Speaker 0:  imagine a rare disease that silently steals a decade of your life, but human doctors routinely miss it for ten whole years. researcher just built a privacy first ai that completely outsmarts medical specialists using nothing but two quick smartphone photos of your hands. so hand photos can outsmart specialist doctors now. cobalt university researchers trained a deep learning model to detect acromegaly, a life shortening hormone disorder by analyzing photos of the back of the hand and clenched fist. the system bypasses facial and palm recognition entirely, delivering an elite diagnostic triage tool without compromising user anonymity. this proves medical ai can perform world class screening using completely non identifiable, low cost, morphological data from a standard smartphone drastically reducing the typical ten year diagnostic delay. what do you think about this, john?

Speaker 1:  i love this. i always love new unique use cases for ai especially for disease detection. and so for those of you that don't know, acromegaly, occurs when the body produces excessive amounts of growth hormone, which in turn increases levels of insulin like igf one. and so unlike gigantism, which, be before growth plates close during childhood, acromegaly develops in adulthood. so since bones can no longer lengthen, the disease causes thickening and enlarging of existing structures. so common manifestations include enlarged hands, feet, jaw, nose, tongue, and internal organs. and so one of the biggest clinical challenges is that symptoms emerge very gradually. so patients may simply notice they need larger rings, gloves, or shoes over many years. we're not talking months or weeks. right? so because these changes occur slowly, both patients and physicians often fail to recognize the disease until significant damage has already occurred. and so we're talking about, diagnostic delays exceeding 10. that's not uncommon. and so, if you're wondering why researchers focused on hands, it's because most previous ai systems for acromegaly detection relied on facial photographs, like looking at the face. so while effective facial recognition, it creates significant privacy concerns, it also limits patient participation. so they train their model using only photographs of the back of the hand and a clenched fist. so they deliberately excluded palm images because palm prints contain highly identical biometric information. so the privacy conscious design made it easier to collect large scale patient data while maintaining confidentiality, which was cool. so the one reason this study stands out is the quality of the training dataset. and i love this because nowadays, the training datasets are getting less and less quality when it comes to training, and that's why we're having issues with a lot of these things. so the researchers themselves collected more than 11,000 hand images, seven from 725 participants in 15 different medical centers across japan, which is a great, great dataset to use, which it's which is part of kind of a larger trend, which i love. so this is great. this is exciting. what do you think about this, craig? do you think this is, there are other potentials for this?

Speaker 0:  well, there's definitely other potentials for this type of technology. and i think now that we have, tools like ai, where every day we're seeing new uses where it's finding, it's mapping our genomes faster than ever before. we're finding defects and way to prove stuff. it's finding new proteins, new treatments, potential treatments for things. doing things that, we couldn't do by ourselves, and it's amazing. and i love where we're going with this. and the fact that, you know, it could take up to ten years to determine this before, and now we can do it so quickly, at you know, with four images apparently. it's a the thought is amazing. like, if you can determine that quickly, send it in, and be done with it, the idea is astounding. think of this update like a master mechanic who can tell your engine's internal timing belt is about to snap just by looking at the subtle uneven wear on the outside of your tires. you don't need to tear down the entire engine to run a full expensive computer diagnostic yet. the ai flags hidden structural damages from just two quick external snapshots. so this is a freshly published clinical study currently moving towards, regional clinical pilot deployments and integration into routine public health checkup. so this, is designed to be a low cost software implementation integrated in public screening infrastructure requiring no special medical hardware. so it'll be easy to implement into, standard checkups and that kind of stuff. any standard digital or smartphone camera paired with an internet connection for cloud based, deep learning inference and zero privacy anxiety. you're not giving away any of your information, facial, face, or fingerprint data. so, this technology is really cool. it, works faster than we've ever thought. the resnet 50 architecture driving this model utilizes a skip connections or residual learning. this mathematical workaround solve the infamous vanishing gradient problem in deep neural networks, which is the exact reason it can map hyper subtle bone density and tissue shifts without getting lost in the data noise.

Speaker 1:  yeah. and the thing is, again, we talked about this. it's not about acromegaly. it's about the growing ability of ai systems to identify diseases from unexpected sources of information. we haven't covered most of these, but we should. so there's the there's some ones i've got listed here. so they were able to use ai to detect heart disease from retinal scans, parkinson's from voice patterns. that kinda makes sense. covid nineteen from ct images, diabetes from eye photographs, and, acromegaly from facial images, which we just talked about. so i love that they're taking a, an outside the box kind of approach to detecting diseases, which is kinda cool.

Speaker 0:  yeah. there's a full article over here on that covers all of this and what other things we it could find. early screening tools in industry are an industry gold mine for health care systems. the primary treatment endoscopic transfenodial surgery to remove, pitular tumors boasts an immensely higher success rate at a fraction of the long term cost it performed before permanent bone and organ enlargement sets in. so while the ai focus on hands, historical clinic history shows that acromegaly, is hoft and retroactively spotted by looking at teeth. the excess growth, hormone widens the jaw creating sudden distinct spacing between lower teeth, a symptom that went unflagged for decades before modern imaging. mhmm. so this is the monumental win for noninvasive privacy first diagnostic tech that could completely eliminate a decade worth of medical waiting times. would you trust an ai smartphone scan to audit your body for hidden diseases, or does medical ai still make you nervous? let us know in the chat.