (optic nerve and optic tract) and the 4th (optic radiations, OR) neuron of the visual pathway as a result of pathomechanisms underlying ischemia cannot be excluded. There are associations between typical cerebral lesions, which account for cerebral ischemia and microvascular risk factors (i.e. arterial hypertension) [5,6], cerebrovascular [7] and cardiovascular diseases [8]. Additionally, anatomy and physiology of retinal arterioles were found to be similar to cerebral arterioles.


Introduction
Glaucoma is the leading cause of irreversible blindness in industrialized countries. Open angle glaucoma (OAG) accounts for about 50% of glaucoma blindness [1]. The pathophysiological mechanisms are not completely elucidated. Both experimental [2,3] and human post-mortem studies [4] have shown that in addition to the damage of retinal ganglion cells also the post-retinal magno-and parvocellular layers of the lateral geniculate nucleus and the primary visual cortex were impaired in glaucoma. Some of the 4 th neurons in the lateral geniculate nucleus have died and surviving neurons were atrophic. A concomitant damage of the 3 rd (optic nerve and optic tract) and the 4 th (optic radiations, OR) neuron of the visual pathway as a result of pathomechanisms underlying ischemia cannot be excluded. There are associations between typical cerebral lesions, which account for cerebral ischemia and microvascular risk factors (i.e. arterial hypertension) [5,6], cerebrovascular [7] and cardiovascular diseases [8]. Additionally, anatomy and physiology of retinal arterioles were found to be similar to cerebral arterioles.
The new non-invasive technique of the diffusion tensor imaging (DTI) is suitable to reconstruct the visual pathway [9]. It is based on the magnetic resonance imaging (MRI) and allows for quantification of limited diffusion of water molecules along fiber tracts such as the visual pathway. DTI parameters give information about the anisotropy sum of the diffusion in all directions [10,13]. The axial diffusivity (AD) measures the water diffusivity parallel to the axonal fibers [15,16], and the radial diffusivity accounts for water diffusivities perpendicular to the axonal fibers [15,16].
The present study used the cluster analysis to find two different age-groups, who are different in RNFL thickness each independently from glaucoma type. The goal was to investigate the involvement of the optic radiation in glaucoma-induced damage of the optic nerve and to separate this effect from the ageing effect.

Patients
In the prospective comparative observational study 45 patients clinically diagnosed with glaucoma of different entities (age 60.3±13.0 years) and 17 subjects diagnosed with normal optic nerve head (age 50.3±14.3 years) were randomly selected from the institutional outpatient clinic. Demographic and history data were interrogated by a questionnaire.
Eyes were assessed by ophthalmologic examination including measurement of the retinal nerve fiber layer thickness (RNFL) by Spectralis Optical Coherence Tomography (Heidelberg Engineering, Heidelberg, Germany), automated white-white perimetry (Octo 900 dG2, Interzeag, Schlieren, Switzerland), and measurement of a 24h-profile of the intraocular pressure.
Study participants completed multimodal magnetic resonance imaging of the brain for detection and staging of cerebral microangiopathy and diffusion tensor imaging of the visual pathway for quantification of diffusion and anisotropy parameters within the optic radiation. The MRI readers were blinded to the diagnosis of the study subjects. The study population is characterized in Table 1. None of the subjects had myocardial infarction or carotid endarterectomy.
The study was conducted in accordance with the Declaration of Helsinki on Biomedical Research Involving Human Subjects. The Clinical Investigation Ethics Committee of the University of Erlangen-Nürnberg approved the study protocol, and written informed consent was obtained from all subjects prior to the study after explanation of the nature and possible consequences of the study.

Magnetic resonance imaging
Magnetic resonance imaging was performed on a 3T high-field scanner (Magnetom Tim Trio, Siemens, Erlangen, Germany) with a gradient field strength up to 45 mT/m (72 mT/m effective). The anatomical data were obtained in a T1-weighted 3D-MPRAGE sequence (TR = 900 ms, TE = 3 ms, FoV [field of view] = 23 x 23 cm, acquisition matrix size = 512 x 256 reconstructed to 512 x 512, reconstructed axial plans with 1.2 mm slice thickness). For detection of microangiopathy, a heavily T2-weighted fluid-attenuated inversion recovery (FLAIR) sequence covering the whole brain was acquired (TR = 10,000 ms, TE = 115 ms, matrix size = 512 x 512).

Diffusion tensor imaging
Diffusion tensor imaging was performed in the axial plane with 5 mm slice thickness using a single-shot, spin echo, echo planar imaging (EPI) diffusion tensor sequence covering the whole visual pathway (TR = 3,400 ms, TE = 93 ms, FoV = 23 x 23 cm, acquisition matrix size = 128 x 128 reconstructed to 256 x 256, number of signal averages = 7, partial Fourier acquisition = 60 %). Diffusion weighting with a maximal b-factor of 1,000 s/mm 2 was carried out along 15 icosahedral directions complemented by one scan with b = 0.

Reconstruction of fiber tracts
Datasets were automatically corrected for imaging distorsions and coregistered in reference to T1-weighted MPRAGE images. These and further calculations such as determining the independent elements of the diffusion tensor, deriving the corresponding eigenvalues and eigenvectors, and reconstructing and volume rendering fibers, were performed with a dedicated software package (Neuro 3D, Siemens, Erlangen, Germany).

Algorithm for semi-automated segmentation of the optic radiation
The previously developed algorithm of our group for automated segmentation of the optic radiation in DTI slices is described in detail elsewhere [17]. Briefly, the algorithm relies on the physiological properties of the optic radiation (anterior-posterior diffusion direction) to obtain an automated estimation of the optic radiation. The system uses a statistical surface evolution technique initialized by the estimated optic radiation to segment the optic radiation. Finally, the midbrain is roughly segmented and the segmented optic radiation is refined based on its relative position to the midbrain. The automated segmentation was accomplished in the slice, which soonest included the two lateral geniculate nuclei (LGN) and the largest area of the optic radiation. Then, the estimated segmented optic radiation was checked for concordance with anatomical knowledge and compared to a DTIbased atlases of white matter anatomy [18]. Segmented anatomical structures not belonging to the optic radiation were removed manually. Such structures were the distal part of the optic tract, the forceps major    of the corpus callosum, which proceeds medial of the optic radiations, and all tracts, which do not connect to the primary visual cortex ( Figure  1 A-C). The mean of the anisotropy and diffusivity parameters within the segmented areas of the optic radiation was calculated and used for further calculations.

Statistical analysis
Analyses were performed using the PASW software (release 18.0, SPSS Inc. Chicago, IL, USA). Hierarchical cluster analysis was performed to select four groups of patients, two middle-aged and two older groups. The two groups each had different RNFL thickness.
The significance of the calculations was confirmed by conventional statistical methods. Normal distribution of the raw data was verified by the Kolmogorov-Smirnov test. Identified groups were compared by means of non-parametric t-test for unpaired samples. For avoidance of a distortion caused by the differently sized measurement values of the DTI-parameters and the RNFL thickness, variables were normalized to z-scores (mean values normalized to 0 and standard deviation normalized to 1) before calculation of the slope. Linear regression analysis was applied to determine the effect of RNFL thickness on the DTI parameters. A p-value ≤ 0.05 was considered to be significant.

Results
The measurement results of the four clusters are shown in Table  2 and 3. Cluster 1 and 2 are characterized by middle age and equal DTI parameters of the optic radiation but different thickness of the retinal nerve fiber layer. Cluster 3 and 4 features an older age and also different RNFL thickness but equal DTI values. The RNFL thickness was not different between the middle-aged group with worse RNFL values and the older group with the better RNFL values within the older groups but the DTI parameters. Cluster 3 and 4 had differently advanced glaucoma as determined by the RNFL values. The slope of the radial diffusivity versus RNFL thickness was different between the two older groups with different RNFL thickness (Table 4). Linear regression analysis showed that in the older patients with the more advanced RNFL reduction (cluster 4) the effect of RNFL reduction was significant on the fractional anisotropy, mean diffusivity, and radial diffusivity of the optic radiations (Table 5).

Discussion
The present study has identified four groups by cluster analysis to investigate the impairment of the optic radiation at different age and with different thickness of the retinal nerve fiber layer. The goal was to identify the variables influencing the condition of the optic radiation.
Comparison of the mean values between groups detected the ageing effect but was not able to prove an additional glaucoma effect in our data. Only the application of regression analysis was able to demonstrate the relationship between clearly reduced RNFL thickness and an impairment of the optic radiation independently from age in advanced glaucoma and anisotropy. This means, that in middleaged glaucoma patients the diffusivity in the optic radiation appears to be unchanged despite a reduced RNFL thickness. An ascending degeneration from the third to the fourth neuron is not supported by our data in this group. The two older groups consisted predominantly (cluster 3) or exclusively (cluster 4), respectively, of glaucoma patients, but they had a different RNFL thickness, which determines the different progression of optic nerve degeneration. This finding points at an impairment of the axonal integrity/ demyelination of the optic radiation in older glaucoma patients with reduced RNFL thickness, which is independent from ageing effects. OD, right eye; OS, left eye; FA, fractional anisotropy; MD, mean diffusivity; AD, axial diffusivity, RD, radial diffusivity; RNFL, retinal nerve fiber layer thickness.  In general, the fractional anisotropy (FA) and mean diffusivity (MD) have been described to denote the axonal integrity of neuronal tracts [10,13,14], whereas the radial diffusivity (RD) increases with the degree of demyelination or glial cell morphology [15]. MD and FA are known to be sensitive to white matter pathologies, but with lacking specificity of the differentiation between different pathologies (e.g. acute stroke, multiple sclerosis, cerebral tumors). Thus, the explanatory power of these measures is limited [20][21][22].
The increase of the mean diffusivity in the optic radiation was also observed with ageing and is thought to agree with the pattern of a chronic ischemic process independent from other risk factors and diseases [23]. Ageing is associated with an increase of the radial diffusivity. There is close agreement between neuropathology and imaging measures suggesting that ageing underlies the process of myelin breakdown [24].
The impact of ischemic processes on the visual pathway has been examined by in vivo DTI in a mouse model of optic nerve degeneration resulting from transient retinal ischemia [25]. The death of retinal ganglion cells caused an early and prolonged decrease of diffusion anisotropy (FA) in the optic nerve. A pattern of gradual MD increase after a sharp early decrease of MD in the optic nerve was seen also in stroke [10]. During transient ischemia AD and RD have shown distinct patterns with an early and prolonged decrease of AD and a delayed and sustained elevation of RD [25][26][27]. Ostensible normalization of MD during follow-up was interpreted as a result of combined axonal injury and demyelination [10]. This pattern reflects the histological profile of axonal injury preceding myelin injury without significant destruction of overall axonal cytoskeletons [25][26][27]. However, these findings are attributed to acute impairment of the visual pathway within a few weeks. In chronic optic nerve atrophy another pattern of DTI parameter changes could be the result of damage through several years.
An association between defects of the retinal nerve fiber layer and signs of cerebral ischemia has been described particularly in older male subjects, who suffer from arterial hypertension [28]. Indeed, the older clusters in our study had a high prevalence of arterial hypertension and also the retinal nerve fiber layer thickness was reduced compared to the middle-aged clusters.
In conclusion, in middle-aged glaucoma patients degeneration of the optic radiation may not be detected by the methods applied in this study. In older glaucoma patients damage of the axonal integrity and demyelination of the optic radiation is suggested to be predominantly induced by ageing, which may cover the additional effect by advanced glaucomatous RNFL impairment.