Sigmoid fits to locate and characterize cortical boundaries in human cerebral cortex.

Quantitative evaluation of neuropathology within the cortex often requires a significant investigator time commitment. Here we elaborate on a method of quantifying the distinctiveness of the gray-white matter boundary using function fitting methods (Avino and Hutsler, 2010) and demonstrate that it can also be adapted to reliably identify the location of the gray matter/white matter (GM-WM) boundary in microscopic images, even when that boundary is indistinct. Multiple images of the gray-white matter boundary were acquired from sixteen subjects. Density profiles across the cortical layers were acquired and sigmoid functions were iteratively fit to the density profiles until a best fit was found. The slope of the resulting sigmoid was used to describe both the position and distinctiveness of the GM-WM boundary. Subsequently, two raters laid cortical boundaries on the same set of images and agreement between the raters, as well as agreement between each rater and the transverse-based boundaries, was assessed. Computer-generated boundaries showed reliably higher agreement with each individual rater, relative to the agreement between individual raters. Error between the raters and the transverse-based boundaries was associated with those images where the boundary was less distinct as assessed by the sigmoid slopes. These findings suggest that transverse-based boundaries are superior to user-generated boundaries. Furthermore, these findings suggest that rater-based boundary definitions in both neurotypical and pathological cases may become unreliable as the number of cell profiles found in the subplate region increases, as is the case in both autism and schizophrenia.