A growing paradox is emerging in medical AI: some of the strongest evidence exists for imaging-based AI tools—across colonoscopy, mammography, retinal imaging, CT scans, and pathology—yet many remain poorly integrated into routine clinical care.
Eric Topol argues that AI-assisted colonoscopy alone now has evidence from 44 randomized trials showing improved adenoma detection, while retinal AI models can predict risks for cardiovascular, neurologic, and metabolic diseases from routine eye exams. Despite this, implementation has lagged due to reimbursement gaps, workflow challenges, and lack of coordinated deployment.
At the same time, generative AI tools like ChatGPT are being rapidly adopted by both physicians and patients—even though real-world clinical evidence for diagnosis and treatment support remains limited. The article highlights a widening disconnect between where AI has the strongest validation and where healthcare is actually deploying it.