Neuropathologists assess vast brain areas to identify diverse and subtly differentiated morphologies. Alzheimer's disease pathologies have different density distributions in grey matter (GM) and white matter (WM), making the task of separating GM and WM necessary to neuropathologic deep phenotyping. Standard methods of segmentation typically require manual annotations, where a trained observer traces the boundaries of GM and WM on digitized tissue slide images using software like Aperio ImageScope or QuPath. This method can be time-consuming and can prevent the analysis of large amounts of slides in a scalable way. In this paper, we propose a CNN-based approach to automatically segment GM and WM in ultra-high-resolution whole slide images (WSIs) by transforming the segmentation problem into a classification problem. Contrary to the traditional image processing segmentation method, our technique is flexible, robust, and efficient with the accuracy of 77.43% in GM and 79.42% in WM on our hold-out WSIs.