We describe a new, observer-independent procedure for identifying boundaries between cortical areas. The method is useful for images obtained from sections which provide microstructural information on the cortical laminar pattern, e.g., Nissl-, myelin-, or immunohistochemically stained sections or receptor autoradiographs. The laminar pattern is represented by profile curves extending from the cortical surface to the white matter boundary. These profiles are constructed from digitized images. Digitization is based on the grey level index (Nissl) or densitometry (myelin, immunohistochemistry, receptor autoradiography). The shapes of neighboring profiles are compared by calculating their distances according to feature vectors extracted from the profiles. Profiles derived from a homogeneous area can be expected to be similar in shape and hence show low distance values between each other. Maximum distances can be found between profiles which lie on opposite sides of a structural boundary. The Mahalanobis distance was found to be more sensitive and to yield greater spatial resolution than other distance measures such as the Euclidean distance. Cell-stained sections of the human neocortex were analyzed. The method not only verified boundaries which had been defined by visual inspection, it also revealed new ones which had not been detected visually. The procedure offers an important supplement to the traditional methods based on visual inspection which, for the first time, is based on quantitative data and therefore offers a new level of reproducibility and observer independence. Anatomical atlases based on this procedure thus provide a new tool for the interpretation of structural data obtained from functional imaging techniques.
Copyright 1999 Academic Press.