Computed detection and quantitative morphometry of Alzheimer senile plaques

High-throughput quantification of Alzheimer's disease pathological markers in the post-mortem human brain

Quantitative analysis of amyloid plaques and neurofibrillary tangles is central to many Alzheimer's disease studies. A novel approach for quantitative immunohistochemistry of plaques and tangles has arisen from the need to account for the heterogeneous expression pattern of these markers in the human brain. This approach aims to overcome the human bias inherent to many sampling strategies, to account for the effects of tissue shrinkage resulting from antigen-retrieval procedures, and to accelerate the analysis of large sample sets by using a high-throughput quantification system. The procedure entailed three coordinated steps: acquisition of montaged images of entire tissue sections, randomised sampling across the cortex, and automated quantification of the selected samples with morphometric image analysis software. Two-dimensional estimates of plaque and tangle densities were obtained from the superior temporal gyrus and middle temporal gyrus of Alzheimer's disease and normal human brains. Results showed a robust correlation between the numbers of plaques and tangles quantified by automated image analysis and those acquired by manual counting. Correction for antigen-retrieval tissue shrinkage ensured that density measurements were not over-estimated. The value and applicability of this assay was demonstrated by the statistically significant differences observed between the averaged densities of plaques and tangles within different investigational groups. We report an accurate and objective approach to the quantification of plaques and tangles in human brain tissue. Implementation of a randomised sampling strategy coupled with a reproducible automated quantification system will facilitate more rigorous comparison of quantitative data derived from different immunohistochemical studies.

Assessing statistical applications in publications on Alzheimer's disease

BACKGROUND/AIMS

To evaluate statistical applications in publications on Alzheimer's disease (AD).

METHODS

Three instruments/checklists were developed: Assessment of Statistical Reporting (ASR; 44 items), Survey of Statistical Designs (SSD; 10 items), and Survey of Statistical Methods (SSM; 7 items). After a pilot testing on 5 AD publications, the instruments/checklists were revised and tested for reliability with a sample of 30 AD articles and for validity with another sample of 10 AD articles from MEDLINE.

RESULTS

Item-specific test-retest and interrater reliability for ASR ranged from 0.29 to 1.0 with the associated standard errors (SEs) ranging from 0.01 to 0.31. The test-retest reliability (intraclass correlation coefficient = 0.94, 95% CI: 0.88-0.97) and the interrater reliability (intraclass correlation coefficient = 0.84, 95% CI: 0.69-0.92) for the overall score of ASR were high. The correlational validity of the ASR with a published checklist was also high (r = 0.74, SE = 0.24). The item-specific test-retest reliability in SSD and SSM ranged from 0.58 to 1.00 with the associated SEs ranging from 0.01 to 0.32. The item-specific interrater reliability in SSD and SSM ranged from 0.17 to 1.00 with the associated SEs ranging from 0.01 to 0.22.

CONCLUSIONS

This study suggested that it was feasible to assess statistical applications in AD publications.

BioVision: An application for the automated image analysis of histological sections

We describe a computer application, "BioVision", that can be trained to quickly and effectively classify and quantify user definable histological objects (e.g., senile plaques, neurofibrillary tangles) within single or double-labeled immunocytochemically stained sections. For a given image population, BioVision is interactively trained (in Independent User Mode) by an investigator to perform the desired classifications. This training yields a statistical model of the different types of objects occurring in the target image population. The resulting model can then be used (in Automated User Mode) to classify all objects in any image or images from the target population. BioVision simplifies the quantification of complex visual objects and improves inter-rater reliability. The program accomplishes classification in two major stages: pixel classification and blob classification. In pixel classification, each pixel is assigned to one of some number of substance classes, based on its chromatic properties and local context, reflecting basic histological distinctions of interest. In the blob classification phase, the image's pixels are first partitioned into "blobs": maximal connected sets of pixels assigned to the same substance class. Then, based on its size, shape, textural and contextual properties, each blob is assigned to a histological object class. A Bayesian classifier is used in each of the pixel and blob classification stages. We report several tests of BioVision. First, we applied BioVision to classify senile plaques and neurofibrillary tangles in several test cases of Alzheimer's brain immunostained for beta-amyloid and PHF-tau and compared the results to those produced by experienced investigators. BioVision was trained to classify Plaque-type blobs as either plaques or plaque-type nonentities, and tangle-type blobs as either tangles or tangle-type nonentities. BioVision classified the objects with an accuracy comparable to the trained investigator. Next, we applied BioVision to the task of counting all the tangles in hippocampal images from 22 Alzheimer's disease (AD) cases selected to span a broad range of dementia levels from the tissue repository of UC Irvine's Center for the study of Brain Aging and Dementia. The tangle counts produced by BioVision proved to be significantly better predictors of the cases' adjusted MMSE scores than any of tangle load, age at death, post mortem interval or the interval between the last MMSE score and death.

Combining correlated diagnostic tests: application to neuropathologic diagnosis of Alzheimer's disease

This article studies the problem of combining correlated diagnostic tests to maximize the discriminating power between the diseased population and the healthy population. The authors consider all possible linear combinations of multiple diagnostic tests and search for the one that achieves the largest area under the receiver operating characteristic (ROC) curve. They discuss the statistical estimation of the optimum linear combination test and the associated maximum area under the ROC curve. Their approach is based on the assumption of multivariate normal distribution of the multiple diagnostic tests. They also present the application of the proposed techniques to the neuropathologic diagnosis of Alzheimer's disease based on brain lesions from 5 different brain locations using a data set from the Washington University Alzheimer's Disease Research Center.

Neuropathologic Criteria for Diagnosing Alzheimer Disease in Persons with Pure Dementia of Alzheimer Type

Universally accepted neuropathologic criteria for differentiating Alzheimer disease (AD) from healthy brain aging do not exist. We tested the hypothesis that Bielschowsky silver stained total, cored, and neuritic senile plaques (TSPs, CSPs, and NSPs, respectively), rather than neurofibrillary tangles (NFTs), best discriminate between the 2 conditions using rigorously defined nondemented (n = 7) and AD (n = 35) subjects with no known co-morbidities. We compared lesions in 3 neocortical regions, in hippocampal CA1, and in entorhinal cortex in 19 men and 13 women between 74 and 86 years at death. The Clinical Dementia Rating (CDR) was used to assess degree of cognitive impairment within a year of demise. Neocortical TSP measures provided the highest correlation with expiration CDR: area under the curve (AUC) = 0.986 with 97.8% sensitivity at 90% specificity with an estimated cut-point of 6.0 TSP/ mm2. All SP measures yielded higher estimated AUC and sensitivity for 90% specificity compared to NFTs. Derived TSP cut-points applied to 149 persons with clinical AD regardless of their neuropathologic diagnosis yielded a sensitivity of 97% and specificity of 84% for TSPs in the 3 neocortical areas. Thus cut-points based on both diffuse and neuritic SP in neocortical regions distinguished nondemented and AD subjects with high sensitivity and specificity.

Multiscale detection and analysis of the senile plaques of Alzheimer's disease

Senile plaques (SP) are one of the characteristic neuropathologic lesions of Alzheimer's Disease (AD), and studies of SP cortical distribution, density (number of SP/mm2), and morphology are expected to lead to new information about the mechanism and pathogenesis of AD. We describe a digital image analysis procedure to detect SP, and to measure SP size, shape, and total fractional area in digital micrographs of silver-stained tissue sections. This histology is nonspecific so the program detects all the significant stained objects and a classifier sorts the SP from other tissue elements. SP vary greatly in size and form, and detection is based on multiscale template correlation. Three independent comparisons of computed versus expert-determined SP densities produced correlation coefficients greater than 0.8. The program found 94,000 SP in 2800 digital images of tissue sections from 42 postmortem cases including healthy aged controls and severely demented subjects.