The Next Medical Device Wave: Sensors, Robotics, Imaging, and Intelligent Software

A plastic disc about the size of a coin sits on the back of someone’s upper arm. It tracks glucose in the fluid just under the skin, and every 15 minutes a new reading and trend appear on a phone. For years, a device like that was something you got through a doctor. In March 2024, the U.S. Food and Drug Administration cleared the first over-the-counter continuous glucose monitor—one people could buy without a prescription.

It is a small object, but it points at a big shift. Medical devices are becoming smaller, more connected, more automated and increasingly driven by software. This field guide looks at four fronts—sensors, imaging, robotics and intelligent software—and, just as importantly, at what separates a clever gadget from a device that measurably helps patients.

First, what counts as a medical device?

In U.S. regulation, “medical device” covers an enormous range, from tongue depressors to implanted heart devices to software. The FDA sorts devices by risk. According to its guide on how to study and market a device:

• Class I devices present minimal potential for harm.
• Class II devices carry higher risk and need additional “special controls.”
• Class III devices sustain or support life, are implanted, or present potentially unreasonable risk, and generally need the most stringent review.

There are three main routes to market. A 510(k) shows a device is “substantially equivalent” to one already legally marketed. A De Novo request is for novel low-to-moderate-risk devices with no such predecessor. Premarket approval (PMA), for Class III devices, requires scientific evidence of reasonable assurance of safety and effectiveness. Note the vocabulary: most devices are cleared or authorized, and only PMA devices are formally approved. News stories often blur the distinction.

Sensors: the clinic moves onto the body

The over-the-counter glucose monitor shows both the promise and the careful boundaries of consumer sensing. The FDA said the system is intended for adults who do not use insulin, including people managing diabetes with oral medicines and people without diabetes who want to understand how diet and exercise affect their glucose. It reports readings every 15 minutes through a smartphone app. The agency also said it is not for people with problematic hypoglycemia (dangerously low blood sugar) and that users should talk to a health care provider before making medical decisions based on its output.

That last caveat is the crux of the whole sensor wave. A continuous stream of numbers is not the same as better health. The open questions are about interpretation: which readings should change what someone does, and how do people avoid acting on noise? More data can create more reassurance, more anxiety, or more unnecessary follow-up—sometimes all three.

Imaging: new physics, not just new software

Much of the excitement in imaging is about artificial intelligence, but some advances come from the hardware itself. In 2021, the FDA cleared the first photon-counting CT scanner. A CT (computed tomography) scanner builds cross-sectional images from X-rays. According to the manufacturer’s announcement of the clearance, conventional detectors first convert X-rays into visible light, while a photon-counting detector converts individual X-ray photons directly into electrical signals and counts them, preserving information about their energy. The company claims benefits including sharper images and the potential for lower radiation and contrast doses; those are the company’s claims, and independent studies across different clinical uses are the way to judge them.

MRI is going through its own reinvention, with faster scans and new reconstruction methods. Our piece on the MRI reinvention covers that story, and you can explore how scans are built in the MRI Explorer.

Robotics: from steady hands to decisions

Today’s widely used surgical robots are controlled by surgeons, who steer instruments from a console. The research frontier is more autonomy. In a study described by the National Institute of Biomedical Imaging and Bioengineering, a Johns Hopkins team’s Smart Tissue Autonomous Robot (STAR) performed laparoscopic intestinal anastomosis—sewing two ends of intestine together—in pigs with minimal human help. The researchers reported that STAR was more consistent in suture spacing and depth than expert surgeons in their tests.

It is worth being precise about what that is: a preclinical study in synthetic tissue and a small number of animals, not a treatment approved for people. Soft tissue moves, bleeds and varies between patients, which is exactly why autonomy there is hard, and why the path to clinical use will require extensive human studies and regulatory review. There is also a design question that engineering alone cannot settle: when a machine makes more of the decisions, who is responsible when something goes wrong, and how should a surgeon supervise it?

The next medical device wave will not be judged by how impressive the machine looks, but by whether patients who use it do better than patients who don’t.

Intelligent software: the device you cannot touch

A growing share of medical devices are partly or wholly software. The FDA maintains a public list of AI-enabled medical devices authorized for marketing in the U.S. Radiology accounts for the large majority of entries, which fits the field’s long history with digital images. The FDA notes the list is not comprehensive; it identifies devices mainly by AI-related terms in their authorization summaries.

AI raises a regulatory puzzle: software can be retrained and improved, but a device is usually authorized as a fixed product. In December 2024, the FDA finalized guidance on predetermined change control plans (PCCPs), which let manufacturers describe planned modifications to AI-enabled software—and how they will develop, validate and implement them—up front, so that changes within that plan do not each require a new marketing submission.

For a closer look at where AI helps in cancer screening, and where false alarms matter, see Can AI Spot Cancer Earlier?

What could slow the wave

• Outcome evidence: many devices are authorized on technical performance or equivalence, not on proof that patients live longer or feel better.
• False alarms: more sensitive sensors and algorithms can flag more things that turn out to be nothing, driving extra tests.
• Generalization: software trained in one hospital or population may perform differently elsewhere.
• Workflow: a tool clinicians do not trust, or that adds clicks, may go unused.
• Security and privacy: connected devices carry sensitive data and need protecting.
• Payment: a device can be authorized and still struggle if insurers will not pay for it.

How to read a device headline

1. Was the device cleared, authorized through De Novo, or approved via PMA—or is it still in research?
2. Was it tested in people, animals or simulations?
3. Does the evidence show better accuracy, or better patient outcomes?
4. Who is it for, and who is it explicitly not for?
5. If it uses AI, how was it validated, and on whom?

Devices may be heading toward something even more ambitious: combining sensor streams, images and software models into a personal simulation. We explore that idea in What If Every Patient Had a Digital Twin?