White Matter Changes with Normal Aging

Charles R.G. Guttmann1, MD, Ferenc A. Jolesz1, MD, Ron Kikinis1, MD, Ron J. Killiany3, PhD, Mark B. Moss3, PhD, Tamas Sandor1, PhD, Marilyn S. Albert2, PhD

1 Department of Radiology, Brigham & Women's Hospital
2 Departments of Psychiatry and Neurology, Massachusetts General Hospital, Harvard Medical School
3 Departments of Anatomy and Neurobiology, and Neurology, Boston University School of Medicine

Corresponding author:
Dr. Marilyn S. Albert
Massachusetts General Hospital
Psychiatry/Gerontology (149-9124)
Bldg 149, 13th Street
Charlestown, MA 02129
Phone: 617/726-5571 FAX: 617/726-5760

Keywords: Aging - Brain, White Matter - Brain, Volumes - Magnetic Resonance Imaging - Image Processing

Materials and Methods
Figure Legends


We evaluated brain tissue compartments in seventy-two healthy volunteers between the ages of 18 and 81 years with quantitative MRI. The intracranial fraction of white matter was significantly lower in the age categories above 59 years. The CSF fraction increased significantly with age, consistent with previous reports. The intracranial percentage of gray matter decreased somewhat with age, but there was no significant difference between the youngest subjects and the subjects above 59. A covariance adjustment for the volume of hyperintensities did not alter the foregoing results. The intracranial percentage of white matter volume was strongly correlated with the percentage volume of CSF. The finding of a highly significant decrease with age in white matter, in the absence of a substantial decrease in gray matter, is consistent with recent neuropathological reports in humans and non-human primates.


There is convincing data from magnetic resonance imaging (MRI) that overall brain volume decreases with advancing age, as reflected in a relative increase in intracranial fluid compartments on MRI images [1-10]. Several MRI-based studies of age-related volumetric changes in the gray and white matter of healthy volunteers have been reported [1-8]. Some of these studies detected changes in gray matter structures but did not observe or report significant changes in the white matter [3-6]. Other studies demonstrated white matter changes but were not able to demonstrate a link between these changes and overall loss of brain parenchyma (atrophy) [1,2,7,8].

Increasing histopathologic evidence suggests that cellular changes in the cortex are far less extensive than previously assumed [11]. Localizing brain changes to specific tissues or structures is important for understanding the underlying pathogenesis of damage and the pathophysiologic consequences. MRI provides localized measurements, however these measurements are difficult to quantify reliably, because of high variability of the MR scanners' variation of the gain field over the volume of interest due to radiofrequency inhomogeneity and their instability over time, as well as the limitations of image processing strategies in accounting for these confounding factors. To address these issues, we have developed a fully automatic algorithm to segment the contents of the intracranial cavity (ICC) into tissue classes, which is a valid and reproducible method for quantifying volumes of white matter, gray matter, CSF, and hyperintensities [12-13].

We have applied this method to 72 healthy individuals to characterize the distribution of tissue classes as a function of age, based on high resolution multislice images that covered the entire brain without gaps. We demonstrate significant loss of white matter with normal aging and show a strong correlation between the volume of white matter and the volume of CSF, a measure of overall brain atrophy.

Materials and Methods


Seventy two subjects participated in the study as part of a large study of normal aging [9]. There were 22 men and 50 women. The number of subjects per age group was as follows: 18-39 = 10 (4 women, 6 men), 40-49 = 9 (7 women, 2 men), 50- 59 = 8 (4 women, 4 men), 60-69 = 23 (16 women, 7 men), >69 = 22 (19 women, 3 men).
Subjects with a history of alcoholism, psychiatric illness, epilepsy, chronic lung disease, hypertension, heart disease, kidney disease, cancer or severe head trauma were excluded from the study. In addition, the subjects received a series of laboratory tests to rule out serious systemic illness. They had also been administered neuropsychological testing to eliminate the inclusion of persons with cognitive symptoms of a dementing illness in the early stages (e.g., the Wechler Memory Scale, the Wechsler Adult Intelligence Scale, etc). All subjects provided informed consent consistent with the institutional IRB regulations.

MRI Assessment

MRI examinations of the brain were performed on a 1.5 Tesla scanner (Signa, General Electric Medical Systems, Milwaukee, WI). Proton density and T2-weighted axial images were acquired using two interleaved dual-echo (echo times (TE) = 30 and 80 msec), long repetition time (TR = 3,000 msec) sequences. Contiguous 3 mm thick slices covered the whole brain from the foramen magnum to the higher convexity, with an in plane voxel size of 0.94mm (24 cm field of view with a 256 X 192 acquisition matrix). Variable receiver bandwidth was used for first (10.67kHz) and second (4.27kHz) echo. Standard flow compensation was achieved using first-order gradient moment nulling. Spins were saturated in an 8cm-thick slab inferior to the imaging volume in slice-selective direction. Scan duration was kept at 11 minutes and 36 seconds by using the 1/2 Fourier technique.

Computerized Analysis of MRI Images

The proton density and T2-weighted images were analyzed using semi-automated and fully automatic computerized procedures, with the operators 'blinded' to the age and gender of the subjects. All scans were reviewed by study staff and those with pereceptible motion artifact were eliminated from the analysis.

After identification of the intracranial cavity (ICC), with minimal operator interaction [12-13], the contents of the skull were segmented using a fully automated approach [14]. This computerized procedure has been demonstrated to be valid [14-15] and highly reliable [16]. The overall analysis employs a mixture of a 2D and 3D analysis. The core algorithm (known as the EM segmenter) analyzes each slice separately. Other aspects of the analysis utilize a 3D analysis, as follows:

The first step in the procedure is to identify the intracranial cavity using a semi-automated program [12-13]. During this step in the analysis of the MRI images, an operator makes a number of standardized decisions, based on anatomy, in order to enable the program to outline the intracranial cavity with accuracy (e.g., ensuring that the optic nerve, that connects the ICC with the orbits, is completely transected). This outline (or 'mask') is then used to determine the total volume of the intracranial cavity.

In the second step of the procedure, an automatic computerized program (the EM segmenter), classifies the region within the intracranial cavity (as defined by the 'mask') into four tissue categories, on a pixel-by-pixel basis: white matter, gray matter, cerebrospinal fluid (CSF), and hyperintensities (lesions). This intracranial cavity tissue segmentation program is a fully automated multi-step procedure and is therefore not biased by operator intervention [14]. The classification procedure, which utilizes an iterative non-parametric statistical method for optimizing the final classification for each pixel, is described in detail elsewhere [14], and can be briefly outlined as follows: The automated program begins by classifying pixels according to empirically derived cut-off points for each tissue class, utilizing information about the signal intensity of each pixel. Then a map is made of the differences between the signal for each pixel and the mean value of the tissue class to which it belongs. This difference map reflects the 'gain' of the image (i.e., the spatial variation in signal measurement), and permits the estimation of the gain field in the image. Lastly, this information is iteratively used to generate additional difference maps that are increasingly closer to an accurate reflection of the tissue classes to which each pixel should be assigned.

In the third step, partially volumed pixels (i.e., pixels that are partially brain and partially CSF, or partially bone and partially CSF), are identified by utilizing information about head anatomy. This step is necessary because partial voluming primarily presents a problem in tissue classification at the boundary between the brain (or bone) and CSF; the algorithm, described in step two (above) tends to classify such partially volumed pixels as hyperintensities. Thus, in the third step, an algorithm searches the boundaries between brain (or bone) and CSF and uses a set of rules (which differ depending upon whether the brain-CSF boundary is at the surface of the brain or near the ventricles) to optimize the identification of tissue and CSF [14]. Figure 1 provides two examples of MRI images and the segmentation results, after step 2 and step 3 of the segmentation procedure.

The pixels within each tissue category identified by the above procedures are then summed and expressed in milliliters by multiplying the total number of pixels by a constant 0.00264. The value of 0.00264 was derived from the following formula [(field of view/matrix size)2 x slice thickness]. These volumes are then normalized using the total volume of the intracranial cavity and expressed in percent of that volume: tissue volume fraction [%] = 100 x number of pixels x 0.00264/intracranial cavity volume.

The validity of the automated segmentation procedure has been assessed by comparing the automated analysis with manual analyses by five experienced operators [14]. The mean values from the 5 experienced operators did not differ significantly from the automated algorithm. The results of this automated method were also demonstrated to be significantly better than earlier segmentation programs utilized in our laboratory [12].

The reproducibility of the automated procedure has been assessed by examining 20 patients twice within 30 minutes, using the MRI protocol described above [16-17]. Between each of the scanning sessions, the patients exited the scanner and were repositioned in the machine by different radiological technicians. The image analysis was then performed by two independent operators ('blinded' to the status of the subjects) on two occasions, separated by two months. To statistically examine reproducibility of the tissue class estimates reported here, an intercorrelation matrix (Pearson's r) was created that examined the relationship between the measurements for each of the scans at two points in time, for the two operators. For the measurements of gray matter, the correlations varied between 0.93 and 0.99. Student's t test indicated that there were no significant differences between the measures of total gray matter between operators or across times of acquisition and measurement (p values varied from 0.14 to 0.96). The intercorrelation of the white matter measures varied from 0.995 to 0.999; none of the measures of total white matter were significantly different from one another by t-test (p values varied from 0.20 to 0.74). For the hyperintensities, the intercorrelations varied from 0.98 to 0.99, which also were not significantly different from one another by t-test (p values varied from 0.16 to 0.82). The intercorrelation of the CSF measures varied from 0.98 to 0.99. The mean difference between the CSF measures for the two MRI scans by the same operator was 0.01% of the total amount of CSF, which approached statistical significance (p=0.06). Comparisons of the CSF measures for the two operators or for the two time periods was not statistically significant (p = 0.36 and 0.67, respectively).

Statistical Analysis

The total volumes (expressed as percent of ICC) for gray matter, white matter, CSF and hyperintensities, derived from the segmentation procedure in the present study, were analyzed by Pearson Product Moment Correlation, analysis of variance (ANOVA), and analysis of covariance (ANCOVA).


Analysis of variance was used to assess group differences in brain tissue volume measures. These analyses demonstrated that the percent CSF increased significantly with age (F (4,70) = 12.42, p < 0.0001); the mean percent CSF increased from 7.1%, among the subjects 18-39, to 13.4% in the group of subjects above 69 years of age (Table 1). Conversely, the mean percent for the volume of white matter significantly decreased with age (F (4,70) = 9.41; p < 0.0001), from 38.9% to 33.0% for the group 18-39 and the group above 69 years of age, respectively. Figure 2 delineates the difference in the quantified volumes of white matter graphically, by showing the number of subjects that fall above and below a line that represents two standard deviations below the mean of the youngest subjects. There was no sigificant difference between the genders, after adjusting for age.

The percentage of gray matter was approximately 48% among the oldest and youngest age groups, but showed a decrease during the intervening decades. A statistically significant decrease was seen between the 18-39 group and the group 50-59 years old, (F (4,70) = 3.46; p < 0.013).

Post-hoc planned comparisons indicated that the oldest age groups were significantly different from the younger ones in relation to CSF and white matter volume percentages. That is, the 60 and 70 year olds had significantly greater CSF volumes than that of the group younger than forty. In addition, the 60 and 70 year olds had a significantly smaller white matter volume than the group below forty. For gray matter, there was a trend for a decrease between the youngest subjects and the group between 50 and 59 years. However, only the difference between the group 18-39 and 50-59 reached statistical significance at the 5% level. Figure 3 provides examples of the differences observed.

The magnitude of the age-related differences were reflected in the correlations with age. The percentage of CSF (r=0.68) and the percentage of white matter (r=-0.57) were significantly correlated with age (p < 0.0001). There was also a strong and highly significant correlation between the percent volume of the white matter and that of the CSF (r=-0.77, p < 0.0001). Neither the percentage of gray matter, nor the percentage of hyperintensities ("lesions") were significantly correlated with age (r=-0.15, p=0.22 and r=0.13, p=0.29, respectively).

The volume of the regions classified as hyperintensities (i.e., "lesions") accounted for 0.25% of the total ICC volume among the subjects 18-39, and 0.29% in the group above 69 years. There was a trend for the group above 39 years of age to have more hyperintensities than the group between 18 and 39, but the difference was not statistically significant at the 0.05 level. We, nevertheless, examined the potential contribution of hyperintensities to the observed decrease in white matter with age. An ANCOVA was performed in which the age-related difference in white matter volume was assessed, covarying the volume of hyperintensities. This procedure did not alter the conclusions based on the ANOVA. The percent white matter still demonstrated a highly significant decrease with age (F (5,69) = 9.02, p < 0.0001). The ANCOVAs pertaining to CSF and gray matter volume percentages, likewise, produced results that were almost identical to the ANOVAs.


Recent neuropathologic studies have challenged the classical view that normal aging is characterized by substantial neuronal loss in the cortex with age. Neuropathologic studies in humans [18-21] indicate that, with advancing age, neuronal loss in the cortex is either not significant or not as extensive as most earlier reports had suggested [22-27]. Rather, the findings of the present report, together with those from studies in aged non-human primates, suggest that alterations in white matter and subcortical neuronal loss may represent the predominant neuroanatomic changes in normal human aging.

These recent investigations differ from earlier studies (e.g., [22]) in both methodology and in subject selection criteria. Technologic advances have automated cell counting procedures, standardized the selection of tissue blocks for cell counts and minimized age-dependent differences in tissue shrinkage. In addition, the neuropathologic studies in humans have omitted any subjects with neurologic disorders, such as Alzheimer's disease, that can markedly alter brain structure and function.

Our results indicate that there is a highly significant age-related decrease in the percentage of white matter and a concomitant increase in the percentage of CSF on MRI. The percentage of gray matter showed only a relatively small decline with age; the only significant difference was found between subjects 18 and 39 and those in their fifties. The volume of hyperintensities changed only slightly with age and did not contribute significantly to the decline in white matter with age.

The presence of a highly significant decrease with age in total white matter, strongly correlated to the overall degree of brain atrophy, has not, to our knowledge, been previously reported in humans in vivo. This decrease, however, is consistent with recent neuropathological reports in humans, cited above, that show a minimal loss of cortical neurons with age [18-20]. There are comparable neuropathological data in monkeys. As in the studies in humans, cited above, studies in non-human primates have taken great care to select subjects without evidence of age-related illnesses that can alter brain function (e.g., kidney disease, lung disease, etc). Minimal neuronal cortical loss with age in monkeys has now been demonstrated in the striate cortex [28], motor cortex [29], entorhinal cortex [30], frontal cortex [31] and in selected parts of the hippocampal formation [32-33]. In addition, a recent study in monkeys based on MRIs, like our findings in humans, reports significant decreases in white matter volume with age, but minimal decreases in gray matter volume [34].

It should be noted that the measures utilized in the present study are consistent with focal changes in small areas of gray matter, such as selected areas of cortex or small subcortical nuclei. There are, in fact, consistent reports of substantial age-related neuronal loss in selected subcortical regions involved in neurotransmitter systems important for cognitive function, such as the basal forebrain [34], the dorsal raphe [35-36], or the substantia nigra [37-38]. For example, there is approximately a 50% neuronal loss with age in the basal forebrain and 35-40% loss in the dorsal raphe. This compares with an approximate loss of 5% in CA3 of the hippocampus. (See [39-40] for reviews). It should, however, be noted that the survival of neurons does not guarantee that they are functionally intact. Studies concerning functions, such as membrane properties, neurotransmitter synthesis, and neurotransmitter receptor composition should clarify this issue (e.g., [41]).

Although the present data cannot determine the precise nature of the alterations in white matter that account for the age-related decline in volume of white matter on MRI, recent reports based on both humans and monkeys, indicate changes in white matter with age that are consistent with the data reported here. Morphologic comparisons between white matter in old and young monkeys reveal an age-related decrease in myelin staining, indicating a loss of myelin with age [39], an increase in membrane-bound vacuoles in the dendrites and the neuropil, and increases in lipofuscin and inclusion bodies in microglial cells and astrocytes [28-29, 31]. It should be noted that a gradual loss of myelin with advancing age is consistent with the present findings, since a gradual and diffuse change in myelin is unlikely to produce classical hyperintensities, such as those seen in an acute process like multiple sclerosis.

The age-related increase in CSF found here is consistent with previous reports based on CT scans (e.g., [42]) and MRIs [1,2,9]. It is now generally accepted that CSF increases with age, even in the absence of disease. The subjects in the present investigation were specifically selected to be 'optimally healthy'; this aspect of the population is most likely the explanation for the limited findings with respect to hyperintense lesions, which likely correspond to the presence of serious systemic disorders, such as heart disease and lung disease.

The present MRI findings also differ from those of the only other comparable MRI studies in humans in which age-related differences in the volume of gray matter and white matter were examined [1-8]. Some of these studies detected changes in gray matter structures but did not report significant changes in the white matter [3-6]. Other studies demonstrated white matter changes but were not able to demonstrate a link between these changes and overall loss of brain parenchyma (atrophy) [1,2,7,8]. Most studies imaged only parts of the brain and/or used lower spatial resolution (e.g., larger slice thickness) than we did in this study. For example, most of the MRI studies cited above used an image analysis procedure which divided each cortical slice into an inner 55% and an outer 45% portion; the inner portion was not included in the analysis. The subcortical compartments of each slice were only analyzed for CSF and were not further separated into gray and white matter, since the authors felt that their automated method was prone to misrepresenting partially volumed white matter as gray matter, especially near the margins of the ventricles where hyperintensities were likely to appear. It therefore appears likely that analyses based on these procedures inadvertently omited regions of subcortical white matter that undergo the most dramatic change with age.

The results presented above must be considered in light of the limitations of the automated analysis applied to the MRI images. When white matter abnormalities are "mild" (i.e., when the white matter hyperintensities do not appear brighter than the normal gray matter in the images) it is possible for hyperintensities to be classified as gray matter. Thus, misclassification of abnormal white matter into the gray matter tissue class may partially mask a decrease in gray matter volume. However, we attempted to examine the potential impact of tissue misclassification by repeating all analyses with an adjustment for the volume of hyperintensities. This did not alter the findings of a slight decrease in total gray matter and a substantial decrease in the volume of white matter with age.

It should be noted that, to our knowledge, no image processing technique has yet been able to overcome the above limitations in a satisfactory way. Recent progress in automatic white matter segmentation techniques, using explicit digital models of brain anatomy, holds the promise to further improve the discrimination between gray matter, white matter and hyperintensities, and should allow the further analysis of these issues [43-44].

Taken together, these data suggest that the "classic" concept of significant age-related cortical neuronal loss must be revised and that white matter changes with age deserve comparable scrutiny to that granted gray matter. These findings also suggest that aging is associated with changes in the brain that can differ substantially from those seen in primary age-related brain diseases, such as Alzheimer's disease. Investigations related to this issue should have implications for the treatment of age- related change in cognition.

Figure Legends

Figure 1: Examples of the Segmentation of MR Images into Tissue Classes: The first two images provide examples of a proton density weighted (first column) and T2-weighted (second column) head MRI image, one from a young subject (top row - 37 year old female) and one from an elderly subject (bottom row - 75 year old female). The images in the next two columns are examples of the segmented images, after implementation of the multi-step segmentation procedure. The first step (not shown) involves identification of the intracranial cavity, using a semi-automated method. The second step, shown in the third column, involves the classification of the intracranial contents into grey matter (grey), white matter (green), cerebrospinal fluid (blue) and hyperintense lesions (yellow). The third step, shown in the last column, corrects for partial voluming artifacts (magenta) that, in step 2, had produced spurious classification of pixels as hyperintensities, at the interface of brain/CSF and bone/CSF. The sections shown are at the level of the lateral ventricles.

Figure 2: Examples of Segmentation Results in Young and Elderly Subjects: The images on the left are representative examples of the results of the segmentation algorithm from four young subjects, shown in axial section at the level of the lateral ventricles. The images on the right are examples, at the equivalent anatomic level, from four elderly subjects. The age and gender of each subject are noted.

Figure 3: Differences in Volume of White Matter with Advancing Age: This figure presents the white matter volume, as a percent of intracranial cavity volume [%ICC], for each subject (female subjects - open circles; male subjects - filled squares). A horizontal line is drawn two standard deviations below the mean for the 20 year old subjects. Note that only subjects 60 and older fall below the horizontal line.

Table: MRI-derived brain tissue volumes by age category


This work was supported by the National Institutes of Health (Grants # P01-AG04953 and # P01-CA67165, Contract # N01-NS-0-2397); RK was supported in part by the Whitaker Foundation. The authors wish to thank Dr. William M. Wells III for stimulating discussions and technical assistance, as well as Dr. Kenneth Jones and Dr. Mary Hyde for statistical consultation and assistance with data analysis. Mr. Mark Anderson, Mrs. Marianna Jakab, and Dr. Andre Robatino are gratefully acknowledged for technical assistance.


1. Jernigan TL, Press GA, Hesselink JR. Methods for measuring brain morphoologic features on magnetic resonance images. Arch Neurol, 1990; 47: 27-32.

2. Jernigan T, Archibald S, Berhow M, Sowell E, Foster D, Hesselink J. Cerebral structure on MRI I: Localization of age- related changes. Biol Psychiat, 1991; 29: 55-67.

3. Pfefferbaum A, Mathalon DH, Sullivan EV, Rawles JM, Zipursky RB, Lim KO. A quantitative magnetic resonance imaging study of changes in brain morphology from infancy to late adulthood. Arch Neurol, 1994; 51: 874-887.

4. Lim KO, Zipursky RB, Watts MC, Pfefferbaum A. Decreased gray matter in normal aging: an in vivo magnetic resonance study. J Gerontol, 1992; 47: B26-30.

5. Murphy DG, DeCarli C, Schapiro MB, Rapoport SI, Horwitz B. Age-related differences in volume of subcortical nuclei, brain matter, and cerebrospinal fluid in healthy men as measured with magnetic resonance imaging. Arch Neurol, 1992; 49: 839-845.

6. Coffey CE, Wilkinson WE, Parashos IA, et al. Quantitative cerebral anatomy of the aging human brain: a cross-sectional study using magnetic resonance imaging. Neurology, 1992; 42: 527-536.

7. Christiansen P, Larsson HB, Thomsen C, Wieslander SB, Henriksen O. Age dependent white matter lesions and brain volume changes in healthy volunteers. Acta Radiologica, 1994; 35: 117-122.

8. Harris GJ, Schlaepfer TE, Peng LW, Lee S, Federman EB, Pearlson GD. Magnetic resonance imaging evaluation of the effects of ageing on grey-white ratio in the human brain. Neuropath Appl Neurobiol, 1994; 20: 290-293.

9. Matsumae M, Kikinis R, Morocz IA, Lorenzo AV, Sandor T, Albert MS, Black PMcL, Jolesz FA. Age-related changes in intracranial compartment volumes in normal adults assessed by magnetic resonance imaging. J Neurosurg, 1996; 84: 982-991.

10. Tanna NK, Kohn MI, Horwich DN, et al. Analysis of brain and cerebrospinal fluid volumes with MR imaging: impact on PET data correction for Atrophy. Part II. Aging and Alzheimer dementia. Radiology, 1991; 178: 123-130.

11. Wickelgren I. For the cortex, neuron loss may be less than thought. [news]. Science, 1996; 273: 48-50.

12. Kikinis R, Shenton M, Gerig G, et al. Routine quantitative analysis of brain and cerebrospinal fluid spaces with MR imaging. J Magn Reson Imaging, 1992; 2: 619-629.

13. Cline H, Lorensen W, Kikinis R, Jolesz F. Three-dimensional segmentation of MR images of the head using probability and connectivity. J Comp Assist Tomog, 1990, 14: 1037-1045.

14. Wells WM III, Grimson WEL, Kikinis R, Jolesz FA. Adaptive segmentation of MRI data. IEEE Trans Med Imag, 1996; 15: 429-442.

15. Guttmann CRG, Weiner HL, Kikinis R, Hsu L, Ahn SS, Hohol MJ, Orav EJ, Khoury SJ, Hafler DA, Mackin GA, Jolesz FA. A prospective study of multiple sclerosis: correlation between MRI, clinical and immunological findings. Proc Soc Mag Res, 1995; 2: 1292.

16. Guttmann CRG, Kikinis R, Anderson MC, Jakab M, Weiner HL, Jolesz FA. Quantitative follow-up of patients with Multiple Sclerosis using MRI: Reliability. Submitted.

17. Kikinis R, Guttmann CRG, Metcalf D, et al. Quantitative follow-up of patients with multiple sclerosis using MRI: Technical Aspects. Submitted.

18. Haug H. Macroscopic and microscopic morphometry of the human brain and cortex. A survey in the light of new results. Brain Pathol, 1984; 1: 123-149.

19. Leuba G, Garey L. Comparison of neuronal and glial numerical density in primary and secondary visual cortex. Exp Brain Res, 1989; 77: 31-38.

20. Pakkenberg B, Gundersen H. Neocortical neuron number in humans: Effect of sex and age. J Comp Neurol, 1997; 384: 312-320.

21. Miller A, Alston R, Corsellis J. Variations in age in the volumes of grey and white matter in the cerebral hemisphere of man: Measurements with an image analyzer. Neuropath Appl Neubiol, 1980; 6: 119-132.

22. Brody, H. Organization of cerebral cortex III. A study of aging in the human cerebral cortex. J Comp Neurol, 1955; 102: 511-556.

23. Brody, H. Structural changes in the aging nervous stytem. Interdiscipl Top Gerontol, 1970; 7: 9-21.

24. Colon, EJ. The elderly brain. A quantitative analysis of the cerebral cortex in two cases. Psychiatr Neurol Neurochir, 1972; 75: 261-270.

25. Shefer V. Absolute number of neurons and thickness of the cerebral cortex during aging, senile and vascular dementia, and Pick's and Alzheimer's diseases. Neurosci Behav Physiol, 1973; 6: 319-324.

26. Henderson G, Tomlinson B, Gibson P. Cell counts in human cerebral cortex in normal aduls throughout life, using an image analysing computer. J Neurol Sci, 1980; 46: 113-136.

27. Anderson JM, Hubbard BM, Coghill GR, Slidders W. The effect of advanced old age on the neurone content of the cerebral cortex. Observations with an automatic image analyser point counting method. J Neurol Sci, 1983; 58: 233-244.

28. Vincent S, Peters A, Tigges J . Effects of aging on neurons within area 17 of rhesus monkey cerebral cortex. Anat Rec, 1989; 223: 329-341.

29. Tigges J, Herndon J, Peters A. Neuronal population of area 4 during life span of rhesus monkeys. Neurobiol Aging, 1992; 11: 201-208.

30. Amaral D. Morphological analyses of the brains of behaviorally characterized aged nonhuman primates. Neurobiol Aging, 1993; 14: 671-672.

31. Peters A, Leahu D, Moss M, McNally K,The effects of aging on Area 46 of the frontal cortex of the rhesus monkey. Cerebr Cortex, 1994; 6: 621-635.

32. Rosene D. Comparing age-related changes in the basal forebrain and hippocampus of the rhesus monkey. Neurobiol Aging, 1993; 14: 669-670.

33. West M, Coleman P, Flood D, Troncoso J. Differences in the pattern of hippocampal neuronal loss in normal ageing and Alzheimer's disease. Lancet, 1994; 344; 769-772.

34. Rosene D, Lai Z, Killiany R, Moss M, Jolesz D, Sandor T, Albert M. Age-related loss of white matter with preservation of gray matter n the forebrain of the rhesus monkey: An MRI study. Neurobiol Aging, in press.

35. Kemper T. The relationship of cerebral cortical changes to nuclei in the brainstem. Neurobiol Aging, 1993; 14: 659-660.

36. Kemper T, Moss M, Rosene D, Killiany R. Age-related neuronal loss in the nucleus centralis superior of the rhesus monkey. Acta Neuropath. in press.

37. McGeer P, McGeer E, Suzuki J. Aging and extrapyramidal function. Arch Neurol, 1977; 34: 33-35.

38. Severson J, Marcusson J, WInblad B, Finch C. Age-correlated loss of dopaminergic binding sites in human basal ganglia. J Neurochem, 1982; 39: 1623-1631.

39. Peters A, Rosene D, Moss M, Kemper T, Abraham C, Tigges J, Albert M. Neurobiological bases of age-related cognitive decline in the rhesus monkey. J Neuropath Exp Neurol, 1996; 8: 861-874.

40. Kemper T. Neuroanatomical changes associated with aging in the central nervous system. In Clinical Neurology of Aging, ML Albert & J Knoefel (eds). New York: Oxford University Press, 1994, 3-67.

41. Barnes C, Suster M, Shen J, McNaughton B. Multistability of cognitive maps in the hippocampus of old rats. Nature, 1997; 388: 272-275.

42. Stafford J, Albert M, Naeser M, Sandor T, Garvey A. Age related differences in computed Wang P, Jernigan T. Morphometric studies using neuroimaging. Neurol Clin, 1994; 12; 789-802.

43. Warfield S, Dengler J, Zaers J, Guttmann CRG, Wells WM III, Ettinger GJ, Hiller J, Kikinis R. Automatic identification of grey matter structures from MRI to improve the segmentation of white matter lesions. J Image Guided Surg, 1996; 1: 326-338.

44. Zijdenbos A, Evans A, Riahi F, Sled J, Chui H-C, Kollokian V. Automatic quantification of multiple sclerosis lesion volume using stereotactic space. In Proceedings of the Fourth International Conference on Visualization in Biomedical Computing (VBC), Hamburg, Germany, 1996.