Calibrated AI models improve the accuracy of identifying and quantifying the risk of hypertrophic cardiomyopathy (HCM)
Let’s talk about precision in healthcare. Not “close enough”, but actual, actionable accuracy. That’s becoming the new standard, and AI calibration is powering it. One recent example is what Mount Sinai and Viz.ai pulled off with an algorithm called Viz HCM. It’s FDA-approved to detect a cardiac disease called hypertrophic cardiomyopathy, HCM for short. This is a condition where the heart muscle thickens and limits blood flow. Often it’s genetic, and finding it early can change lives.
Now, what’s new here isn’t just detecting HCM. It’s the algorithm’s added ability to deliver real probability scores. Not vague labels like “high risk,” but something better: a 60% likelihood you’ve got HCM. That level of transparency gives doctors clarity, and patients more confidence in their care choices. The team validated these scores by comparing the algorithm’s predictions against nearly 71,000 real patient ECGs gathered between March 2023 and January 2024. It flagged 1,522 potential cases. Every one of those was cross-checked with imaging and clinical records. The result? The calibrated model was on target.
It’s deployed and doing the job. Dr. Joshua Lampert, Director of Machine Learning at Mount Sinai Fuster Heart Hospital, explained it clearly: “You have about a 60 percent chance of having HCM”, not maybe, not possibly, but a number grounded in data and designed for decision-making. When it comes to clinical diagnostics, that’s a game changer.
For C-suite leaders in health tech or any company shaping how we interact with data: this model shows it’s not enough just to identify risk. Quantifying it means clinicians can prioritize time and resources based on patient-specific probability, not generalized suspicion. That’s not just a medical win. It’s a smarter, leaner approach to healthcare delivery. Whether you’re running a health system or scaling a medical AI product, this is the kind of outcome that justifies the investment.
Calibrated AI supports effective integration into real-world healthcare systems
High-performing tech isn’t enough if it doesn’t integrate into real operations. In healthcare, that means AI has to align with clinicians’ workflows, not disrupt them. The recent study using the Viz HCM algorithm at Mount Sinai proves this can be done, and done well. By calibrating the AI’s output to reflect actual disease probability, the system added immediate value to clinical decision-making. This goes beyond scanning ECGs for signs of hypertrophic cardiomyopathy. It optimizes how hospitals respond, who they prioritize, and how follow-ups are scheduled.
The key here is implementation. That’s usually the toughest part, getting proven technology into a clinical environment without slowing things down. The Viz HCM study shows it’s possible. A calibrated output, like “there’s a 60% chance this patient has HCM”—can be plugged into existing systems. Clinicians don’t need to interpret black-box scores or vague probability bands. They see risk clearly quantified and can act immediately. Patients with actual risk are brought to the front of the line, and unnecessary testing is reduced downstream.
According to Dr. Girish N. Nadkarni, Chair of the Windreich Department of Artificial Intelligence and Human Health and Director at the Hasso Plattner Institute for Digital Health at Mount Sinai, this setup works because it doesn’t chase performance benchmarks for the sake of it. It sticks to a clear outcome: supporting decision-making while improving care delivery. Dr. Nadkarni put it simply: “It’s not just about building a high-performing algorithm, it’s about making sure it supports clinical decision-making in a way that improves patient outcomes and aligns with how care is actually delivered.”
For executives assessing AI deployment across other sectors, this is a solid reference point. It reminds us that the real challenge isn’t just making AI work, it’s making AI usable. And when calibration turns ambiguous outputs into real, clinically relevant decisions, that’s where the value scales. The lesson’s clear you build systems that work where real-world decisions are being made.
AI tools continue to expand their role in cardiovascular risk screening
AI in healthcare is scaling, fast. And its application in cardiovascular screening is pushing the boundary even further. We’ve already seen success calibrating models to assess hypertrophic cardiomyopathy risk. But the field is growing. A separate research initiative earlier this month showed an AI model capable of identifying heart failure risk using single-lead electrocardiograms. It worked under high-noise conditions and was still accurate. That’s a big deal if you’re thinking about screening at scale.
This model was tested across multiple patient groups, including a sizeable cohort from the Yale New Haven Health System. Over 192,000 patients were evaluated, and the model flagged 22.2% as positive for elevated heart failure risk. Across a median follow-up of 4.6 years, 3,697 patients (1.9%) developed heart failure. That jump, moving from broad population-level screening to specific predictive insight, gives healthcare teams a clear advantage in early detection and resource planning.
Here’s what matters for leaders: this isn’t experimental. This is real-world validation. The evidence supports single-lead ECG as a viable, scalable input for AI-driven screening tools aimed at serious health outcomes. Minimal equipment, widely available data, and effective prediction, it checks all the boxes for scalability and cost-efficiency without sacrificing impact.
If you’re an executive leading health systems, digital health tools, or enterprise-scale diagnostics, this momentum signals strong ROI potential. It shows that strategic application of AI goes beyond hospital walls. These technologies can now underpin neighborhood-level screening programs, corporate health initiatives, even long-term insurance forecasting. It’s a practical and forward-looking investment that aligns with how healthcare systems are evolving, driven by data, not guesswork.
Key takeaways for decision-makers
- Calibrated AI boosts diagnostic precision: Leaders in health systems should adopt calibrated AI tools, like Viz HCM, that move beyond flagging risk to quantifying it, enabling more accurate and actionable clinical decisions.
- Integration is the real differentiator: High-performing AI models must deliver practical value by fitting cleanly into existing workflows, helping clinicians act faster and prioritize care based on real, individualized probabilities.
- Scalable AI screening is here: Executives should invest in AI-driven diagnostic tools that validate at scale, single-lead ECG models are already predicting heart failure risk accurately across large populations, making broad-screening programs both feasible and effective.
