Discussion

We have developed, for the first time, a semiautomatic method to isolate the LGN, which is capable of estimating the volume on the basis of high-resolution MR structural images. The robustness of this method was validated by 55 healthy controls ranging in age from 20 to 67 years. The main findings were that LGN volumes range from 52 to 102 mm³ (mean, 77 mm³) in the right and 66–105 mm³ (mean, 86 mm³) in the left hemisphere. These values agree reasonably well with the reported volumes in postmortem studies.^9,11 Thus, MR structural imaging is a valid and reliable method to evaluate LGN volume in vivo, when used in combination with our semiautomatic analysis method. In addition, we demonstrated LGN volume shrinkage in normal aging (see below).

The LGN is a major relay station of the retinofugal pathway, modulating the transmission of visual information that travels to the visual cortex. The characterization of the normal LGN and possible degenerative changes after nervous system damage are still a matter of debate, from a clinical as well as a methodologic point of view (Table 1). However, the precise determination of LGN in vivo is technically challenging due to its relatively small size and subcortical location.

In our study, LGN volumes were found to be comparable with volumes obtained by other postmortem studies. For example: Zvorykin¹¹ reported a 2- to 3-fold volume range (66–152 mm³) among 17 individuals. Putnam⁹ reported volumes of 77–115 mm³ but only studied 3 individuals. Andrews et al²⁸ studied 15 specimens and determined the volume in the right (91.1–154 mm³; mean, 121 mm³) and left hemispheres (91.9–157 mm³; mean, 115 mm³). They also reported a 2- to 3-fold variation in the size of LGN between subjects.

Rather than using postmortem tissue, we developed a more reliable method to measure the LGN, which enables direct stereotactic target determination. Measurements done in vivo would substantially improve accuracy and aid in observing pathologic changes. The results obtained in our study are consistent with the postmortem results,^9,11 showing comparable estimates of the average LGN volume and a high degree of variability (approximately 2-fold) with good test-retest reliability and consistency (Table 2).

Other studies have used proton density–weighted images¹² and fMRI to determine LGN volume in vivo.^8,16,28 The proton density images were found to be susceptible to changes in water content and generally provided good visualization of white matter but poor visualization of gray matter. However, differences in relaxation times and too low proton density among thalamic nuclei caused low contrast and quantitative uncertainties.¹³ The activation of neuronal populations in the LGN associated with visual stimuli is an alternative method to estimate LGN volume in fMRI studies, but fMRI experiments are much more involved, hence less practical, for routine clinical application.

There is considerable controversy about the precise relationship of the presumed LGN activated area and the actual volumes (Table 1). Due to the partial volume effects and the activation from adjacent structures such as the pulvinar,²⁰ BOLD-based activation of the LGN inevitably introduces bias to measurements. In addition, the activation scope of the LGN is close to noise level and most likely depends on the global normalization procedure.¹⁵ In extreme clinical cases, where permanent visual impairment or even blindness occurs, with fMRI methods, it is very difficult to acquire reliable data. Such issues have harmed the consensus of fMRI results and ultimately more widespread application in visual system research with human subjects.

Manual delineation of the LGN might produce more accurate results. However, this would be extremely time-consuming, and it is potentially prone to subjective bias, especially when the native image itself is of low-quality resolution or contrast. Our semiautomatic method improves the precision and speed of a prior region-of-interest-based method of Korsholm et al²¹ by using the FreeSurfer subcortical pipeline.^29,30 This pipeline is fully automated and user-independent because it uses a neighborhood function to encode spatial information, a forward model of the MR scanner parameters to improve sequence independence, and a nonlinear function to account for morphologic differences between the atlas and the individual brain. With the help of the obtained VDC mask in FreeSurfer, it is possible to easily exclude non-LGN voxels, which originate mainly from the hippocampus, white matter, or CSF.

Kastner et al²⁰ approximated the LGN as a cube with a mean side length at 4.89 ± 0.27 mm, which is consistent with the results by Chen et al.³¹ In addition, this closely matches the observations of Gupta et al,⁸ who estimated the side length at 4.74 ± 0.54 and 4.83 ± 0.95 mm for the right and left LGN, respectively. Therefore, a sphere VOI with a 3-mm radius was used as a mask to extract the LGN after the region-growing algorithm in this study. Moreover, we also noted that the posterior commissure usually appeared in the same section together with the LGN. By following the optic tract, it was also possible to trace of the LGN in the coronal plane.

With this new LGN imaging method, our study investigated the effects of aging on LGN volumes to test the value of our procedure in a medical context. Because each LGN represents only one-half of the visual fields, we summed up and compared the bilateral LGN volumes, accounting for visual input from each eye. The bilateral LGN volumes in our samples decreased with age (r = −0.512, P = .000) after regressing out ICV. This is an important observation because it might indicate that the loss of vision with progressing age may also have a brain-based component. However, further study of age-related changes is needed to demonstrate that our method is sufficiently sensitive to monitor longitudinal changes such as degeneration in the LGN. In fact, this semiautomatic method may lay the basis for further studies of the diagnosis and treatment of neurologic and ophthalmologic diseases and may also very well be useful for neurosurgical intervention.

With regard to the clinical observation of LGN volume changes with age, our general finding is consistent with those of animal studies, which have already reported LGN degeneration in normal aging.^32⇓–34 Diaz et al³² compared the volume change of the dorsal LGN as a function of age and found significant decreases of neuronal volume attenuation and numeric attenuation of neurons in old rats. Similarly, de la Roza et al³³ and Vidal et al³⁴ reported a significant reduction in the diameters of both the soma and nuclei of neurons of the dorsal LGN in aged rats. As suggested in their study, aging and partial loss of visual input appeared to induce small changes in the morphology of most of the dorsal LGN neurons. However, to the best of our knowledge, LGN volume changes associated with aging have never been studied in the human brain, and it now needs to be determined whether LGN volume loss is strictly a result of reduced afferent input from the retina or whether it is caused by LGN cell loss.

No sex difference in LGN volumes in the samples was observed, but the volume of the right LGN was, on average, larger than that of the left. Nevertheless, the difference is smaller than the variance introduced by repeated measurements. In addition, given that most subjects have 1 dominant eye, it would be of interest to determine if the asymmetry of the LGN volume is related to unilateral dominance. Additional studies are required to evaluate the relationship of LGN volume and eye dominance, including larger samples and a correlation with functional measures of vision.

In the field of neuroimaging, it is still a technical challenge to precisely visualize and distinguish the irregular and variable laminar patterns of the human LGN because this would require higher resolution. Therefore, it is imperative to apply our method to data obtained from high-resolution (7T) MR imaging, in which anatomic landmarks are more distinct and can be more easily identified.