Oxford spinout RADiCAIT uses AI to make diagnostic imaging more affordable and accessible — catch it at TechCrunch Disrupt 2025

If you’ve ever had a PET scan, you cognize it’s an ordeal. The scans assistance doctors observe crab and way its spread, but the process itself is simply a logistical nightmare for patients. 

It starts with fasting for 4 to six hours earlier coming into the infirmary — and bully luck to you if you unrecorded rurally and your section infirmary doesn’t person a PET scanner. When you get to the hospital, you’re injected with radioactive material, aft which you indispensable hold for an hr portion it washes done your body. Next, you participate the PET scanner and person to effort to prevarication inactive for 30 minutes portion radiologists get the image. After that, you person to support physically distant from the elderly, young people, and large women for up to 12 hours due to the fact that you’re virtually semi-radioactive.  

Another bottleneck? PET scanners are concentrated successful large cities due to the fact that their radioactive tracers indispensable beryllium produced successful adjacent cyclotrons — compact atomic machines — and utilized wrong hours, limiting entree successful agrarian and determination hospitals. 

But what if you could usage AI to person CT scans, which are overmuch much accessible and affordable, into PET scans? That’s the transportation of RADiCAIT, an Oxford spinout that came retired of stealth this period with $1.7 cardinal successful pre-seed financing. The Boston-based startup, which is astatine Top 20 finalist successful Startup Battlefield astatine TechCrunch Disrupt 2025, has conscionable opened a $5 cardinal rise to beforehand its objective trials.  

“What we truly bash is we took the astir constrained, complex, and costly aesculapian imaging solution successful radiology, and we supplanted it with what is the astir accessible, elemental and affordable, which is CT,” Sean Walsh, RADiCAIT’s CEO, told TechCrunch. 

RADiCAIT’s concealed condiment is its foundational exemplary — a generative heavy neural web invented successful 2021 astatine the University of Oxford by a squad led by the startup’s co-founder and main aesculapian accusation officer, Regent Lee.  

Left: CT scan. Middle: AI-generated PET scan from RADiCAIT. Right: Chemical PET scan.Image Credits:RADiCAIT

The exemplary learns by comparing CT and PET scans, mapping them, and picking retired patterns successful however they subordinate to each other. Sina Shahandeh, RADiCAIT’s main technologist, describes it arsenic connecting “distinct carnal phenomena” by translating anatomical operation into physiological function. Then the exemplary is directed to wage other attraction to circumstantial features oregon aspects of the scans, similar definite types of insubstantial oregon abnormalities. This focused learning is repeated galore times with galore antithetic examples, truthful the exemplary tin place which patterns are clinically important. 

The last representation that goes to doctors for reappraisal is created by combining aggregate models moving together. Shahandeh compares the attack to Google DeepMind’s AlphaFold, the AI that revolutionized macromolecule operation prediction: Both systems larn to construe 1 benignant of biologic accusation into another.  

Walsh claims the squad astatine RADiCAIT tin mathematically beryllium that their synthetic oregon generated PET images are statistically akin to existent chemic PET scans.  

“That’s what our trials show,” helium said, “that the aforesaid prime of determination has been made erstwhile the doctor, radiologist, oregon oncologist is fixed a chemic PET oregon [our AI-generated PET].” 

RADiCAIT doesn’t committedness to regenerate the request for PET scans successful circumstantial therapeutic settings, similar radioligand therapy, which kills crab cells. But for diagnostic, staging, and monitoring purposes, RADiCAIT’s exertion mightiness marque PET scans obsolete.   

RADiCAIT team, from Left: JP Sampson, COO; Sean Walsh, CEO; Sina Shahendeh, CTO; Regent Lee, CMIO.Image Credits:RADiCAIT

“Because it’s specified a constrained system, there’s not capable proviso to conscionable request for diagnostics and theragnostics,” Walsh said, referring to a aesculapian attack that combines diagnostic imaging (i.e., PET scans) with targeted therapy to dainty diseases (i.e., cancer). “So what we’re looking to bash is soak up that request connected the diagnostic side. PET scanners themselves should prime up the slack connected the theragnostic side.” 

RADiCAIT has already begun objective pilots specifically for lung crab investigating with large wellness systems similar Mass General Brigham and UCSF Health. The startup is present pursuing an FDA objective proceedings — a much costly and progressive process that’s driving RADiCAIT’s $5 cardinal effect round. Once that’s approved, the adjacent measurement volition beryllium to bash commercialized pilots and show the commercialized viability of the product. RADiCAIT besides wants to tally the aforesaid process — objective pilots, objective trials, commercialized pilots — for colorectal and lymphoma usage cases.  

Shahandeh said RADiCAIT’s attack to utilizing AI to output valid insights without the burdens of hard and costly tests is “broadly applicable.” 

“We’re exploring extensions crossed radiology,” Shahandeh added. “Expect to spot akin innovations linking domains from materials subject to biology, chemistry, and physics wherever nature’s hidden relationships tin beryllium learned.” 

If you privation to perceive much astir RADiCAIT join america astatine Disrupt, October 27 to 29 successful San Francisco. Learn much here.