Today, Schmidt Sciences is excited to announce 28 new AI2050 Fellows tackling the defining AI challenges of our time, from understanding its risks to expanding what it can help us discover. schmidtsciences.org/ai2050-f…

Apparently, yesterday @midjourney pivoted from AI image generation to...whole body ultrasound , presumably AI-augmented. I spent some time trying to find hard data and did not come up with much beyond the video. Some thoughts based on the X reactions today. 1) "Nobody's ever done this before." This seems to be a variant on ultrasound tomography, with Butterfly sensors arranged in rings. Ultrasound tomography is not new, with commercial systems available for breast imaging. The system proposed here seems very similar to Garrett et al 2024 (pmc.ncbi.nlm.nih.gov/article…) which provided fuzzy images of the abdomen and extremity. By collating sound wave return from the ring array the system can attempt to minimize artifact from bone and gas. So, we have known this is possible since 2024. 2) "So easy for the patient." The system requires the patient to submerge themselves in water. For commercial breast UT, this simply requires laying on top of a shaped tank, but not going fully in water. As some people have difficulty lying on a DEXA or CT scanner (the latter of which can image in seconds) I will be interested to see how this is received by patients. 3) "This will revolutionize medical imaging." Ultrasound is limited by bone, air (lung, stomach, bowel) and depth of penetration. In the abdomen, this system can "see" around bowel and spine by adding together the full set of images. There is no workaround for the head or lungs and you'll notice that's why they didn't offer any pictures of those areas. It also means it won't be great at screening inside the stomach, intestine, or colon. I also note the volunteers appear fairly skinny. Ultrasound is always more limited in heavier patients. 4) "But AI can fix it!"...not really. Current DL-based reconstruction techniques require at least some undersampling of a region in order to reconstruct the image. You'll get a lovely picture of the outside of the skull. I can AI-upsample a fuzzy photograph but that doesn't mean what comes out of it actually existed. We have a variant of this issue already with MRI DL reconstruction. 5) "But this is better than MRI in the 1970s!" Yes, true. But the competition is not 1970s MRI, it's modern CT/MRI/US and most especially low-field MRI. For brain, for example, low-field MRI is already diagnostic quality and doesn't need shielding. A low-field scanner costs 50k and can be used in an ICU or put in a van. Why would I send a patient to get an experimental full body US when there's whole body MRI available that's already diagnostic quality? science.org/doi/abs/10.1126/… 6) "The FDA has no idea how to regulate this." This one made me laugh. There's an entire set pathway for this. Commercial ultrasound tomography already exists as an easy reference of a similar technology in the application. If they haven't submitted to the FDA, it's because they plan to try at a later date, or because they're not planning on submitting it at all. Do I think this is new and exciting? Yes. It looks like it's going to be great for body composition, and I do think there will be improvement in the future. Is it currently medical-grade diagnostic quality? No, not based on what they showed us. Apparently in-person there was a great hand demo. I don't see why that would be an improvement over routine US or MRI in visualizing hand soft tissues. To quote @khakrish: "It feels like all the same problems as full body MRI with the added problem of an unproven imaging modality and no FDA clearance."

The Hallmarks of Cancer brought clarity through abstraction - but cancer is more complex than we can reduce. honored to build on their legacy in our @CellCellPress perspective on generative models as unified, multimodal systems for cancer discovery 📄 msft.it/6013QhkUS !