In this chapter, we describe a step-by-step implementation of an automated anatomical MRI feature extractor based on artificial intelligence machine learning for classification. We applied the DenseNet—a state-of-the-art convolutional neural network producing more robust results than previous deep learning network architectures—to data from male (n = 400) and female (n = 400), age-, and education- matched healthy adult subjects. Moreover, we illustrate how an occlusion sensitivity analysis provides meaningful insights about the relevant information that the neural network used to make accurate classifications. This addresses the “black-box” limitations inherent in many deep learning implementations. The use of this approach with a specific dataset demonstrates how future implementations can use raw MRI scans to study a range of outcome measures, including neurological and psychiatric disorders.