Medical datasets are getting increasingly larger and richer in terms of both amount and heterogeneity of collected data. The analysis and evaluation of these enormous datasets poses a challenge to clinicians and experts in healthcare systems around the world. Thanks to its effectiveness processing and extracting knowledge from large datasets, it is expected that machine learning will become a standard tool in healthcare in the following years, as it will help clinicians in screening, diagnosis, and patient-tailored treatment suggestion. Computer vision will be a fundamental piece of this revolution, whereas more than 90% of the medical data are from images. In this context, I will present a general overview on how we are using computer vision and machine learning in a variety of medical problems, with a special focus on ophthalmology. We will dive into the details of a reinforcement learning method for faster and more efficient patient-attentive perimetry testing, and into the challenges of making computer vision systems robust to unexpected inputs using a novel technique for out-of-distribution (OOD) detection.