Parkinson’s Disease (PD) is one of the most relevant neurodegenerative disorder. It is mainly caused by a loss of dopamine neurons leading to a reduction in the neurotransmitter dopamine, which is essential in the control of movement. While the diagnosis of PD is mainly clinical, new markers are being used with high accuracy in the later stages of the disease, where symptoms are clear. However, the early stages of the disease, when symptoms start to evolve and treatments could potentially be more effective, are yet to be explored. In this work we explore the low-dimensional latent space of the Parkinson’s Progression Markers Initiative (PPMI) DaTSCAN imaging dataset, with a twofold objective: to perform an early diagnosis of PD, and to link the low-dimensional representation of the images to symptomatology. Different unsupervised methods have been used to extract the features (ISOMAP and PCA), and the resulting space is evaluated by means of binary or multiclass classification, and linear regression, using Support Vector Machines (SVM). We obtained a diagnosis of PD with an Area Under the ROC Curve (AUC) above 0.94 for three different variables, and a relevant link between the Unified Parkinson’s Disease Rating Scale (UPDRS) and the imaging composite features with even for a simple linear model. These results pave the way to explore latent representations in PD and study the progression of the disease and its symptomatology.