EP5: Revolutionising Healthcare Diagnosis with AI | Deep Infinity

AI in healthcare promises huge efficiency gains — but in a sector where mistakes can cost lives, efficiency alone isn't enough. In a recent episode of the UK AI podcast, Tim spoke with Rama Boya (CEO) and Prasanthi Enuga (COO), co-founders of Deep Infinity, about how their company is using generative AI and proprietary large language models to reduce misdiagnosis, ease clinician workload, and support — never replace — human medical decision-making.

Balancing Efficiency with Human Oversight

Rama is clear that AI in clinical settings must be built around patient safety first. Because every patient and clinical scenario is unique, Deep Infinity's approach avoids one-size-fits-all workflows, instead building multiple checkpoints into each solution to minimise risk while maximising accuracy. The goal, Rama explains, is a clinician-in-the-loop model — using AI to handle repetitive, time-consuming tasks like documentation and data analysis, freeing clinicians to focus on the expert judgement only humans can provide.

This approach also addresses a real-world problem: patients falling through the cracks due to delayed diagnosis or missed communication between departments. By using AI to track patient records, follow-ups, and recommendations, Deep Infinity aims to reduce that risk closer to zero.

Bridging the Gap Between AI Technologists and Medical Professionals

One of the biggest early challenges for Deep Infinity was bringing together two very different disciplines: AI technologists and medical professionals operating within decades-old clinical procedures. Rama describes initial hesitancy from clinicians unsure whether AI tools would help them or add to an already heavy workload.

Overcoming that required a gradual, trust-building approach — introducing AI tools step by step, demonstrating value, and directly addressing clinicians' concerns. Rama notes that adoption has accelerated significantly over the past year, as clinicians increasingly see AI as a way to reduce documentation burden and free up more time with patients.

AI-Powered Chest X-Ray Analysis: A Practical Case Study

Prashanti explains that Deep Infinity's core mission — reducing misdiagnosis — shaped the design of their product from the outset. A key use case is chest X-ray analysis, where Deep Infinity's proprietary AI models are trained to detect abnormalities at an earlier stage than manual review alone.

Since only around 10–20% of chest X-rays show abnormalities, Deep Infinity's models are designed to quickly identify and prioritise the normal cases, allowing radiologists to focus their expertise on the smaller subset of urgent, abnormal cases. This also speeds up the reporting process significantly — cutting a task that traditionally takes around 30 minutes down to just a few minutes, while automatically routing urgent cases into a prioritised queue.

The underlying technology combines proprietary AI models, trained on millions of chest X-ray images and benchmarked using standard clinical accuracy, sensitivity, and specificity measures, built on top of existing open-source foundation models.

Tackling Bias and Diversity in Medical AI Data

A critical challenge in healthcare AI is data diversity — large datasets are often skewed toward majority ethnic populations, potentially reducing diagnostic accuracy for underrepresented groups. Rama acknowledges this as an ongoing challenge, compounded by varying data regulations between countries that restrict how much and what type of patient imaging data can be used.

Deep Infinity's approach is to build region-specific proprietary models, fine-tuning a global base model with regional datasets — for example, using South Asia-specific data for models deployed in that region — to better reflect the populations they serve.

Scaling AI Across a Complex Healthcare System

With over 200 NHS trusts in the UK alone, each with its own departments, hierarchies, and procurement processes, scaling healthcare AI solutions is a significant challenge — particularly for a startup balancing data privacy, security, and compliance requirements alongside the slow, trust-based process of running pilots.

Rama's advice for other companies navigating this space: focus relentlessly on the customer's actual pain points rather than leading with the technology itself. Understanding what genuinely matters to an IT department or clinician — and demonstrating value in their terms rather than emphasising "AI" for its own sake — is, in his view, the key to avoiding lengthy, unproductive sales cycles.

Navigating Healthcare Regulation

Prasanthi highlights the significant regulatory burden involved in bringing AI medical tools to market, particularly for image analysis tools, which can take well over a year to achieve compliance. Relevant frameworks include GDPR, DTAC, NHS toolkit requirements, and NICE guidelines, alongside medical device classification requirements that vary depending on a product's risk level.

The Future: Agentic AI and Multimodal Diagnostics

Looking ahead, Rama describes Deep Infinity's work building an AI radiology agent — a system capable of not just detecting abnormalities, but automatically triggering next steps, such as booking a follow-up CT scan when a chest X-ray result is inconclusive, and notifying both clinician and patient as part of an integrated workflow. He notes that agentic AI is already established in sectors like fintech, but healthcare has been slower to adopt it due to regulatory complexity — though that is changing.

Prasanthi points to multimodal AI as the next major shift in the space — moving beyond single-scan models (X-ray, CT, or MRI individually) toward systems capable of combining multiple imaging types to build a more holistic view of a patient's condition, informing diagnosis and recommendations more comprehensively than any single model could alone.

Rama also highlights the longer-term potential of quantum computing to enhance machine learning and agentic AI systems in healthcare, citing his involvement in a life sciences and health services working group focused on emerging quantum applications.

What does this mean?

• Successful healthcare AI depends on keeping clinicians firmly in the loop, using AI to remove administrative burden rather than clinical judgement.
• Trust-building with medical professionals is a gradual process, requiring clear communication of value beyond the technology itself.
• Regional, fine-tuned AI models are one practical approach to reducing bias in medical AI trained on non-diverse datasets.
• Regulatory compliance (GDPR, DTAC, NICE, medical device classification) remains one of the biggest barriers to scaling AI in healthcare.
• Agentic AI and multimodal diagnostics are set to be the next major developments in AI-driven healthcare.