Dia Trambitas-Miron and David Talby on what makes NLP in Health/Pharma/Biotech challenging and unique.
Subscribe: Apple • Android • Spotify • Stitcher • Google • AntennaPod • RSS.
This week’s guests are Dia Trambitas-Miron (Head of Product) and David Talby (CTO) of John Snow Labs, the startup behind the popular open source project, Spark NLP. The company also has a suite of products including an NLP platform targeted specifically for the healthcare, pharmaceutical, and biotech sectors. In this episode we discuss key NLP use cases in the healthcare sector, and what makes NLP in Health/Pharma/Biotech challenging and unique. We also look at broader trends in NLP and machine learning and discuss how those trends impact applications in healthcare.
Dia Trambitas-Miron:
Recent deep learning and machine learning models have topped human performance in different tasks. But they are not 100% accurate. This is why we want to keep a human in the loop. We offer tools for clinicians and for health experts to quickly analyze text, but we also give them evidence from where this data was extracted, so that they can judge at the end if the data is correct or not. At the end of the day, they are the ones taking the responsibility for the decision or the clinical decision taken.
David Talby:
There are two things you need to do on your own, if you want to do this. One is just the algorithms, the neural architecture themselves, because deep learning models are just different for healthcare. The other thing you absolutely need is the data. We have a sizable team of clinicians who do this, you need people who are specialists.
Highlights in the video version:
- Introduction to Dia Trambitas-Miron and David Talby
What are current applications of NLP in healthcare and life sciences?
Unstructured text to structured data
Information extraction accuracy in healthcare and human in the loop
End-to-end models
Beyond specificity of the models, what are the other challenges for NLP in healthcare?
Tuning and customizing models
What data gathering is needed to train healthcare models?
What does it entail to build these models?
No tolerance for error in the medical domain
Are large language models relevant to healthcare NLP?
Give us a sense of the people you work with in healthcare
How much data do domain experts need to start a model?
How much model tuning do the data science teams you interact with do?
Do you have specific product features around privacy and confidentiality?
What is the first step to build a model from a large number of doctor notes?
De-Identification
Healthcare NLP maturity compared to other sectors
ML Community and confidential computing technologies
Spark NLP open source project
Bleeding Edge Models and challenging problems
Healthcare NLP and support for other languages
OCR
Healthcare NLP Summit
Related content:
- A video version of this conversation is available on our YouTube channel.
- Parisa Rashidi: “Machine Learning in Healthcare”
- Moshe Wasserblat: “Democratizing NLP”
- Yoav Shoham: “Making Large Language Models Smarter”
- Nicholas Boucher: “Imperceptible NLP Attacks”
FREE Virtual Conference:
[Image: Medical Records in Hospital from Storyblocks.]