AI 4 Healthcare Newsletter

Welcome to our newsletter, a meticulously curated platform intended to deliver the most recent news, breakthroughs, and published research from the exciting crossroads of AI and Healthcare.

This is the third week of our newsletter. We hope you find the content valuable. Please share your feedback and comments with us. Our goal is to create the best newsletter covering AI in healthcare.

AI News in Healthcare

‘Oppenheimer’ is a must-watch for everyone who works in AI and health care. A valid question, are we building AI in healthcare like Oppenheimer built the atomic bomb? “Oppenheimer’s story illustrated how there is a disconnect between the goals of science — such as transparency, collaboration, and truth-seeking — and political goals.” Understanding this is particularly urgent now, since the public is continuously learning about how AI will change industries, including health care.

Machine Learning Tools Help Predict Alzheimer’s Progression. An interesting article that talks about experimentation with various tools to predict Alzheimer’s progression in patients that show no symptoms or mild cognitive impairment. More research is needed but an interesting idea that can change the way we approach a disease.

Duke Health and Microsoft partner to 'ethically harness' generative AI. A five year partnership has begun that will combine patient care with technology innovation. This will include the creation of a Duke Health AI Innovation Lab and Center of Excellence.

New AI ultrasound tech is first to land FDA approval to enhance prenatal care: ‘Better health outcomes’. This article states that Sonio, a company based in France, has gained FDA approval for AI technology, that has the ability to scan tens of thousands of ultrasounds to detect warning signs of fetal health issues. Why this is important: This article title is wrong. FDA granted clearance not approval. A clearance means that the AI system can not operate on its own without humans. To date, there is no FDA approved AI system on the market.

AI use in breast cancer screening as good as two radiologists, study finds. It's both troubling and disappointing when a reputable publication like The Guardian reports a misleading title. The study (for more details, please see below) actually demonstrated that AI + radiologist equals 2 radiologists reviewing mammograms. In some countries, like the UK and Sweden, mammograms should be read by two radiologists. Using AI to replace one of them increases efficiency. However, such a study may not translate well to the US, where only one radiologist is required to interpret the study. No insurance company or patient would accept an AI system reading their mammogram independently. What do you think?

Cognizant expands generative AI partnership with Google Cloud, announces development of healthcare large language model solutions. Why this is important: I actually do not know what Cognizant is doing and honestly you should not care. I included this article because we're likely to witness countless companies building large language models on healthcare data and raise a lot of money doing so. So, buckle up. 

AI News in the World

Meta released AudioCraft, a new generative AI tool that can turn text into audio, create any type of music, melody, or soundtrack, and create any sound effect or noise. The sky is the limit for generative AI 😁.

Cities Where AI Threatens Employment the Most (2023). This article breaks down in detail which jobs and which cities are at most risk for potential job loss due to AI. Not sure if this is true but thank God, our city will not be threatened.

Medium (American online publishing platform) decided: "We are a home for human writing, full stop." They plan to crack down on AI-generated text on their platform. Why this is important: This trend to combat AI-generated content is likely to continue. The question arises: how will we discern between AI-generated and human-generated content in the future as it becomes increasingly challenging to differentiate the two? 😅.

In a survey of nearly 13,000 people—from executive suite leaders to middle managers and frontline employees—in 18 countries, only 14% of employees said they received AI training. Why this is important: You can not push AI efforts at your organization without educating your employees.

YouTube experiments with AI auto-generated video summaries. I love the idea, hope it works 😂.

Biden-⁠Harris Administration Secures Voluntary Commitments from Leading Artificial Intelligence Companies to Manage the Risks Posed by AI. Great idea. We need more efforts like this. The question is: Can any government legislation keep up with the pace of AI development these days? No further comment at this time.

The birth of “Chief of generative AI” job title.  Many compnaies like Amazon and others started to creat new positions like prompt engineer, and Chief of Generative AI.

Impactful Publications

In a study published in The Lancet Oncology, researchers looked at AI assisted mammography reading vs. standard reading by two radiologists. Summary of the study: A randomized, controlled, population-based trial was conducted with women aged 40-80 years eligible for mammography screening at four sites in Sweden. A total of 80,033 women were included in the trial. AI-supported screening resulted in 244 screen-detected cancers and 861 recalls from 39,996 participants, with standard screening resulting in 203 cancers and 817 recalls from 40,024 participants. The AI-supported mammography screening showed a similar cancer detection rate to standard double reading but reduced the screen-reading workload by 44.3%, indicating that the use of AI in mammography screening is safe. Why this is important: the first prospective randomized trial to show AI+ Radiologist = 2 radiologists. Unfortunately, many media and tech companies interpret the study as AI is better than 2 radiologist. This is not what the study showed.

Some radiologists do not view AI as a good assistant. Researchers conducted an experiment with professional radiologists that varies the availability of AI assistance and contextual information to study the effectiveness of human-AI collaboration. Interesting findings: 1) Providing AI predictions does not uniformly increase diagnostic quality. 2) Providing contextual information does increase quality. The observed errors in belief updating can be explained by radiologists’ partially underweighting the AI’s information relative to their own and not accounting for the correlation between their own information and AI predictions. The researchers concluded, unless the documented mistakes can be corrected, the optimal solution involves assigning cases either to humans or to AI, but rarely to a human assisted by AI.

Researchers at Stanford and other academic center built Med-Flamingo: a Multimodal Medical Few-shot Learner. This is a generative AI model that can use text and images and come up with an answer to a medical question (please see image below). The model performed well on USMLE tests with images. Why this is important: The future of healthcare could lie in constructing multimodal large language models that utilize clinical, genomic, imaging, and other types of data to drive personalized and more accurate diagnosis and treatment recommendations.

Check out our courses from AI 4 Healthcare

ChatGPT for Healthcare. Discover all you need to know about ChatGPT and its applications in healthcare. Join over 2000 students worldwide who have already enrolled in this course.

No code-low code Machine Learning for Healthcare. This is the only course that provides hands-on examples of machine learning in healthcare. Start building real models immediately upon completion of the course.

Artificial Intelligence and Machine Learning in Healthcare (The Basics). Learn everything you need to know about the basics of AI application in healthcare, explained in simple terminology, without requiring any coding or mathematical experience.