Many patients with epilepsy do not achieve adequate pharmacologic control of their seizures and must consider surgical options. Many such patients undergo temporal lobectomy and experience a marked reduction in the frequency and severity of their seizures. However, many are less fortunate. One suspected factor for the latter group is the limited ability of clinical imaging to delineate subtle epileptogenic abnormalities, leading to subtotal resection of lesional tissue. A long range goal in this field is to increase the sensitivity and specificity of detecting such abnormalities by “training” MRI with pathology, feature analysis and machine learning. A key component of this is the ability to segment histopathology to facilitate its mapping to co-registered MRI. A foundational step in this process is to determine whether or not algorithms are capable of detecting the architectural abnormalities in cortical dysplasia on the basis of these segmentations. In brief, reliable semi-automated segmentations were developed to extract a number of features including neuron size, clustering, eccentricity, field-fraction and polarity. Feature analyses using t-Distributed Stochastic Neighbor Embedding (t-SNE) plots demonstrate a non-random association between selected features and diagnostic categories. These results indicate that automated algorithms are capable of distinguishing dysplastic from normal cortex on the basis of semi-automated segmentations.