“Smart Vet: Autocompleting Sentences in Veterinary Medical Records”, 2019-03-19 (; backlinks; similar):
Every day, veterinarians write tens of thousands of medical records, mostly in standard formats following the SOAP structure: “Subjective”, “Objective”, “Assessment”, and “Plan”. These notes record the findings of their physical exams and observations of their patients, and take countless hours to write.
We present in this paper a new system that we call “Smart Vet” that assists veterinarians in the writing of their notes by suggesting autocompletions for their sentences as they are writing them within the sections of their medical records.
To enable this, we present two approaches: an end-to-end deep learning system that models this task as a seq2seq neural machine translation problem (ie. translate a given sequence of sentences that correspond to the existing medical record into the following sequence that corresponds to the next sentence the veterinarian would want to write) and a transformer-based language modeling system based on OpenAI’s recent advancements.
Based on the success of this latter method, we evaluate this system live in a medical records application, and successfully see our autocompletions being used in production 12.46% of the time—a remarkable success.