Matthew B Bloom MD, MSEE

6/ Our results highlight the immense promise of self-supervised learning to improve diagnosis of rare diseases, enabled by algorithms like CheXzero. At the same time, much work remains to translate these advances into real clinical impact. Excited to share this work! Major thanks to my amazing co-authors @agarwaln_econ, Ray Huang, @alex_moehring, @TobiasSalz, @feiyangkathyyu. Let us know your thoughts! Paper here: aeaweb.org/conference/202…

Hundreds of AI tools have been built to catch covid. None of them helped. The key is paying more attention to clean datasets. technologyreview.com/2021/07/30/103…

@EHechenbleikner @EricKnauerMD @EdLinEmory @EzraTeitelbaum @SAGES_Updates @p_g_j_ So glad the course was helpful! I love it when we gain a new convert!!

Before & After LCBDE today in OR with @EricKnauerMD. I ❤️ choledochoscopy!! First learned this at the SAGES CBDE course Aug 2019 at Cedar-Sinai. Hope more courses will be offered in the future!@EmorySurgery @EdLinEmory @EzraTeitelbaum @SAGES_Updates @p_g_j_

“Those of us in machine learning are really good at doing well on a test set,” -Andrew Ng X-Rays the AI Hype - IEEE Spectrum spectrum.ieee.org/view-from-the-…

I wish this story had distinguished training & testing. It's not an "open secret" that training data is error-prone, it's well known. Training is tolerant of noisy labels. A smaller test set with correct labels IS needed to assure AI works as intended. wired.com/story/foundati…

What Matters Most for Older Surgical Patients journals.lww.com/annalsofsurger…

My abstract was accepted by @SAGES_Updates for presentation at sages2021.surgery #SAGES2021 "Therapeutic Laparoscopy in Blunt Trauma"

@vineettiruvadi Have you considered the venerable Cushing’s forceps with teeth? A personal favorite, but not for everyone...

Are there 'Debakey with Adson teeth'? Seems to me zero physics-reasons for that to not exist #surgery #medtwitter

@vineettiruvadi @EricTopol @NatMachIntell @caromitreka @CamImaging @jhfrudd @Cambridge_Uni @EvisSala I disagree. Academic journals have recognized the need for authors to apply proper statistical methods and describe in detail their use. As outlined in this paper, most authors using AI algorithms on their datasets are not properly using data techniques against bias.Suboptimal!

@MatthewBloomMD @EricTopol @NatMachIntell @caromitreka @CamImaging @jhfrudd @Cambridge_Uni @EvisSala That's not really the issue. AI in medicine is still at the 'creative engineering' phase - you can't operationalize that. There is no 'proper AI methodology criteria' when we're still in the exploration phase. These are natural growing pains, and *it's not a bad thing*.

There've been > 300 #AI models, >2,000 studies for covid medical imaging (chest X-ray, CT) diagnosis.Systematically reviewed here: *"None of the models are of potential clinical use due to methodological flaws and/or underlying biases"* nature.com/articles/s4225… @NatMachIntell

In January, the FDA released their AI/ML Action Plan, which describes their "multi-pronged approach" to advance oversight of ML-enabled devices—with a focus on ensuring patient safety, algorithm transparency and real-world results. Learn more here: bit.ly/3s0885B

@Mike_TruittMD @traumadoctors @SWSCongress @WesternTrauma @TopKniFe_B @BKendallMD @docmartin22 @EAST_TRAUMA @elliotthaut The same reason I don’t chose any drink that’s served with a cocktail straw...

Two recent examples of why Texas tubes are better than pigtails. Two different pigtails. 2 diff patients. 2 failures. In the last 5 days. @traumadoctors @SWSCongress @WesternTrauma @TopKniFe_B @BKendallMD @docmartin22 @EAST_TRAUMA @elliotthaut #MedTwitter

Facebook Wants to Build a Smartwatch. news.softpedia.com/news/facebook-…

Really interesting results! Goes against the common belief that routine screening only diagnoses more DVTs but doesn't have a benefit.

Now available in the Google Fit app for Pixel phones: Mobile Device Heart Rate & Respiratory Rate Tracking! Learn more about how by using your phone's camera you can measure your heart rate and respiratory rate. Learn more: goo.gle/3btfwQZ

Google’s deep learning finds a critical path in AI chips | ZDNet zdnet.com/article/google…. AI chipsets >> graphics cards

We need novel ways to monitor patients at a distance. What hidden information is in the video feed? The doctor will Zoom you now technologyreview.com/2021/02/23/101…

Mark Barry, MD @ShibaniPati @UCSFSurgery identify that early cryoprecipitate after trauma and hemorrhagic shock protects endothelial barrier in vitro and reduces lung injury, inflammation, and vascular permeability in vivo in a murine model #fellowshipofthesnow #WTA2021

A @StanfordAIMI team takes on saliency maps: “We show that saliency maps are consistently worse than expert radiologists at localizing a variety of pathologies on chest X-rays, suggesting that deep learning explainability is far from ready for use in the clinical setting.”