Automating medical diagnostics has emerged as a promising field of research that has the potential to ease the burden on clinics and enhance patient outcomes. Despite their numerous benefits, medical specialists remain hesitant to place their trust in automated approaches. It is postulated that the solution may lie in probabilistic machine learning, which can improve trustworthiness by enabling diagnostic models to express a level of confidence in their diagnoses. The aim for this work is to develop a general approach to estimate uncertainty in automated diagnostics. This will be illustrated through an application that uses Mueller polarimetry, which is suitable for various reasons. As a powerful, new diagnostic technique, bringing Mueller polarimetry closer to clinical implementation is valuable to researchers who use it. It also constitutes a relatively low-dimensional problem that matches the probabilistic approach well. Such an approach can help estimate and mitigate the different types of uncertainty that currently undermine model predictions, thereby reducing specialistsâ€^TM own uncertainty towards automated diagnostics.