How has scientific knowledge in medicine evolved since the mid-twentieth century? We study this question using Medical Subject Headings, a controlled medical taxonomy created and used by the National Library of Medicine since the1940s. We create a high-dimensional classification model to characterize medical literature based on novelty. A paper publication in our analysis is more novel if it is assigned with a newer combination of MeSH keywords. On the population of medical publications from MEDLINE, we find a close association between the MeSH-based novelty measure and the advancements in medical sciences. For example, novelty surged during the War on Cancer years and then again at the turn of the millennium when genetic engineering technologies like CRISPR were invented. We then compare these results with a novelty measure generated by a domain-specific large language model based on paper abstracts and introductions. Interestingly, the AI characterization of scientific advancements over time follows a similar pattern, but the local variations are much smaller and sometimes inaccurate.