Color image analysis in neuroanatomical research: Application to senile plaque subtype quantification in Alzheimerʹs disease

Many problems in neuroanatomy and neuropathology require the collection of large data sets and would benefit from a method that allows for rapid quantitative analysis to be carried out on a routine basis. An example is the quantification and subtype classification of the number of senile plaques in post-mortem Alzheimerʹs disease tissue. A method to reliably automate the analysis of plaques and their underlying subtypes would allow more rigorous and quantitative correlations to be investigated. Computer assisted image analysis of data typically utilizes gray scale images. These methods, however, are only applicable to quantification of objects labeled with a single marker. We sought to extend this type of analysis to double-labeled tissue sections so we could quantify dual labels separately based on their peroxidase color characteristics, analyze the resultant occurrence of overlap between the two labels, and classify senile plaques into discrete subtypes. We present a method for semi-automated color image analysis which allows one to identify separate labels based on histogram mapping of hue, saturation and value as well as apply overlapping feature detection algorithms. The technique is application driven, so that a trained observer can set threshold or object criteria and verify the desired results. These methods were able to yield total “amyloid load” and “dystrophic neurite load” values, generate plaque histograms based on total size, and subtype plaques into diffuse/primitive and neuritic/classical categories. By adjusting feature criteria, we were able to achieve promising agreement (Fisherʹs R to Z correlation of 0.94) between a human observer and the computer algorithm in the classification of plaque subtypes using three AD cases.