Home / SmartTech / Nvidia researchers describe an AI-based clinical speech transcription system

Nvidia researchers describe an AI-based clinical speech transcription system



At the 2020 Conference on Machine Intelligence in Medical Imaging, held practically this year, Nvidia researchers presented a paper describing an AI system that captures and transcribes the speech of clinical patients. The system identifies clinical words and assigns the words to a standardized health database. Tasks that the researchers say could ease the pressure on clinicians if they experience pandemic-related overload.

The co-authors suggest telemedicine as one possible use of the system, an area that has seen unprecedented demand during the coronavirus pandemic. In March, virtual health consultations grew 50%, according to Frost and Sullivan research. General online doctor visits are expected to reach 200 million this year.

The heart of the researcher system is a BERT-based language model that is trained in a self-monitored manner on a text dataset. (Self-supervised learning is a means of training models to perform tasks without providing labeled data.) Bio-Megatron, a model with 345 million parameters ̵

1; in-model configuration variables – recorded and learned patterns from 6.1 billion words extracted from PubMed were. a search engine for abstracts on life science topics.

After pretraining, the model was refined using a clinical natural language processing dataset created by a previous National Center for Biomedical Computing agreement funded by the National Institutes of Health (NIH). It was then incorporated into an automatic speech recognition component that performs word identification and compares words with concepts in the Unified Medical Language System (UMLS), an ontology developed by the National Library of Medicine at the NIH.

In experiments with Nvidia V100 and T4 graphics cards, the researchers report that Bio-Megatron achieved an accuracy of 92.05% after 1 millisecond processing, taking into account precision and recall. “This opens up significant new capabilities in systems where patient, clinical and researcher response is paramount. An automated speech recognition model that can extract and relate important clinical concepts from clinical conversations can be very useful,” you write. “We hope that our contribution will help achieve faster and better patient responses, which will ultimately lead to improved patient care.”

Nvidia’s contribution to the research community comes after Microsoft co-authors proposed a cutting-edge biomedical language model called PubMedBERT. They said they achieved industry-leading results on tasks like named company discovery, evidence-based medical information extraction, document classification, and more.


Source link