
At LSI Europe '24, a panel of seasoned investors and experts gathered to explore the intersection of artificial intelligence and diagnostic imaging. Titled "AI Infrastructure and the Future of Imaging," the session was led by AI lawyer Natasha Allen and featured insights from Vishal Gulati (Recode Ventures), Imran Hamid (LifeArc Ventures), Sahir Ali (Modi Ventures), and Soyoung Park (formerly of VU Venture Partners). With the rise of general-purpose AI models and a surge in healthcare data, the conversation tackled a pressing question: What will it take to realize the promise of AI in medical imaging?
What Makes a Medtech AI Company Stand Out?
According to Gulati, the investment thesis around AI has evolved rapidly. In earlier years, investors prioritized access to proprietary data. Today, he explained, "most of the value is in the assets—what assets you produce." The panel agreed that competitive differentiation now hinges on how companies translate data into proprietary clinical or commercial tools.
"Tell me again, what is your go-to-market strategy?" Gulati said, emphasizing that engineering prowess alone is no longer enough. Reimbursement planning, clinical workflow integration, and regulatory foresight are essential.
Ali agreed, noting that AI's real potential lies in its ability to integrate diverse biological data streams. "We can now collapse imaging, pathology, and genomics into one cohesive view," he said, referencing opportunities in precision oncology. That fusion, he argued, can change how clinicians assess risk and personalize treatment.
Park added that her firm prioritized three things when evaluating an AI healthtech pitch: "Do you really have AI technology? Do you have your own data? And how do you train it?" The quality of training data and methodology can define the success of an algorithm, especially in life-critical applications like radiology and pathology.
Infrastructure Gaps Are Slowing Progress in AI in Medical Imaging
Despite the promise of AI, infrastructural limitations loom large. Hamid raised the issue of GPU access in healthcare systems. "Mass General did a deal with NVIDIA several years ago. They're still using HDX, the 2016 MacBook equivalent," he said. In contrast, companies like Meta and Microsoft are acquiring the latest hardware at a massive scale. "Right now, it seems that PepsiCo has more GPUs than all of the healthcare systems put together," he added.
Ali and Park echoed this concern, pointing to the limited budgets and slow IT cycles of many health systems. Park observed, "A lot of hospitals are still using paper and pen. We're only beginning to see the move toward administrative automation."
Gulati believes this infrastructure gap threatens healthcare’s place in the AI race. "Healthcare organizations are at the bottom when it comes to access to GPUs," he said. Without robust computing power, training and deploying effective models becomes nearly impossible.
Challenges in Adoption: Reimbursement, Resistance, and Relevance
The panelists emphasized that the adoption of AI in healthcare hinges on more than just tech readiness. "The biggest friction to adopting digital pathology wasn't regulatory," said Hamid. "It was pathologists themselves." Even FDA-approved tools languish without provider buy-in.
Ali emphasized the importance of building trust with radiologists rather than replacing them outright. "AI needs to help with workflow first," he said. He cited Rad AI as a promising example, where the technology assists with summarizing impressions rather than interpreting scans independently.
Park agreed: "Companies that started with a clear reimbursement strategy have done better. Engineering-focused startups that ignored that are struggling."
The panel also discussed geographic variation. According to Hamid, many challenges are universal. "The gaps aren’t from the companies," he said. "They’re at the system level. There’s a lack of urgency among healthcare providers and payors."
Opportunities for Innovation in the Infrastructure Layer
Several panelists suggested that the most promising plays in AI are at the infrastructure level—especially middleware. "Imagine an app-store-like environment with accreditation, integration, and commercial layers," said Park. Gulati added that the success of past platforms like Uber was contingent on a broader tech ecosystem—namely smartphones. "We need the same kind of enabler for AI in healthcare."
Looking globally, Park pointed to opportunities in emerging markets. In regions where there are only a handful of radiologists per million people, "AI could leapfrog traditional systems altogether," she said. Gulati agreed, citing parallels with mobile banking platforms like M-PESA.
The Future Outlook for AI in Medical Imaging
As the panel closed, Allen asked for predictions on the future of AI in healthcare. Gulati called AI a "generational platform," likening its impact to that of the internet or mobile phones. "We are in the play mode now," he said. "The killer apps of generative AI in healthcare have yet to be found."
Hamid cautioned that success will come from focusing on practical applications. "Back-end workflow optimization, imaging summaries, triage tools—these are the things that will scale first," he said.
Park agreed, emphasizing the responsibility of today’s innovators. "We're building the fundamentals now that future generations will rely on," she said.
The opportunity for AI in medical imaging is massive—but so are the obstacles. As the panelists made clear, companies that succeed will be those that pair strong technology with thoughtful strategies for reimbursement, infrastructure, and provider engagement. The future may be unwritten, but the path to impact is becoming clearer.