Ammar Sabilarrohman
For decades, the standard for progress in medicine was defined by physical innovation: the development of a more potent pharmaceutical, the refinement of a surgical technique, or the installation of a massive, state-of-the-art diagnostic machine. Today, however, the most significant "medical miracles" are emerging from an entirely different realm. They are being born from digits, patterns detected in vast seas of data, and algorithms operating in an abstract space that no surgeon’s hands can reach.
At the center of this shift is Aidoc, a company that has successfully moved artificial intelligence from the realm of academic theory to the front lines of clinical practice. With 32 FDA clearances—the most in its category—and a footprint spanning over 1,600 hospitals worldwide, Aidoc is currently analyzing more than 70 million patient cases annually.
In a recent episode of KFF’s Business of Health, host Chip Kahn sat down with Aidoc co-founder and CEO Elad Walach to discuss the evolution of clinical AI, the hurdles of regulatory adoption, and the future of a healthcare system where AI acts as a ubiquitous, silent guardian.
The Genesis: From National Security to Healthcare
The trajectory of Aidoc is rooted in a fundamental lesson Walach learned during his tenure leading an AI division for national security in Israel: "Cutting-edge algorithms mean nothing if they are not actionable."
Walach’s transition to healthcare was driven by both professional curiosity and personal history. He describes a "healthcare bug" in his own life, influenced by his father, who worked in IBM research and pushed for early AI integration in medicine. The impetus to formalize this mission came after his family experienced the trauma of a preventable diagnostic error.
"When I finished my service, my co-founders and I spent about a year to a year and a half just sitting in hospitals," Walach recalls. "We didn’t know much about healthcare. We spent that time observing the recurring problems: the lack of access and the reality that amazing clinicians were barely holding on in an overwhelmed system."
This period of observation led to the founding of Aidoc in 2016, a moment Walach identifies as the advent of "deep learning" in medicine—the first time AI became accurate enough to perform at a "physician-grade" level.
Chronology: The Road to Widespread Deployment
The journey from a small startup to an enterprise platform running in 1,600 hospitals was not achieved by simply pitching a product, but by establishing a foundation of trust.
- 2016: Aidoc is founded with a focus on solving the "intractable problems" of medical imaging, starting with the detection of brain hemorrhages.
- The Early Years: The team navigated the regulatory landscape, learning that the FDA is not a "bureaucratic wall," but a partner in patient safety. With over 30 submissions, Aidoc helped define how the FDA reviews and validates clinical AI.
- The Scaling Phase: By focusing on "clinical-grade" accuracy, Aidoc moved from single-use models to a multi-pathway platform.
- Present Day: Aidoc has transitioned from point solutions to enterprise-wide integration, allowing health systems like WellSpan to expand from six to over 20 care pathways in a matter of months.
"The key to adoption was never just the model-building," Walach explains. "It was understanding that in healthcare, you don’t partner on product—you partner on vision, roadmap, and the ability to execute."
Supporting Data: The Cost of Inaction
The urgency for this technology is underscored by sobering statistics regarding diagnostic error and healthcare efficiency. A study from Johns Hopkins, cited by Walach, suggests that diagnostic errors and delays contribute to approximately 400,000 deaths annually in the United States.
Furthermore, the operational burden is compounding. Recent data indicates that between 2022 and 2023, wait times for outpatient imaging diagnoses doubled. "We threw labor at the problem for years," Walach notes. "But the problem keeps growing. The only way out is technology."
Aidoc’s platform functions as a "second set of eyes." In an emergency department, when a patient undergoes a CT scan, the AI analyzes the images as they are being generated. If it detects a critical condition—such as a pulmonary embolism, aortic dissection, or brain bleed—it automatically flags the case for the radiologist, ensuring the most urgent patients are seen first. This shift from a "first-come, first-served" queue to a "triage-by-acuity" system can be the difference between a successful intervention and a catastrophic outcome.
Official Responses and Operational Challenges
The integration of AI into hospital workflows brings two major challenges to the forefront: alert fatigue and data drift.
Managing Alert Fatigue
"Alert fatigue" occurs when clinicians are bombarded with too many notifications, leading them to ignore important findings. Walach argues that the solution is not just better software, but higher accuracy. By building a "foundation model"—a sophisticated AI capable of identifying 100+ diseases simultaneously with 99.7% specificity—Aidoc minimizes false positives. "When you run 20 models at once, you’re going to false-alarm the physicians to death unless your accuracy is near-perfect," he says.
The Problem of Data Drift
Data drift refers to the phenomenon where AI models lose accuracy over time as the clinical environment changes—whether through new scanning hardware, updated protocols, or organizational changes. Walach admits that even for the market leader, accuracy can drift by 10% every 18 months. Consequently, hospitals must maintain a "human in the loop" approach, employing "AI babysitters" or data specialists to monitor and calibrate the system continuously.
Implications: The Future of Clinical Practice
The long-term implication of this technology is a shift toward "proactive" rather than "reactive" care.
From Reactive to Proactive
Walach points to the example of Mercy Hospital in St. Louis, which used AI to identify "calcium scores"—a key predictor of heart disease—in patients undergoing routine chest CT scans for unrelated injuries. By identifying these high-risk patients who were previously "unmanaged," the health system could proactively reach out to them. This changes the role of the hospital from a place of acute intervention to a partner in long-term wellness.
The Role of the Physician
Contrary to the fears of some, Walach does not envision a future where AI replaces the radiologist. "I don’t believe in the ‘we’re going to replace the radiologist’ paradigm," he asserts. Instead, he sees AI as a tool for collaboration. By automating the identification of routine or "bottom of the license" work, physicians are freed to focus their expertise on complex cases that require human judgment.
The "Seatbelt" Analogy
Looking toward 2026 and beyond, Walach envisions a world where AI-assisted diagnosis is as common as wearing a seatbelt. "I think it will be a standard of care," he says. "No diagnostic encounter should happen without this AI layer supporting it."
While the integration with Electronic Health Record (EHR) systems like Epic and Oracle remains a complex technical hurdle, the industry is trending toward openness. The ultimate goal is a seamless, invisible layer of intelligence that supports clinicians at the point of care, ensuring that every patient benefits from the most accurate and timely diagnostic insights available.
As Chip Kahn concluded, the transition to this AI-driven future is not "unsettling"—it is, in every sense, the next great leap forward in the evolution of modern medicine.
Ali Ikhwan
Asep Darmawan
