January 8, 2026
When the robot blinks, Reddit explodes
AI Misses Nearly One-Third of Breast Cancers, Study Finds
AI misses 1/3? 'It found 2/3!'—stat wars erupt as doctors call for assist
TLDR: An AI tool missed around 30% of cancers in a study of patients who already had cancer, while a quick MRI method rescued about 80% of those misses. Commenters argued over stats, demanded better context, and pushed for AI to assist doctors, not replace them—layered screening matters.
The headline says AI missed almost one‑third of breast cancers, but the comments instantly turned it into a stat war. One camp cheered, “Hey, that means it found nearly two‑thirds!” while the skeptics demanded real numbers, not vibes. emil‑lp dropped a nerdy zinger—“I can detect 100% by ‘Return True’”—to drag any single-number hype, calling for the full breakdown of hits and misses, not just one shiny stat. Meanwhile, levocardia fact‑checked the whole thing: this was a retrospective case series on women already diagnosed with cancer, not a broad screening study, which means no true negatives and no classic “confusion matrix” to argue over. Translation: great for testing AI on known tumors, not proof it’ll catch everything in the real world.
Here’s the tea: in 414 women already known to have cancer, the AI system missed about 30.7%, especially in dense breast tissue and tiny tumors. But a fast MRI trick called diffusion‑weighted imaging (DWI)—it looks at how water moves in tissue, no contrast dye—rescued roughly 80% of those misses. Docdeek did the back‑of‑the‑napkin math: you still end up with a chunk left uncaught. Moosturm chimed in with the crowd‑pleasing stance: keep AI as an assistant, not the boss. The vibe? Fierce debate, sharp jokes, and a reminder that smarter screening might mean stacking tools, not worshipping one. Read the original study here
Key Points
- •In a cohort of 414 women with confirmed breast cancer, an AI-CAD system missed 127 cancers (30.7%).
- •Missed cancers were strongly associated with dense breast tissue and smaller tumor size (≤2 cm nearly five times more likely to be missed).
- •Radiologist review of diffusion-weighted MRI alone detected 83.5% and 79.5% of AI-missed lesions for two readers, respectively.
- •DWI performed best for tumors >1 cm and for cancers not visible on mammography, but accuracy declined for lesions <1 cm.
- •Study limitations include a single-institution, retrospective design with only cancer-positive cases; authors call for prospective multicenter trials.