ColLooks like AI spots breast cancer before anyone knows there is breast cancer to spot. In a huge new study, AI looked at routine mammograms in screening clinics and caught patterns human eyes can miss three to six years before diagnosis, then nudged the scans into the pay-attention pile. Not replacing radiologists. More like handing them a flashlight, a second brain and fewer reasons to squint at shadows. This is a robot I’ll happily root for.
Signs of Breast Cancer Could Be Spotted 3-6 Years Before Diagnosis Using AI Screening, Shows Massive Study https://www.goodnewsnetwork.org/signs-of-breast-cancer-could-be-spotted-3-6-years-before-diagnosis-using-ai-screening/
EndicottRoad59@kibcol1049 A.I. should not be used for cancer diagnosis.
@EndicottAuthor As I understand it, this flags up early investigation by real medics, it doesn’t diagnose at all. It identifies possible cancer well in advance of current time scales, that means more people can be given treatment earlier, if real doctors also agree when the AI flags it up. What part of this is a bad thing?
doctorambientTo be clear, this isn't the same AI that generates useless chatbots. This is standard machine learning stuff that's been going on for decades in early diagnosis of cancer. Unfortunately everything's being rebranded with the same name now!
Alex@rtnVFRmedia Suffolk UK@doctorambient @EndicottAuthor @kibcol1049
its advanced pattern matching, same as "AI" cameras that are being used on the roads in both UK and AU to detect speeding, use of mobile phone when driving and even potential DUI - these things deliver valid results, and aren't going to go away even if the bottom falls out of the more frivolous and pointless uses of the tech..
Mensch, Marina@doctorambient This is why we should name LLMs. They are only a part of A.I. development @EndicottAuthor @kibcol1049
@energisch_ I agree with this completely! LLMs have gobbled up all the attention.
MycotropicWe did the same thing with lung nodules successfully. Machine learning that's based on exquisitely curated CT data can be very successful when it's used to identify patterns in new scans. LLMs are stupid and they're trained on primarily uncurated text, human expert curated targets for ML and DL can perform quite well.
@mycotropic @EndicottAuthor Brilliant stuff, thanks.