Medical AI shows paradox of underused image tools and widely adopted unproven LLMs
Updated
Updated · erictopol.substack.com · May 3
Medical AI shows paradox of underused image tools and widely adopted unproven LLMs
11 articles · Updated · erictopol.substack.com · May 3
The review says 44 colonoscopy trials, a mammogram trial of more than 100,000 women and new retinal models have not translated into routine care, while 72% of surveyed US doctors use generative AI.
It attributes the gap to weak implementation, reimbursement and orchestration for validated imaging AI, even as about 40 million US adults use chatbots daily despite limited real-world evidence for diagnosis or treatment decisions.
The article argues proven AI for scans, retina images and colonoscopy should be deployed now, while LLMs for clinical decision-making need prospective trials, safety monitoring and outcome data before broad adoption.
Why does medicine adopt unproven AI for paperwork over validated AI that finds cancer years earlier?
Is your doctor's AI assistant using your private health data without your consent?
When medical AI makes a fatal error, who is legally responsible: the doctor, the hospital, or its developer?
Healthcare AI Adoption Surges 2.2x Faster Than Economy (2025–2026) Amid Imaging AI Integration Challenges
Overview
Between 2025 and 2026, healthcare AI adoption surged over twice as fast as the overall economy, driven by rising global healthcare costs and high administrative burdens. Large language models (LLMs) rapidly integrated into clinical workflows, fueled by strong consumer demand, technical maturity, and significant venture funding. In contrast, specialized medical imaging AI faced slower adoption due to complex regulations, high operational costs, and lack of reimbursement, especially impacting rural and resource-limited areas. This created a paradox of fast LLM uptake versus lagging imaging AI. Addressing these barriers through hybrid AI models, reimbursement reforms, and equity-focused investments is essential to unlock AI's full potential for improving diagnostics and healthcare access worldwide.