AI-Powered RADiCAIT Makes Medical Imaging Affordable at TechCrunch Disrupt 2025
▼ Summary
– PET scans are medically valuable for cancer detection but involve lengthy fasting, radioactive injections, and post-scan isolation from vulnerable groups.
– Access to PET scans is limited in rural areas because scanners require nearby cyclotrons to produce short-lived radioactive tracers.
– RADiCAIT uses AI to convert accessible CT scans into synthetic PET scans, aiming to replace PET for diagnostics with a more affordable and simpler solution.
– The startup’s AI model learns by mapping CT and PET scan patterns, generating images statistically similar to real PET scans, as shown in clinical trials.
– RADiCAIT is conducting clinical pilots for lung cancer and pursuing FDA trials, with plans to expand to other cancers and radiology domains.
Transforming medical diagnostics, RADiCAIT’s breakthrough AI platform converts widely available CT scans into synthetic PET images, potentially eliminating the need for complex, expensive PET scanning procedures while expanding access to critical cancer detection tools.
Anyone who has undergone a PET scan understands its demanding nature. These scans play a vital role in identifying cancer and monitoring its progression, yet the procedure itself presents significant challenges for patients. It begins with a mandatory fast lasting four to six hours before the hospital visit. For individuals in rural communities, the situation is even more difficult, as local hospitals often lack the necessary PET equipment. Upon arrival, patients receive an injection of a radioactive tracer. They must then wait approximately one hour for the substance to circulate through their system before entering the scanner. The imaging process requires lying completely still for about thirty minutes. Afterwards, due to residual radioactivity, patients must avoid close contact with children, elderly individuals, and pregnant women for up to twelve hours.
A further complication involves the geographic limitations of PET technology. These scanners are predominantly located in large urban centers. This is because the radioactive tracers they require have an extremely short lifespan and must be manufactured in nearby cyclotron facilities, making widespread deployment in regional or rural hospitals practically impossible.
RADiCAIT, a startup originating from Oxford University, proposes a novel solution. The company’s foundational AI model can generate a synthetic PET scan from a standard CT scan, a far more common and economical imaging method. Recently emerging from stealth mode with $1.7 million in pre-seed funding, the Boston-based firm is now a Top 20 finalist in the Startup Battlefield at TechCrunch Disrupt 2025. They have initiated a $5 million funding round to propel their clinical trials forward.
Sean Walsh, the CEO of RADiCAIT, explained the core concept. He stated that they have effectively replaced the most constrained, complex, and costly imaging solution in radiology with the most accessible, simple, and affordable one—the CT scan.
The technology relies on a generative deep neural network developed in 2021 at the University of Oxford by a team led by the company’s co-founder, Regent Lee. This model is trained by analyzing and comparing countless pairs of CT and PET scans, learning to recognize the intricate patterns and relationships between anatomical structure and physiological function. Sina Shahandeh, RADiCAIT’s chief technologist, describes the process as connecting distinct physical phenomena. The system is trained to pay particular attention to specific features, such as certain tissue types or abnormalities, through repeated exposure to diverse examples. This enables the model to discern which patterns hold clinical significance.
The final diagnostic image presented to physicians is the product of multiple integrated models working in concert. Shahandeh likens their methodology to Google DeepMind’s AlphaFold, the AI that transformed protein structure prediction, as both systems excel at translating one form of biological information into another.
Walsh asserts that his team can mathematically demonstrate the statistical similarity between their AI-generated PET images and traditional chemical PET scans. He confirmed that their trial data shows doctors, radiologists, and oncologists make decisions of identical quality whether they are reviewing a conventional PET scan or RADiCAIT’s synthetic version.
While RADiCAIT does not aim to replace PET scans in all scenarios, such as radioligand therapies that directly destroy cancer cells, its technology could render traditional PET scans obsolete for many diagnostic, staging, and monitoring applications. Walsh highlighted the severe supply-demand imbalance in the current system. He explained that their goal is to absorb the overwhelming diagnostic demand, thereby freeing up physical PET scanners to handle the growing needs of theragnostics, which combines diagnostics with targeted therapy.
RADiCAIT has already commenced clinical pilots for lung cancer testing in collaboration with major health systems, including Mass General Brigham and UCSF Health. The pursuit of an FDA clinical trial, a more rigorous and costly endeavor, is the driving force behind their current $5 million seed round. Following potential FDA approval, the next phase will involve commercial pilots to demonstrate the product’s market viability. The company also plans to replicate this development pathway for colorectal cancer and lymphoma applications.
Shahandeh believes the underlying principle of their technology—using AI to extract valid insights without burdensome and expensive tests—has broad applicability. He mentioned they are actively exploring extensions across other areas of radiology and anticipate similar innovations that can uncover nature’s hidden relationships in fields ranging from materials science to biology, chemistry, and physics.
(Source: TechCrunch)
