From 2024 to 2025, doctors at Yale Medicine reduced the time it took to evaluate patients with abdominal aortic aneurysms by 73%.
This improvement, for a condition where fast treatment saves lives, wasn’t down to employing more doctors or nurses, or a new breakthrough drug discovery. It was thanks to the Yale team using an AI tool that analyzes images from patients’ scans, combined with other medical records, to support clinical decision-making.
“It's another pair of eyes that reduces a chance of an overlooked finding, that's the feeling from the clinicians,” said Tom Valent, chief business officer at Aidoc, the Tel Aviv-based AI medical imaging company that built the platform.
At a time when global health systems are under increasing pressure, due to aging populations, rising rates of disease and shortages of skilled healthcare workers, AI can help doctors make sense of complex medical images more quickly and consistently. It’s also being used to bring medical imaging into areas of healthcare that have remained stubbornly analog, making treatments cheaper and more convenient for patients.
The hospital that doesn't talk to itself — yet
Aidoc’s platform can be used to analyze medical images from scans across different areas of medicine, ranging from cardiology to neurology and radiology.
One of the big problems that AI can solve in healthcare is joining up different parts of the system which don’t normally interact, Valent explained. If Aidoc’s vision models are given a new patient scan to analyze, the platform combines that data with information from electronic medical records, which can give doctors important context they might otherwise not have had.
“Oftentimes in big hospitals, there's a coordination issue between different specialists,” Valent told The Infinite Loop. “We basically make care in the health system higher quality, more proactive, and more connected between different specialties.”
Aidoc’s platform analyzes medical images to speed up triaging. Credit: Aidoc
Aidoc says more than 1,600 hospitals are currently using its technology, and that the platform can help spot urgent cases that might be at the back of a queue, before a busy doctor has a chance to look at a scan. In medical areas like radiology that can make a big difference if a cancer patient is assessed faster.
“In hospitals with big backlogs and long turnaround times, the radiologist knows that the chance of a case at the bottom of their queue having something urgent, and not being seen in a time period that’s potentially critical, is much lower with Aidoc,” said Valent.
Using AI models to make decisions on who might receive medical care fastest raises concerns over whether algorithmic bias might mean that certain demographics of patients receive care before others. This is something that Aidoc invests heavily in through its AI safety team, and is a crucial step in receiving regulatory approval.
“We have more than 60 people that are focused on AI safety, both in the development stages and in the post-market stages, when the algorithm is already running in production,” said Idan Bassu, chief R&D and AI officer at Aidoc.
Half the cost, ten times the patients
AI is also bringing medical imaging into areas of healthcare that have previously relied on extremely manual workflows.
Barcelona-based Impress uses AI to help orthodontists create treatment plans, based on scans of their teeth that are analyzed by a vision model trained on more than 500,000 patient cases. Traditionally, orthodontists will look at a patient’s teeth in person, requiring people to make regular trips to a clinic to assess progress.
Impress uses data from an initial in-clinic scan, combined with proprietary AI, to predict the outcome of a treatment. Credit: Impress
IImpress reduces these visits by letting patients take scans of their teeth at home with a smartphone, with the visual data then analyzed by AI that recommends the next steps in the treatment plan. The company says this allows doctors to treat 10 times more patients in the same amount of time, reducing the cost of Invisalign-style treatments to around €3,000, compared to around €6,000 at traditional clinics.
An Impress patient taking an at-home scan to monitor treatment progress. Credit: Impress
Impress CEO and co-founder Vlad Lupenko explained how every patient case is overseen by a doctor, and that AI shouldn’t be used to write clinical oversight out of the process. The risks of AI being used to take decision-making out of doctors’ hands were laid bare in 2020, when a chatbot made by the now bankrupt healthtech company Babylon missed symptoms of a heart attack in a patient.
“AI's an additional tool that helps to increase the efficiency of doctors’ time and their accuracy, but it will never replace the doctor,” Lupenko said.
By combining imaging and AI technology with clinical expertise, to make treatments cheaper and more convenient, Impress, which describes itself as Europe’s biggest provider of clear aligner treatments, has grown to treat nearly 50,000 patients every year.
Compute where the patient is
Impress's platform analyzes more than 180,000 scans every week. Aidoc's system processes 60 million patient cases per year. At that scale, the demands on cloud infrastructure are significant — and in healthcare, they come with constraints that don't apply elsewhere.
“Flexibility is critical. We need to be able to scale GPU and compute capacity dynamically,” Impress CTO Yerzhan Tashbenbetov told The Infinite Loop.
On top of regulation-compliant infrastructure, Aidoc adds that it needs cloud compute that’s geographically close to hospitals, to ensure fast speeds and support timely clinical decisions.
“One of the reasons for multi-regional compute is latency. We want the cloud hardware to sit as closely as possible to the customer,” said Idan Bassu, chief R&D and AI officer at Aidoc, adding that the huge quantities of data that Aidoc processes demand cutting-edge efficiency architectures.
“You have the ingestion of a very high volume of data you need to handle. You need a cloud environment that can process hundreds of millions of events from both scans and medical records every day. We need very, very efficient data streams.”
Aidoc and Impress are solving different problems with the same underlying bet: that AI trained on enough images, running close enough to the patient, can do something no doctor working alone can.
