The average age of the population is steadily increasing; therefore, we need to look for earlier signs of deterioration in cognitive ability. The earlier that abnormalities are spotted, the sooner and better they can be treated, and MRIs can spot early signs of disease before any symptoms become visible. An MRI is a 3-dimensional medical imaging technique used in radiology to form pictures of the anatomy and the physiological processes of the body in both health and disease. MRI scanners use strong magnetic fields, magnetic field gradients, and radio waves to generate images of the organs in the body. Signs of physical changes include gray-matter loss, the brain shrinking, and the decline of white-matter. Inspired by 2-dimensional and 3-dimensional convolutions, the aim of this project is to use machine learning on the Cam-CAN ageing neuroimaging dataset. Due to the non-linearity nature of ageing, it is hard to comprehensively model using standard linear statistical techniques such as GLM (Generalized Linear Models). The goal will be to predict the age of participants and physiological variables (cognitive ability) from MRIs. Augmentation of linear models by e.g. quadratic terms only partially address this problem since it fails to address other relationships. As an alternative, deep neural networks are known to learn arbitrary non-linear relationships. In recent years, Convolutional Neural Networks (CNNs) have been very successful at extracting features from raw images thereby circumventing the necessity for prior feature selection / dimensionality reduction through methods such as ICA (Independent Component Analysis) Results are encouraging, with roughly 40% accuracy from a 7-class Classification model and achieving Mean Square Error and Mean Absolute Error of around 0.7, achieving better than the results of more basic models like Support Vector Machines. From these results, we believe Convolutional Neural Networks are a justified choice when mapping the relationships between MRIs and non-linear variables. However, we believe better results can be achieved if a larger dataset presents itself, and if greater computational power is available to handle the demanding nature of certain Convolutional Neural Network models. This is because hardware constraints and a limited dataset imposes restrictions on the type of Network architecture we can develop, such as the depth of the model.