Comparison of manual and semi-automated segmentation methods to evaluate hippocampus volume in APP and PS1 transgenic mice obtained via in vivo magnetic resonance imaging

Highlights

• A semi-automatic and a more accurate manual method were developed to detect difference in volumes of hippocampus between mice.

• The semi-automated segmentation was unable to detect the same level of differences.

• Manual segmentation is a more reliable segmentation method for small structures.

Abstract

Background

Magnetic resonance imaging (MRI) of transgenic mouse models of Alzheimer's disease is valuable to understand better the structural changes that occur in the brain and could provide a means to test drug treatments. A hallmark pathological feature of Alzheimer's disease is atrophy of the hippocampus, which is an early biomarker of the disease. MRI can be used to detect and monitor this biomarker.

Method

Repeated measurements using in vivo 3D T₂-weighted imaging of mice were used to assess the methods. Each mouse was imaged twice in one week and twice the following week and no changes in volume were expected. The hippocampus was segmented both manually and semi-automatically. Registration was done to gain information on shape changes. The volumes from each mouse were compared intra-mouse, between mice and to hippocampus volume values in the literature.

Results

A reliable method was developed which was able to detect difference in volumes of hippocampus between mice when performed by a single individual. The semi-automated segmentation was unable to detect the same level of differences. The semi-automated segmentation method gave larger hippocampus volumes, with 78–87% reliability between the manual and semi-automated segmentation. Although more accurate, the manual segmentation is laborious and suffers from inter- and intra-variability.

Conclusion

These results suggest that manual segmentation is still considered the most reliable segmentation method for small structures. However, if performing longitudinal studies, where there is at least one year between imaging sessions, the segmentation should be done all at once at the end of all the imaging sessions. If segmentation is done after each imaging session, with at least a year passing between segmentations, very small variations in volumes can be missed. This method provides a means to quantify the volume of the hippocampus in a live mouse using manual segmentation, which is the first step toward studying hippocampus atrophy in a mouse model of Alzheimer's disease.

Introduction

Alzheimer's disease (AD) is the most common form of dementia and cognitive decline in the elderly population. One of the pathological features of AD is atrophy of the brain and the hippocampus. The neuronal cell death and loss of the synaptic function and structure of the neurons causes loss of brain tissue and severe atrophy of the brain (Terry, 2000). Atrophy occurs before any noticeable clinically changes. It is one of the first indicators of dementia and the hippocampus is one of the first structures in the brain where atrophy occurs (den Heijer et al., 2010). The earlier that AD can be detected the better the outcome for the patient. Ideally detection of AD prior to any presenting of signs is the goal. MR imaging has become an accepted method of detecting the biomarker of brain and hippocampus atrophy (Jack et al., 2010)

Transgenic mouse models of AD have been used to help understand pathology and physiology of this neurological disease. There are a variety of transgenic models that can show one, two or even three of the key features clinical pathologies of AD. In association with imaging techniques, specifically MRI, researchers have been able to use these models to map changes of the whole brain or specific regions of the brain. In addition, in vivo imaging can provide a window into the biology of the living mouse, allowing the possibility of making repeated measurements of biological function and specific molecular or genetic events within a single mouse over time.

Prior to MR imaging to view changes in the hippocampus of the transgenic mice, histological slices were used. Using histological slices, reduced hippocampus volume was shown in the APP transgenic mouse model. Abnormalities in the hippocampus of PDAPP human V717 mutation in the APP gene under the control of platelet-derived (PD) factor-β promoter) homozygous and heterozygous transgenic mice were seen by histological analysis when compared to wild type mice (Dodart et al., 2000). The ratio of the surface area of the hippocampus to the surface area of the whole brain was reduced, there by signifying hippocampus atrophy. The reduction in size was noticeable at 3–4 months of age. Another study, via the use of cytochrome oxidase histochemistry, reported PDAPP V717F transgenic mice exhibited a 34–37% decrease in hippocampus volume as compared to wild type mice (p < 0.001) at both 3.6 months and 17.5 months (Gonzalez-Lima et al., 2001). The location of the volume reduction was isolated to the dorsal hippocampal area.

While automated segmentation methods for mice brain and brain structures are becoming more and more common, the benchmark still remains manual segmentation (Scheenstra et al., 2009). Generally, with small animal manual brain structure segmentation, the operator/researcher manually defines the border of the structure that needs to be segmented by selecting the corresponding voxels/pixels within a region of interest based on visual inspection at gray-white matter borders where there are differences in signal intensities. Manual registration always requires a trained technician to manually segment the regions of interest.

The unofficial gold standard for automated segmentation is atlas based segmentation (Scheenstra et al., 2009). Typically atlas-based segmentation involves registration of the target image to an atlas (by either affine or rigid transformation to align the two images), a labeling step wherein the labels from the atlas are moved to the target image and a final segmentation step where the target image is segmented. Multiple atlases are also used to give more robust and accurate data (Aljabar et al., 2009). This is done to align the target image to the atlas image. Types of atlas based segmentation include: label propagation; label propagation with multiple atlases, and probabilistic atlas-based segmentation (Cabezas et al., 2011).

T₂-weighted, 2D imaging was used on ten PDAPP transgenic mice and ten wild type mice. Images toward the rostral and caudal ends of the hippocampus were performed. The ratio of hippocampus to brain area was 31% smaller in the transgenic mice compared to wild type. In the caudal section a 13% decrease was found. They attributed the decrease in hippocampus to brain area to a decrease in the hippocampus volume because differences in the average area of the brain were not statistically significant between the transgenic and wild type groups for either the rostral or caudal locations (Weiss et al., 2002).

One of the early longitudinal measurements of brain changes with 3D MRI in a mouse model of AD (PDAPP (V717F) mice) showed statistically significant smaller hippocampus volumes as compared to wild-type mice of the same age at certain time points in the study (Redwine et al., 2003). T₂-weighted 3D MRI was done on mouse heads removed from the rest of the body. The mouse heads were collected from four different age groups: 40 days old, 100 days old, 365 days old and 630 days old. The hippocampus was significantly smaller in 100 days old transgenic mice (16.4 ± 0.64 mm³) when compared to wild type mice (18.7 ± 0.66 mm³) with a difference of 12.3%. This volume change occurred before the accumulation of plaque was apparent in the histology slices. The volume loss began at 40 days and continued until 100 days old, but no volume loss was seen between 100 days and 630 days. At 630 days the volume loss in the transgenic mice was not statistically different from the wild type mice (Redwine et al., 2003).

Manaye et al. (2007) imaged double transgenic mice (APPswe, PSEN1dE9) to examine the volume changes in the brain and hippocampus across the adult lifespan of the mice (4–28 months). Histology was also performed. There was no difference found between the transgenic mice and the wild type mice in the volume of the hippocampal formation (30.4 mm³ APP/PS1; 30.7 mm³ wild type). No difference was observed with the histology sections either. These findings were contrary to earlier studies of Redwine et al. (2003) and Weiss et al. (2002), but had similar results to earlier histological studies (Delatour et al., 2006) that found no atrophy of the hippocampus in APP/PS1 mice.

Using APP (K670N, M671L)/PS1 (M146L) double transgenic mice and wild type controls aged 2.5, 6.5 and 9 months Oberg et al. (2008), performed 3D imaging on a 9.4 T horizontal bore magnet with a volume coil for excitation and a mouse brain surface coil for signal detection. Three mice from each group were sacrificed after the imaging for histology. Volume calculations showed that the transgenic mice had smaller hippocampus volume than their age-matched controls.

Transgenic mice containing the Austrian APP mutation APP (T714I) were generated and compared to APP wild type mice and a change in brain volume was found (Van Broeck et al., 2008). The APP-Au expressing the T714I mutation mice were compared to age-matched APP wild type controls at 12 and 20 months of age. 3D magnetic resonance imaging was done. Mice from both groups were imaged at 12 months and 20 months. While total brain volume was significantly smaller in the APP-Au mice compared to the control at both age groups, no significant volumetric differences were seen for the hippocampus volumes. The authors did see volume reduction in some instances but the variance was high between samples so no statistical significance was achieved.

The results of more recent data looking at in vivo volume changes in APP (K595N, M596) PS1 (M146V) mice with MRI were published by Maheswaran et al. (2009). They indicate that the hippocampal formation in the double transgenic mice increased on a paired basis by 11% over 8 months as compared to the wild type mice that saw only an increase of 0.5%.

In summary, there is no agreement in the literature involving the volume change of the hippocampus of transgenic mice expressing high levels of amyloid plaques. Publications have shown both an increase and decrease in the volume of the hippocampus.

There are a lot of unanswered questions and variation within studies of APP and PS1 mice. To address these questions, our research goal was to develop and evaluate a method of in vivo imaging of the hippocampus that optimizes resolution and contrast so determining regions of interest in the brain can be done with small error. Also we aimed to determine the variability of our manual segmentation of the hippocampus by a single user and between users by imaging the same mice multiple times over a short period. Thirdly, using image registration, we aimed to quantify the changes in the hippocampus size and shape and volume over the study period to note any changes. This image registration was also used to create a semi-automated method of segmentation and a comparison between manual segmentation of the mouse hippocampus to this semi-automated method of segmentation was done. Both manual and semi-automated segmentation were done to determine if semi-automated segmentation produced the same volume measurements as the “gold standard” manual segmentation. Ideally one would like to be able to segment automatically to save on time and increase through put, however without sacrificing accuracy. While both manual and semi-automated segmentation of mouse brain structures such as the hippocampus have been published, often the specifics of how manual segmentation is done are not described in detail. Segmentation is usually described as manual segmentation by counting of voxels or use of an automated method. The reliability of the segmentation methods are rarely discussed in the literature. In this study, a complete explanation of the methods used is presented. Repeat measurements on the same mouse during a period of time where no volume changes are expected are used to assess the reliability of the methods.