AI in Healthcare – news picked by GLI /05
Welcome to AI in Healthcare. Our aim with this series is to offer you a meticulously curated compilation of news updates and perspectives personally selected by the Graylight Imaging (GLI) team. Without further ado, we invite you to immerse yourself in the content.
How to shift AI in healthcare from development to deployment
In a paper published in Nature last year [1] researchers from Stanford University offered a data-driven perspective on the innovations and challenges that are defining ML for healthcare. I’m sure we can all agree that over the last decade, significant progress has been made in applying machine learning (ML) to healthcare. As Stanford researchers noted, the success of deploying ML models in clinical settings, however, is dependent on how data is collected, organized, protected, moved, and audited. Additionally, the lack of healthcare-specific benchmarking datasets and robust metrics for evaluating synthetic data quality pose challenges. Despite those complexities, new advanced AI models emerge and some of them aim to evaluate AI-based assessment. Last paper published in RSNA Radiology is a great example here [2]. The authors conducted research to assess AI prediction variability as an uncertainty quantification (UQ) metric for identifying cases at risk of AI failure in cancer diagnosis at MRI and CT across various cancer types, data sets, and algorithms. A total of 18,022 images were employed in the training phase, while 838 images were designated for testing. Identification of groups of patients and data that are prone to artificial intelligence failure in diagnosing is extremely important and, in our opinion, must be performed to successfully implement models in the clinical environment.
2023 Healthcare Provider IT Report
Just a few days ago Bain & Company in cooperation with KLAS Research presented the report ‘2023 Healthcare Provider IT Report: Doubling Down on Innovation’ [3]. According to this survey, which involved 201 US healthcare provider executives, 56% of the participants identified software and technology as one of their top three strategic priorities. This marks a notable increase from the 34% reported in 2022. What is worth mentioning: currently, nearly 6% of respondents have already implemented a generative artificial intelligence (AI) strategy. Remarkably, approximately 50% of them are presently engaged in its development or have imminent plans to do so. What’s more: 70% of health system respondents believe AI in healthcare will have a greater impact on their organization this year than last, shifting AI strategies from the IT department to the C-suite.
Image source: Bain & Company and KLAS Research report ‘2023 Healthcare Provider IT Report: Doubling Down on Innovation’
These figures are impressive, especially when compared to the 2022 results. Will there be a breakthrough in the real-world application of AI? [4] I guess, we’ll see.
Can AI compose clinical documentation?
A report published in JAMA Internal Medicine in 2022 validated that physicians dedicate nearly two hours each day beyond their office hours to complete clinical documentation [5]. What’s more: in a report from The New York Times in 2022, it was suggested that a significant contributor to physician burnout is the disconnect between the profound mission of providing medical care and the bureaucratic aspects of healthcare on the other side [6]. Can AI be a panacea for this pressing problem? The answer is complex.
A recent comparative study [7] on clinical language understanding and large language models (LLM) presents meaningful observations here. The authors provided a comprehensive evaluation of state-of-the-art LLMs, namely GPT-3.5, GPT4, and Bard.
The study emphasizes the need for cautious implementation of LLMs in healthcare settings, ensuring a collaborative approach with domain experts and continuous verification by human experts to achieve responsible and effective use. LLM AI holds great potential in alleviating the challenges associated with clinical documentation and clinician burnout. Can it just simply solve the problem? Of course not, however, AI-driven documentation solutions may assist medical professionals in the battle with the bureaucratic side of their work. The code used in this research is available at https://github.com/EternityYW/LLM_healthcare.
References:
[1] Zhang, A., Xing, L., Zou, J. et al. Shifting machine learning for healthcare from development to deployment and from models to data. Nat. Biomed. Eng 6, 1330–1345 (2022). https://doi.org/10.1038/s41551-022-00898-y
[2] https://pubs.rsna.org/doi/10.1148/radiol.230275
[3] https://www.bain.com/insights/2023-healthcare-provider-it-report-doubling-down-on-innovation/
[4] You might be interested in our latest post, where we discussed the number of health departments utilizing artificial intelligence: https://site.graylight-imaging.com/radiology-ai-landscape-in-numbers-and-money/
[5] https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2790396
[6] https://www.nytimes.com/2022/09/29/health/doctor-burnout-pandemic.html