Spotlight #13: 🧐 Clarifying algorithmic bias, 💊 Personalised drug treatment, 💉 ChatGPT and plastic surgery, 🌧️ Generative AI to improve clinical workflows
AI Health Hub, 11/12/2023
🧐 Algorithmic Bias, Generalist Models, and Clinical Medicine
Paper
This paper is part of Google Research publications in Health and Bioscience, and it is a must-read to grasp the different ways algorithms can be biased, looking at it from a philosophical and ethical angle.
The clinical machine learning landscape is predominantly occupied by narrow models, so trained on specific biomedical datasets for particular clinical tasks. However, generalist models in this field are on the rise. Therefore, this article delves into the different biases inherent in these two model types, emphasizes their ethical implications, and proposes strategies for mitigating algorithmic biases.
💊 AI enabled soft robotic implant monitors scar tissue to self-adapt for personalised drug treatment
News
This research by the University of Galway and Massachusetts Institute of Technology (MIT) combines soft robotic and machine learning to achieve personalised drug delivery. An innovative implantable device has been developed with dual functionality: it delivers a drug but also detects early signs of the body rejecting it. Through the use of AI, the implant can adapt its shape, ensuring consistent drug dosage, as it bypasses the scar tissue build up. Thanks to mechanotherapy, a technique that involves regular movements in the body of soft robotic implants, the build up of scar tissue around the implants can be prevented. With the addition of AI, it is possible to account for the individual patient’s immune response.
The full article can be found here (not open access).
💉 Plastic Surgery and Artificial Intelligence: How ChatGPT Improved Operation Note Accuracy, Time, and Education
Scientific Article
How can large language models (LLMs) be used for plastic surgery? In note taking, of course! We are clearly seeing a pattern to use LLMs for reporting in healthcare, highlighting its great need in supporting providers in their tasks.
In this study, a group from the Department of Plastic and Reconstructive Surgery of Queen Victoria Hospital (West Sussex, United Kingdom) evaluated operative notes for plastic surgery procedures generated using ChatGPT. The AI-generated operative notes took considerably less time to create than human-generated notes (5.1 seconds vs 7.10 minutes; P<.05), with 100% of the ChatGPT notes adhering to the current guidelines. Despite the need to be edited, the overall process of editing remained faster than the traditional method of handwriting or computer typing because the edits were largely less than 5 words and took ∼1-2 minutes on average to edit. Surgeons and patients expressed high satisfaction. In the article, the group highlighted that the text prompt engineering played a crucial role in using ChatGPT for generating operative notes. Overall, the benefits listed were efficiency, consistency, customizability, accessibility, integration and data analysis.
It’s worth checking the full article here. The supplemental online material includes a video showing the process of note generation by ChatGPT.
🌧️ HCA Healthcare Collaborates With Google Cloud to Bring Generative AI to Hospitals
Press Release
HCA Healthcare, Inc. (NYSE:HCA), is one of the leading healthcare providers in the United States and one of the largest corporations. HCA Healthcare announced an additional direction to their ongoing collaboration with Google Cloud set on implementing generative AI to improve clinical workflows. The partnership began in 2021 for what concerned privacy and security of patient data. It carried on with the addition of Augmedix to provide a solution that could easily document medical information during patient visits. The new opportunity HCA Healthcare is targeting through generative AI is patient handoffs between nurses. Using Google’s Cloud LLMs, HCA Healthcare’s Care Transformation and Innovation team (CT&I) built a system that can generate handoff reports, which is undergoing testing with nurses. The long-term goal is to use Med-PaLM 2 LLM to support clinicians.
🥳 I hope you enjoyed the findings!
We are always curious to hear your feedback. Do you have thoughts on the new substack page? Would you like to read more about a specific topic? Maybe you work in this field and would like to contribute with an opinion piece, or suggest one for us to write on?
We would love to hear it, so leave a comment or send us a message 🤓.