Interesting reads ... July 2024

Interesting reads ... July 2024

I discuss the critical role of explainability in healthcare AI, highlighting the necessity for transparent AI decision-making to build trust and maintain ethical standards in medical applications. The paper examines techniques like saliency maps and class activation maps to enhance clarity in AI decisions, emphasizing the importance of continuous improvement and human-centered design for effective and ethical AI integration.

https://www.dhirubhai.net/posts/janbeger_as-ai-technologies-become-increasingly-embedded-activity-7217501001107312640-PQR1

DOI: 10.1016/j.ejrad.2024.111507


Gentile and Malara discuss the application of AI in cancer screening and surveillance, highlighting its role in improving cost-effectiveness through enhanced risk assessments and early diagnosis. They also emphasize the challenges of long biomarker validation times, underrepresentation in trials, and communication difficulties, while noting AI's success in multicentre studies and its potential for continuous improvement and real-time validation.

https://www.dhirubhai.net/posts/janbeger_this-paper-discusses-the-application-and-activity-7218101909587447808-HmSo

DOI: 10.1016/j.esmorw.2024.100046


Christopher Longhurst , Karandeep Singh , Aneesh Chopra , Ashish Atreja, MD, MPH , John Brownstein argue for the establishment of AI Implementation Science Centers to rigorously evaluate the real-world clinical effectiveness of AI models in healthcare settings. They emphasize that local validation and continuous monitoring are crucial for ensuring ethical use, integrating AI into clinical workflows, and maintaining its effectiveness in diverse and dynamic healthcare environments.

https://www.dhirubhai.net/posts/janbeger_this-paper-argues-for-the-establishment-of-activity-7217377144564310016-dBLZ

DOI: 10.1056/AIp2400223


Archana Reddy Bongurala , Dhaval Save MD, Ankit Virmani MSc, and Rahul Kashyap, MBBS, MBA, FCCM , discuss how AI tools like voice-to-text transcription and automated note generation can reduce the time primary care physicians spend on EHR-related tasks, thereby potentially mitigating clinician burnout and improving job satisfaction. However, they also highlight the challenges AI introduces, such as data bias, security concerns, overreliance, regulatory hurdles, and alert fatigue.

https://www.dhirubhai.net/posts/janbeger_this-paper-explores-the-transformative-potential-activity-7215202811343470593-wjjB

DOI: 10.1016/j.mcpdig.2024.05.006


Horiuchi, D., Tatekawa, H., Oura, T. et al. found that GPT-4-based ChatGPT, with a diagnostic accuracy of 43%, significantly outperformed GPT-4V-based ChatGPT, which had an accuracy of 8%, in musculoskeletal radiology. GPT-4-based ChatGPT's accuracy was comparable to that of a radiology resident but lower than a board-certified radiologist, and its use improved diagnostic accuracy for both residents and board-certified radiologists, particularly in nontumor cases and bone tumor diagnoses.

https://www.dhirubhai.net/posts/janbeger_exploring-the-opportunities-and-challenges-activity-7220665994732552192-e6jf

DOI: 10.1007/s00330-024-10902-5


Anjali Rajagopal , Shant Ayanian, MD , Alexander Ryu , Ray Qian , Sean Legler , Eric Peeler, MD, Meltiady Issa, MD, Trevor Coons, MHA, FACHE , and Kensaku Kawamoto, MD, PhD, MHS, FACMI, FAMIA reviewed 139 articles on Machine Learning Operations (MLOps) in healthcare, identifying critical areas such as model monitoring, ethics, workflow integration, and regulatory considerations. They highlighted the need for rigorous studies on the impacts of MLOps practices on patients, pointing out the current focus on statistical model performance rather than practical implementation in clinical settings.

https://www.dhirubhai.net/posts/janbeger_this-paper-provides-a-comprehensive-synthesis-activity-7219559643306491904-592s

DOI: 10.1016/j.mcpdig.2024.06.009


Mohammad?Alkhalaf, Ping Yu , Mengyang Yin, and Chao Deng demonstrated that using the Llama 2 model with zero-shot prompting achieved a 93.25% accuracy in generating structured summaries of nutritional status from unstructured EHR data. They found that incorporating retrieval-augmented generation (RAG) improved summarization accuracy to 99.25%, showing its effectiveness in handling unstructured data by providing additional context, although it did not enhance the model's ability to identify malnutrition risk factors from nursing notes, which stood at 90% accuracy.

https://www.dhirubhai.net/posts/janbeger_applying-generative-ai-with-retrieval-augmented-activity-7213753271822290944-7eRp

DOI: 10.1016/j.jbi.2024.104662


Maurizio Cè, Simona Ibba, Michaela Cellina , Chiara Tancredi, Arianna Fantesini , Deborah Fazzini , Alice Fortunati, Chiara Perazzo , Roberta Presta, Roberto Montanari , Laura Virginia Forzenigo , Gianpaolo Carrafiello , Sergio Papa, and Marco Alì find that while most radiologists have a positive attitude towards AI, with younger and older professionals being especially optimistic, only 36.2% use AI tools daily and just 30% find them decisively beneficial. Additionally, the study reveals that 84% of radiologists emphasize the necessity of the final assessment by a human radiologist, indicating a reluctance to fully rely on AI for decision-making processes.

https://www.dhirubhai.net/posts/janbeger_this-paper-investigates-radiologists-perceptions-activity-7216652367427993600--sjX

DOI: 10.1016/j.ejrad.2024.111590


Tijs vandemeulebroucke argues for a multi-level global ethical approach to AI in healthcare, addressing privacy, bias, power dynamics, and environmental impacts. The paper emphasizes the need for ethical scrutiny at all stages of AI development and integration, considering ecological sustainability and social justice to ensure fair access and minimize health inequities.

https://www.dhirubhai.net/posts/janbeger_the-ethics-of-artificial-intelligence-systems-activity-7222088195176337408-0Hp3

DOI: 10.1007/s00424-024-02984-3


Henning Nilius , Sofia Tsouka , Michael Nagler , and Mojgan Masoodi discuss how machine learning significantly enhances precision medicine by improving diagnostic accuracy, patient stratification, prognosis, and treatment monitoring. They emphasize the need to overcome regulatory, organizational, and methodological challenges to effectively integrate ML into clinical practice, ensuring secure data handling and alignment with clinical workflows.

https://www.dhirubhai.net/posts/janbeger_this-paper-explores-how-machine-learning-activity-7224262505479458816-p10Q

DOI: 10.1016/j.trac.2024.117872



For more, follow me:

LinkedIn: https://www.dhirubhai.net/in/janbeger/

X: https://twitter.com/janbeger

My website: https://janbeger.carrd.co

Alexandre MARTIN

Autodidacte ? Chargé d'intelligence économique ? AI hobbyist ethicist - ISO42001 ? Polymathe ? éditorialiste & Veille stratégique - Times of AI ? Techno-optimiste ?

3 个月
回复
Sarai Pahla

Medical Doctor-Linguist AI/ML Regulator

3 个月

Thanks so much for these newsletters - and for sharing you personal thoughts on XAI!

要查看或添加评论,请登录

社区洞察

其他会员也浏览了