Martin Willemink, MD PhD FSCCT

Tennr just raised $101M Series C to tackle one of the ugliest bottlenecks in healthcare: referrals. Faxed forms. EMR silos. Manual intake. It's a mess, especially in imaging and specialty care, where 1 in 3 referrals fall through. Tennr’s approach? Fix the system without changing provider behavior. What they’re doing right: - Using a referral-specific language model trained on 100M+ documents - No EMR ripouts, no retraining, just automation layered onto existing workflows - Launched the Tennr Network, offering real-time referral visibility for radiologists and specialists - Already processed millions of referrals across hundreds of sites - Revenue has tripled since Series B Backed by a stacked list: a16z speedrun, IVP, GV (Google Ventures), Lightspeed, ICONIQ But here’s what I’m watching: - Can they maintain performance across fragmented EHR and fax-heavy environments? - Scaling orchestration logic across different payers isn’t trivial - At $600M+ valuation, expectations just went through the roof This isn’t just another referral platform. It’s infrastructure. And if they get it right, it quietly unlocks faster care and better outcomes across the entire imaging chain. One to watch. https://lnkd.in/gWM-kp4J

11

1 Comment

This is encouraging news for radiologists; however, the study may be somewhat unfair to the AI models, as it compares specialists with general-purpose AI systems rather than models specifically trained for radiology.

5

In this article, Eric Doherty discusses the growing complexity of #AI-based medical technologies and the regulatory frameworks trying to keep pace. He stresses that in 2025, "it’s not enough to prove performance at launch — manufacturers must demonstrate ongoing safety, monitor real-world performance, and manage transparency in updates and algorithms." A key point he raises: continuous evidence is the new compliance. To read the article, click here: https://lnkd.in/eTvku6qr

5

Day 2 at #MTC25 The MedTech Conference (5–8 October, San Diego). AI in commercial either moves the P&L — or it’s noise. In the last year we’ve: • Won a nine-figure public contract without cutting price; lifetime value now in the hundreds of millions. Single largest award win by any MedTech supplier in the industry. • Automated RFP/RFQ/Tender workflows → 80% faster cycles, +8% revenue, −22% cost. • Paired evidence synthesis and outcome development with ML pricing → 50% faster time-to-value (~8,600 hours back). This isn’t about tools; it’s about earlier, better decisions — and price defence when it counts. If you own contracting (RFx/Tenders), sales/BD, pricing, or market access, I’ll share the 15-minute playbook. I’m on-site — DM me for a live run-through today. If there’s no value in 5 minutes, we stop. #AIincommercial #CommercialisationOS #Dataledautomation #Starategicvaluecapture

13

Google DeepMind just released MedGemma last week, and when we saw its potential for medical image classification, our team at Qmed Asia jumped right in. 🔬 We fine-tuned Google’s new MedGemma model for detecting diabetic retinopathy (APTOS 2019 fundus images) via SFT, and the results are exciting. 🧠 Before fine-tuning -> ⚙️ After fine-tuning (SFT): Accuracy: 47.6% --> 78.2% Sensitivity (avg): 52.4% --> 40.5% Specificity (avg): 88.2% --> 87.5% ✅ Early-stage DR (Class 0) sensitivity jumped to 94.4% 📉 Trade-off: Lower sensitivity for Class 1–3, which we’re improving in the next round (e.g. better sampling, loss re-weighting) 💡 Key takeaway: MedGemma + local fine-tuning = fast-tracked DR AI screening potential, even with limited paired data. We are amazed at how 500 data points can improve the model by almost 30%! Thanks to Google Deepmind team on releasing the model. If you are interested on the model we FT, here is the HF link: https://lnkd.in/grmXRnR6 Let’s build medical AI that works for real-world clinics. Reach out if you're working on fundus, triaging, or imaging AI in healthcare! Kudos to LEE QI ZUN! #MedGemma #HealthcareAI #FundusAI #DiabeticRetinopathy #OpenSourceAI #SovereignAI

107

8 Comments

🚨 The dynamics reshaping healthcare technology appear to be on the verge of reaching dental. Within the next 6–12 months, I believe we’ll start to see similar shifts to those already happening in clearinghouses, where value migrates rapidly between players. In my view, this will begin to unfold more visibly in dental, likely on a larger scale and potentially with new entrants. Sam Altman has often spoken about a winner-take-all model in tech. That philosophy may matter here, because one of the earliest areas of impact could be X-Ray AI. In my opinion, this segment is particularly ripe for consolidation, moving from many vendors to a few dominant ones. OpenAI or XAI are possible contenders, simply because diagnostic imaging is low-hanging fruit for mass consumer adoption of AI in dental. For practices and patients, the product could look “free” under a bundled model, similar to how Amazon Prime works. For the broader ecosystem, the implications could be profound. OpenAI’s growth strategy seems primarily consumer-driven, with healthcare as a natural extension. To illustrate: if 100 million Americans subscribed to an “AI Healthcare Prime” at $20 per month, that alone would represent a $24B annual revenue stream from all healthcare verticals. The economics are staggering. In dentistry, as with the other verticals, the most natural entry point for this strategy is diagnosis and imaging. Once embedded, the downstream effects on workflows, vendor models, and practice economics could be significant. Importantly, this doesn’t require PMS integration, patients could simply photograph an X-ray from any imaging software with their iPhone. From there, the AI handles the rest. I’ve even seen early hints of this in my own offices. The bottom line: this kind of shift feels inevitable. In a winner-take-all landscape, no dental vendor can assume long-term safety. Venture backing offers limited protection when adoption is driven directly by consumers rather than B2B sales. To be clear, I may be wrong, but if I were in Mr. Altman’s position, this is the exact playbook I would consider. Recent hires in consumer and digital health tech at OpenAI, to me, suggest a possible direction of travel. Finally, there are ongoing discussions in the industry about how large AI players could eventually intersect with payers. While nothing is certain yet, that dynamic could add another layer of disruption. Time will tell. In these volatile times, the only real moat is speed. Everything else is noise. #OpenAI #nategross #ashleyalexander #grok #digitaltransformation #AI #RCM

10

Fine-tuning foundation models in medical imaging is now within reach for every clinician, researcher, and health system. On October 23rd, HOPPR and DeepHealth are hosting a live webinar to show just how seamless it is to adapt powerful medical-grade foundation models on HOPPR’s platform; aligning them with your own data, workflows, and clinical needs. We’ve already seen overwhelming demand for this session, and for good reason: this isn’t theory, it’s practical and transformative. Fine-tuning isn’t limited to big tech anymore, it’s something radiology departments, startups, and research groups can do today. If you want to understand where medical imaging AI is headed and how you can harness these models for your own practice or institution, you’ll want to be in this conversation. 🔗 Join us live on October 23rd: https://lnkd.in/g4wQ7BnB #AI #FoundationModels #MedicalImaging #HealthcareInnovation #Webinar #Radiology

19

Spencer Dorn-- being a medical director and additionally, having a role in system leadership, I have found the theory of constraints (and also the SCARF model by Rock) to be a critical approach when thinking about systems engineering or improvement. It is a nuts and bolts approach: 1-Identify: Find the system's biggest constraint (it is often not a focus, unless one really delves into a well constructed process map). 2- Exploit: Make the most of the constraint by optimizing its use (no idle time for the constraint as things need to continuously move through the bottle neck).  3-Subordinate: Align other processes to support the constraint (take every step and role in the process and determine how it or they can relieve the bottle neck).  4-Elevate: If the constraint still limits the process or system, invest in it to increase its capacity.  5-Repeat: Once the constraint is addressed, identify the next one and repeat the process. PDSA. I have been wondering how this will work with AI. In medicine, the constraint is usually the most specialized person or machine whose output cannot be duplicated by anything or anyone else in the process. Spencer Dorn, what aspect of the Theory of Constraints do you think may most be changed with the application of AI in healthcare? In general and from a patient's perspective requiring clinical care? I think access to care is one of the greatest issues for patients today. AI has the ability to increase access to care through #4 above. AI and evolving technologies (SMART, wearable, IoT) will increase access to care for patients in the future. Today, these technologies are probably just adding more patient data (and expectations) into a constrained system that cannot digest or meaningfully use the data. They are in a data tsunami. I hope with AI, there will be disruption in our healthcare system that forces use to reinvest and support our primary care infrastructure which in my opinion is where significant constraint and worsening constraint will continue to occur. Various models of AI should assist with data aggregation, analysis, interpretation, and documentation for primary care clinicians and teams to be better at pattern recognition, more precise and timely interventions, and with the ability to better track a patient's health. what are your thoughts? #UsingWhatWeHaveBetter

7

1 Comment

🧑‍💻 The Tech Sector WILL take over #healthcare research. 🏥 This post by Google Research talks about their work on reading wearable sensor data for use in AI. Imagine a world where your 📱🗣️ phone could motivate you to 'run faster to burn those calories' or 'slow down to recover'. That might not be as far away as you think. Google, Apple and the usual suspects in the #hardware game have easy access to #biometrics across many devices. They are able to bypass a lot of the restrictions most institutions like hospitals and clinics face. And until recently they have often worked in tandem, with Tech companies providing hardware for #health researchers to conduct studies with respect to ethics, dignity and proper scientific conduct. As LSM-2 and similar 'sensor data' models emerge, tech companies will become capable of reading a lot more about users than ever before. I don't mean to be a doomsayer, the story that tech companies kill #privacy has been told a million times already. No, my bigger concern is that Tech companies will more actively try to tap into the healthcare system through AI for biometrics just because they can! As recently as 2024, the Centers for Disease Control and Prevention estimated that about 160,000 heart attacks go undetected in the US every year. Now imagine if Google or a competitor announces that they can now detect their users' experiencing heart attacks and call for aid immediately. I believe we are on the cusp of a major paradigm shift in how acute ilnesses are handled, and more importantly by whom. Am I exaggerating? I would love to hear what other people think about this.

9

2 Comments