San Francisco Radiology Startup Epsilon Health Completes $20 Million Series A Led by AlleyCorp
San Francisco radiology startup Epsilon Health has emerged from stealth mode. CEO Rustin Rassoli disclosed exclusively to Axios that the company has completed a $20 million Series A round led by AlleyCorp; the company’s press release and PitchBook record this round and total funding at $27.6 million, with follow-on investors including Uncork Capital, Renegade Partners, SemperVirens, and Jack Altman, who was then a managing partner at Alt Capital.
Epsilon does not create single-disease or single-modality plugins but aims to become the radiology practice itself. The company contracts licensed radiologists and embeds its self-developed models into the image reading and reporting process, with reports signed by doctors sent back to referring parties and patients. Rassoli, who studied mathematics and computer science at the University of Texas at Austin, noted that he witnessed misdiagnoses and delays at his father's imaging clinic, which motivated him to operate a software company, a model group, and a telemedicine association simultaneously, rather than just selling tools to existing groups.
Operational metrics are used to demonstrate that the gap can be filled. The company claims to have served over 250,000 patients, processing more than 2,500 images daily, and expects to review approximately 1% of the daily X-ray volume in the U.S. by 2026; it has taken on more than half of the scanning volume from a major outpatient imaging provider in the U.S. Arjun Karpur, head of machine learning, previously conducted research at Google DeepMind; Chief Medical Officer Roi Bittane was formerly the Chief Medical Officer and Strategic Head at Envision Radiology, described as the second-largest radiology group in the U.S. with over 500 physicians.
The funds will be used for expansion rather than a single-point model competition. Fierce Healthcare reports that this round will be used for market expansion, hiring, increasing clinical collaborations, and infrastructure. Jack Altman commented on the rapid growth from zero to processing thousands of exams daily in less than 10 months, stating that this structure of "becoming the practice itself" will serve as a blueprint for future healthcare delivery. The industry context includes a labor shortage in U.S. radiology: discussions often cite about 19,000 active radiologists, half of whom are nearing retirement, and a projected need for about 30,000 doctors before 2030; the annual scanning volume in the U.S. is estimated at around 700 million scans.
Unlike products like Aidoc that alert based on lesions, Epsilon keeps medical liability, remote signing, and proprietary real case closed loops within the same entity. Doctors generate reports from home, and models iterate based on doctor feedback, with data remaining within the practice. PitchBook marks it as an early-stage company founded in 2024, with employee numbers still in single digits, having completed a seed round in March 2026 and the Series A round on September 10. Axios places this transaction in the investment trend of "radiology still being the most mature scenario for medical AI payments."
The market mechanism is the buying and selling of capacity, not pure software licensing. Buyers are outpatient imaging chains and referral networks that lack overnight image turnaround speed and scalable physician hours; sellers are traditional radiology groups with hourly labor costs and backlog queues. Funding comes from early-stage funds in New York and cross-industry tech investors into an entity that bundles reading services, model weights, and case data together. Beneficiaries are large outpatient providers that can hand over half their scanning volume for faster reports, and contracted physicians whose capacity is amplified by models; pressured parties are AI vendors that only sell single-lesion software and cannot obtain a complete real report closed loop, as well as traditional groups that cannot scale hiring at the same speed. The event-driven aspect comes from the first round pricing after the stealth period, rather than new drugs or new reimbursement codes.
The aging U.S. radiology workforce and the continuous rise in scanning volume make "who owns the signing authority and data closed loop" a variable that monetizes earlier than "who scores higher." Epsilon has turned this variable into an A round asset priced by AlleyCorp.
Source: Public information
ABAB AI Insight
Rassoli's path is not a transition from a paper company to a clinic. He grew up in his father's imaging center and, after early experiences like Atomic, merged three companies into one legal entity: the physician association is responsible for signing and medical liability, the research group uses real feedback to improve models, and the software layer connects to existing referral processes. The hiring of a machine learning head from DeepMind and a former Chief Medical Officer of Envision indicates that early hires are not just demonstration-level annotators but individuals who can integrate the responsibility chain into the practice structure. Many single-lesion companies in the history of medical AI have stalled at the alert stage, unable to sign final reports; Epsilon retains signing authority within its own network of doctors, effectively securing legal imaging qualifications first.
Capital mobilizes the leverage of reading hours, not creating a new cloud billing line. AlleyCorp leads the round, with Uncork and Renegade following, and Jack Altman personally investing; funds are used for hiring doctors, building infrastructure, and securing large outpatient shares. The motivation is: as U.S. scanning volumes continue to rise and doctors approach retirement, whoever captures more than half the volume of a particular outpatient facility simultaneously gains revenue, training data, and referral stickiness. Software licensing cannot achieve this because hospitals can replace models; turning the practice itself into a product means that switching models equates to switching suppliers and doctors.
Comparable entities include Teladoc, which has platformized its network of doctors, Oak Street, which has turned clinics into data flywheels, and Aidoc and Viz.ai, which sell single-disease alerts into emergency departments. Epsilon is transitioning from "tool layer expansion" to "practice layer control": not pursuing a ranking of all modalities but aiming to push daily X-ray processing volume to a national percentage. Jack Altman refers to this structure as the blueprint for future healthcare, fundamentally transforming labor shortages into an investable operational system rather than a research milestone.
The structural change is a reconstruction of the industry chain. The mechanism is: reimbursements still go to the doctors signing the reports, while the model only changes how many reports a single doctor can sign; thus, value shifts from "algorithm accuracy" to "who can organize a licensed signing network and own the returning data." Traditional groups sell physician hours, software companies sell alerts, and Epsilon sells packages of hours amplified by models. The more severe the shortage, the higher the premium on the packages; once an outpatient facility hands over half its volume to the same practice, the switching cost shifts from an IT project to a continuity of care issue. Pricing power thus moves away from model evaluation to the practice entity that owns the case closed loop.
ABAB News · Cognitive Laws
Tools can be replaced, but signing authority cannot be outsourced to demonstrations.
In a shortage industry, capacity itself is the product.
Data closed loops belong to those doing the work, not to those selling plugins.