Artificial intelligence is making waves across the medical field, with new studies showing promise in predicting eye treatment complications, analyzing heart MRIs, and even designing RNA-based drugs.
While specialized healthcare AI models demonstrate potential, research also cautions against relying on general-purpose AI chatbots for clinical decision-making, highlighting the need for tailored solutions in critical medical applications.
AI Model Shows Promise in Predicting Eye Treatment Complications
A new study suggested AI could help predict complications from treatments for age-related macular degeneration (AMD), a leading cause of vision loss affecting millions of Americans. Researchers from Emory University and Cleveland Clinic developed a machine-learning model that analyzes eye scans to identify patients at risk of inflammatory responses to common AMD treatments.
The study, published in the journal Heliyon, focused on neovascular AMD (nAMD), a severe form of the disease typically treated with anti-VEGF drugs, according to a press release. While effective, these treatments can cause severe eye inflammation in some patients. The AI model, which analyzed optical coherence tomography (OCT) scans, demonstrated accuracy rates of up to 81% in identifying patients likely to develop this complication.
“Our study provides valuable data for clinicians to make better treatment decisions, potentially reducing the dosage or combining these agents with anti-inflammatory drugs to prevent severe complications,” Anant Madabhushi, executive director of Emory AI.Health and principal investigator of the study, said in the release.
The research team analyzed images from 67 eyes in a retrospective clinical trial. While the results were promising, more extensive prospective studies will be necessary to validate the model’s effectiveness in clinical settings. The researchers aim to integrate their algorithms into future clinical trials to test real-time identification of at-risk patients, per the release.
Study Finds General AI Chatbots Unsuitable for Clinical Decision-Making
Healthcare AI firm Atropos’ study revealed popular chatbots like ChatGPT falter in clinical decision-making, underscoring the need for specialized AI in critical medical applications.
According to the company’s paper, published on the open-access platform Arxiv, Atropos tested five large language models — including general-purpose models, a healthcare-specific model, and Atropos’ own ChatRWD beta — on 50 healthcare questions. Nine independent clinicians assessed the models’ responses based on relevance, reliability and actionability.
Results showed that general-purpose LLMs provided relevant information only 2% to 10% of the time. A healthcare-focused model performed slightly better at 24%. Atropos’ ChatRWD, which uses 160 million de-identified patient records, outperformed competitors by providing relevant insights 58% of the time.
The study also tested the models’ ability to answer novel questions. While most LLMs struggled, answering 0% to 9% of such queries, ChatRWD addressed 65% of them. These findings raise questions about the appropriate use of AI in healthcare settings and highlight the potential advantages of specialized models in critical fields like medicine.
AI Heart MRI Analysis Shows Promise
Researchers developed an AI model to analyze heart MRI scans in seconds. The study, published in European Radiology Experimental, demonstrated how AI could reduce the time and resources required for hospital heart image analysis.
The AI was trained on data from 814 patients across multiple NHS trusts and tested on 101 patients from a separate hospital, per a press release. The diverse dataset enhanced the model’s potential for widespread application.
“The AI model precisely determined the size and function of the heart’s chambers and demonstrated outcomes comparable to those acquired by doctors manually but much quicker,” Pankaj Garg, lead researcher from the University of East Anglia, said in the release. “Unlike a standard manual MRI analysis, which can take up to 45 minutes or more, the new AI model takes just a few seconds.”
Postgraduate researcher Hosamadin Assadi emphasized the broader implications.
“This innovation could lead to more efficient diagnoses, better treatment decisions, and ultimately, improved outcomes for patients with heart conditions,” Assadi said in the release.
While the results were promising, the researchers suggested further testing with larger, more diverse patient groups to validate the model’s effectiveness in various real-world scenarios.
AI Pioneer’s New Venture Targets Breakthrough in Drug Development
Jakob Uszkoreit, a figure behind the transformer architecture powering modern AI, shifted his focus to advancing drug development.
As co-founder of biotech startup Inceptive, Uszkoreit is applying generative AI to create more effective, biologically harmonious medicines, CNBC reported.
Uszkoreit, who left Google in 2021, was part of the team that published the seminal “Attention Is All You Need” paper in 2017, laying the groundwork for today’s AI boom.
“There are actually applications … where transformers have been deployed in production long before but to much, much less fanfare,” Uszkoreit said, per the report.
Last year, Inceptive secured $100 million in funding led by Andreessen Horowitz and Nvidia. The company aims to design RNA molecules using AI that can exhibit specific behaviors within biological systems.
“There’s actually this promise of a flavor of medicine that is in much greater harmony with living systems than most existing medicines,” Uszkoreit said, per CNBC, highlighting the potential for advancements in pharmaceutical research and development.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
