‍Key Takeaways:

• Radiomics has the potential to democratize access to clinical expertise to ensure the highest quality of care for all patients. Disease characterization by neural networks enables earlier patient identification, stratification, and optimal treatment decisions
• Radiomics tools developers should consider fit with current practices and clinical workflows while focusing on the high-priority clinician needs. This approach will help ensure broad adoption and maximize clinical ROI, on top of FDA clearance and Medicare reimbursement already secured by some solutions.
• Pharma and life science companies are exploring radiomics both in clinical product development and strategy (e.g., stratifying patient populations, speeding up trial enrollment, and identifying non-responders early) and in real-world settings for market growth (e.g., identifying eligible/untreated patients and increasing treatment rates)
• Radiomics can play a unique role in precision medicine, integrating multi-modal genomic and phenotypic data

‍

Hi Václav, thank you for joining us. To start, could you please share a bit about your background and about Optellum?

Yes, I’m a machine learning and computer vision scientist, and focused on lung cancer early in my studies. My interest was partly driven by a personal experience: my aunt, who has never smoked a cigarette, was diagnosed with stage IV lung cancer and passed away just months later. This experience shaped my view of lung cancer as one of the biggest challenges in oncology.

During my PhD at Oxford, we were one of the first in the world to be applying modern machine learning to support treatment planning for lung cancer. Years later, I reunited with my PhD supervisors to tackle these challenges on a large scale by founding Optellum. Advances in Computed Tomography (CT) technology, cloud imaging archives, and GPU architectures for training large neural networks have now made this vision a reality.

At Optellum, which is based in Oxford, UK, with a growing U.S. commercial team, our mission is to help physicians diagnose lung disease as early as possible and provide optimal treatment for their patients. We are proud to be the first and only FDA and EU MDR-cleared AI decision support software for lung cancer diagnosis, reimbursed by Medicare. Our platform is already making a significant impact, helping clinicians diagnose patients and save lives across leading health systems. Additionally, we’re collaborating with major partners like Intuitive Surgical, Johnson & Johnson, GE Healthcare, and several top biopharma companies. They help us accelerate expanding access to Optellum for more clinicians and patients around the world and with enhancing the platform with new modules supporting AI-guided precision treatment.

‍

Thank you so much for that background. Many aren't too familiar with ‘radiomics’, and I think even some of us within the space may define it differently. So, before we get too deep into the conversation, I’d like to ask, how you define radiomics?

I completely agree—radiomics is still a relatively new field, and its definition can vary. At its core, radiomics involves transforming pixels from standard medical images, which are already ubiquitously available in every hospital, into digital biomarkers. These biomarkers are similar to conventional metrics like blood glucose levels—quantifiable, reproducible, and ideally tied to objective clinical outcomes. The goal is to move away from the subjective “eyeballing” of images and instead provide measurable, consistent data that drives real clinical utility at a scale.

‍What do you see as the main applications?

Radiomics can be thought of across several horizons in terms of clinical impact and complexity.

It starts with the detection and quantification of “hand-crafted” image features, the most straightforward application. One of the early examples was automating the measurement of features that trained clinicians already assess, such as lesion size or 3D volume, within clinical workflows and therapeutic development. However, this doesn’t really touch the surface of the full potential of radiomics.

The next step is in precision disease characterization. Moving beyond simple metrics like size, although these are important as they are in imaging guidelines, radiomics can enable what the FDA refers to as CADx, or computer-aided diagnosis. One example is systems that can determine whether a lesion is malignant or not, leveraging deep neural network models that analyse the full richness of image pixels to uncover subtle patterns.  Such systems can allow clinicians at all levels to perform at the top of their license, democratizing access to expertise that typically resides only with top specialty-trained physicians and so expanding access to quality care.

The final frontier is outcomes prediction. By using pre-interventional data, we can predict disease prognosis and even treatment response. Ultimately, we move beyond pixels to integrate multimodal data, combining longitudinal imaging with clinical information and molecular biomarkers, to help guide the right choice of treatment for the right patient.   

‍

That’s a great framework. Can you give us some examples of how radiomics can improve patient care? Also, what do you see as the gaps and unmet needs in the current practices where Optellum and other radiomics tools have a chance to improve patient care?

The huge potential of platforms like Optellum’s is driven by the ubiquitous availability of CT scans, with nearly 100 million taken annually in the U.S. alone– enabling “opportunistic screening” of patients before symptoms appear. For example, I have a small nodule in my lung, like millions of other patients with suspicious lesions found incidentally in scans taken for unrelated reasons. Most are not cancerous, but identifying those that are is crucial - essentially finding the needle in the haystack. Our platform analyzes all scans and radiology reports across the health-system to identify at-risk patients, helping prioritize those who should be rapidly escalated to specialists for follow-up procedures or treatments. This is the essence of precision medicine - I’ll break it down into key levers of how we help providers, by looking comprehensively at the care pathway.

The first lever is automated earlier patient identification. While some areas of medicine are hesitant about full automation, particularly among radiologists, we've focused on supporting clinical teams in automating scanning radiology reports for incidental findings, with no interruption to their workflows. Instead of hiring additional staff to manage this, our tool reviews thousands of reports in a second, identifying suspicious cases to prevent them from slipping through the cracks, thus reducing patient safety incidents and leakage.

Once you automate patient identification, it opens a floodgate: There are so many of these patients that have not been followed, even in emergency room CT scans alone.  That’s why, after automated identification, comes prioritization and AI staffing efficiency. That means stratifying patients through the use of an AI and radiomics-based malignancy risk score, reimbursable by CMS Medicare. This can help providers to drive efficient care orchestration. For example, to decide which patients need accelerated referral to a nodule clinic or remain managed by their PCP.

The final level is augmenting human capabilities, or AI upskilling. Rather than replacing experts, our goal is to upskill clinical teams to perform at the top of their license. In the case of clinical trials, pharma often turns to elite academic institutions like MD Anderson for expertise, yet the vast majority of lung cancer diagnoses happen in community settings. We're aiming to bring AI-powered expertise to every clinician and democratize access to world-class care regardless of location. And even within elite centers, we can help turn every nurse practitioner or fellow into a top thoracic oncology expert – enabling them to prioritize the right patients and so freeing up surgeons and interventional pulmonologists’ time to focus on procedures.

“…Rather than replacing experts, our goal is to upskill clinical teams to perform at the top of their license. In the case of clinical trials, pharma often turns to elite academic institutions like MD Anderson for expertise, yet the vast majority of lung cancer diagnoses happen in community settings. We're aiming to bring AI-powered expertise to every clinician and democratize access to world-class care regardless of location…”

‍

Thinking about the ways that these tools can improve patient care, what do you see as the state of radiomics adoption today? What technologies are used, what's at the cutting edge, and what are the barriers that are preventing the next generation of radiomics tools from becoming universal?

As we approach the end of 2024, we’re seeing notable progress in radiomics adoption. Acute stroke care is a great example, where the urgency of treatment has driven radiomics to become standard of care. In oncology, particularly lung cancer, we’re past the early adoption phase and witnessing exponential growth. Our initial customers are scaling up from initial sites into system-wide adoptions, which signals that we’ve moved beyond the visionaries into broader acceptance. What’s driving this? The evidence for accuracy and safety is solid, and FDA clearances across various specialties have paved the way. Reimbursement will be a key factor in making these tools widely adopted. Real clinical ROI—such as increased volumes of curative treatments and staffing efficiencies—also plays a significant role.

That said, deployment by hospital IT remains a bottleneck. Many institutions face concerns around cybersecurity and safety, with IT directors hesitant to adopt specialized applications. This can be overcome by demonstrating clear clinical value and engaging with service line leaders at the executive level. Partnerships with trusted enterprise IT companies also help ease these concerns.

Another important factor is aligning radiomics with treatment. When you create synergies between the patient, provider, and therapeutic vendors, you strengthen the overall ecosystem.

Lastly, even with payer reimbursement and proven ROI for administration, building a great user-friendly product that integrates seamlessly with clinical workflows is critical. This means ensuring the technology is not only effective and satisfies FDA requirements, but also makes lives easier for clinicians, creating a real win-win.  Complex coordination of all these factors is challenging but exciting, and creates significant barriers to entry.

“…This means ensuring the technology is not only effective and satisfies FDA requirements, but also makes lives easier for clinicians, creating a real win-win.  Complex coordination of all these factors is challenging but exciting, and creates significant barriers to entry…”

‍

Matching the existing clinical workflow is critical, this is top-of-mind for clinicians as they’re using new tools and technologies.

Switching gears, what value do radiomics tools have for life sciences and pharma companies? It would be great to hear about what is currently done, and also the future potential.

Therapeutic device companies are often ahead in this area, as the integration of computer technology and CT scans is central to their image-guided therapy solutions. However, pharma companies have traditionally overlooked the potential of radiomics. Many have not even collected raw DICOM images and have relied solely on reports from CROs.

A decade or so ago, the focus was on genomics and NGS as the key to solving many challenges. While genomics certainly offers value, it's now becoming clear that integrating phenotypic and other clinical data provides a much richer, more complete picture of a patient’s condition. Currently, the use of radiomics in pharma is limited, but this is changing rapidly. Visionary organizations are now recognizing its potential and are starting to collect this data. This shift opens up significant opportunities for future impact by enabling earlier patient identification, more comprehensive patient insights, and better-targeted therapies, providing a competitive edge for life science companies willing to invest in radiomics – accelerating their clinical development and growth.

“…While genomics certainly offers value, it's now becoming clear that integrating phenotypic and other clinical data provides a much richer, more complete picture of a patient’s condition…”

‍

What are some of the use cases for radiomics for life sciences and pharma companies?

At the highest level, I see three key use cases for radiomics, with the most exciting being the integration of the first two.

The first is clinical product development strategy and operations, where radiomics tools can help accelerate clinical trials and reduce costs, by enhanced precision. Specific examples in early phase development include more robust and accurate identification of responders, allowing for earlier decisions on which programs to progress. In later phases of development, AI and radiomics can help speed up patient enrolment through early identification and cohort enrichment. It can also enable earlier identification and mitigation of adverse events (AEs). Finally, it can help stratify patient sub-populations, for example, to identify responders in failed trials and high-risk patients who may benefit from additional. more aggressive treatment. 

The second exciting use case, where radiomics is gaining significant traction is commercial market growth. Here, AI and radiomics are used by providers in real-world clinical settings, to increase treatment rates of patients who could benefit from already-approved treatments. That has a huge ROI for both pharma and providers – helping them to find the needle in the haystack through early identification and stratification. For example, it helps uncover patients who may have mutations but haven’t received the necessary tests or those who are eligible for treatments but haven’t been properly identified and referred to a specialist. 

The ultimate win-win comes from combining both aspects: partnering in clinical development, especially in Phase II and III trials, and extending these tools as “digital companion diagnostics” into routine clinical practice, with pharma partner resources helping to accelerate development, validation and commercial availability. These digital tools can help clinicians identify more patients for the right therapy, even without being explicitly included on the drug labels. While there’s some conservatism and risk aversion around regulatory and HCP compliance, we’re seeing a shift toward greater receptivity, and I believe there’s much more to come.

“…The ultimate win-win comes from combining both aspects: partnering in clinical development, especially in Phase II and III trials, and extending these tools as “digital companion diagnostics” into routine clinical practice, with pharma partner resources helping to accelerate development, validation and commercial availability…”

‍

How can we drive broader use of these tools among life sciences and pharma stakeholders? What can companies like Optellum do to accelerate adoption, and how can stakeholders on the life sciences and pharma side help to overcome challenges?

One parallel to look at is the evolution of genomic biomarkers. Around 15 years ago, they were met with skepticism, but now they are essential to development and commercialization strategies. Pharma is data-driven, and solutions need solid data backing. We're seeing the same shift with radiomics—pharma is starting to fund validation efforts for radiomics tools, which accelerates their development and ensures alignment with the needs of the pharma industry. This early-stage investment is a win-win, helping pharma while enabling the growth of radiomics.

A challenge is that life sciences companies don’t always capture imaging data, but this is an education gap that is quickly being addressed. Some companies are already ahead in recognizing its value. I see a huge opportunity in positioning radiomics as a key tool to integrate information from diverse data sources—combining imaging, genomics, and phenotypic data in multi-modal AI prognostic and predictive models.

For companies like Optellum, a significant advantage is that CT scans are already ubiquitous in clinical practice. They’re available for nearly every patient and can be used multiple times, unlike biobanked blood and tissue samples that are limited and often competed for with other methods. This makes CT scans a powerful tool to link disparate data sources and drive more comprehensive patient insights without the limitations of other biomarkers.

“…Pharma is data-driven, and solutions need solid data backing. We're seeing the same shift with radiomics—pharma is starting to fund validation efforts for radiomics tools, which accelerates their development and ensures alignment with the needs of the pharma industry…”

‍

You’ve brought up the size of the training datasets that Optellum is using. We’ve heard a bit about the heterogeneity of images, and imaging practices. How much of an issue is that heterogeneity? Is this something that presents a hurdle going forward, and will there need to be greater standardization of the images that go into these algorithms?

For Optellum, heterogeneity in imaging is actually a significant opportunity. To secure FDA clearance, we demonstrated our tool’s performance on a diverse set of real-world data, accounting for huge differences in real-world patient populations, and numerous scanner models, reconstruction algorithms, resolutions, contrast agents, and radiation doses. This is crucial because, to be adopted widely, solutions must be designed to handle heterogeneous, real-world data—not just idealized, standardized data from clinical trials.

To design scalable radiomics tools, we must ensure that our tools can work in practical, real-world settings. Designing solutions for standardized clinical trials alone—where conditions are artificially constrained—doesn’t reflect the realities of patient care. Our tools must be robust enough for everyday use. And building datasets to train radiomics systems from real-world clinical archives, requiring linking images to diagnostics and therapeutics outcomes, is challenging, especially in the case of early diagnosis where no standardized registries exist.  

There’s another advantage of radiomics here: advances in imaging technology have led to lower radiation doses, for example, but radiologists like their new images to look similar to the old ones. This ensures backward and forward compatibility, not available for molecular diagnostics.

One challenge is that high-resolution scans are often deleted weeks or months after they’re taken. With cloud storage, there’s no reason not to store all this data for AI use.

“…To design scalable radiomics tools, we must ensure that our tools can work in practical, real-world settings. Designing solutions for standardized clinical trials alone—where conditions are artificially constrained—doesn’t reflect the realities of patient care…”

‍

How do you expect the radiomics landscape to change over the next five years? Are there any near-term trends or developments that you're most excited about?

Over the next five years, I expect radiomics to shift from being a tool just used by early adopters, into mainstream use. The specific applications will vary across clinical specialties and markets, but in high clinical ROI applications that overcome regulatory and reimbursement hurdles, radiomics will become a standard of care for physicians.

On the pharma side, forward-thinking companies will realize that these tools give them a competitive edge in dramatically accelerating product development and commercial growth. Success stories will drive a virtuous cycle of adoption, with more providers and pharma companies on board. I’m particularly excited about the potential of digital tools powered by radiomics to enable precision medicine in curative settings, such as early-stage non-small cell lung cancer. In these settings, traditional liquid biomarkers are limited, but radiomics can help drive label expansion, for example, by identifying high-risk Stage I-II patients. This area is evolving rapidly, with new therapies emerging every few months. The rapid development creates confusion, especially as it requires novel collaborations across oncologists, surgeons and pulmonologists. AI has the potential to become the objective trusted advisor of a virtual multi-disciplinary tumour board, harnessing the collective expertise of thousands of physicians and consolidating the myriad types of multi-modal patient data, from imaging to diagnostic tests.

Across all disease settings, I see a big future for multi-modal digital biomarkers that integrate radiomics with clinical risk factors, liquid and tissue genomics, and other sources. This is especially critical in complex diseases like advanced NSCLC, where guidelines struggle to keep up with the fast pace of biomarker development and combination therapies, which have increased costs and risks of adverse events. Radiomics holds the potential to unify expertise and democratize access to high-quality care, particularly in community settings.

Finally, I’m personally convinced that AI-powered population health “opportunistic screening”, leveraging standard-of-care imaging acquired for other reasons, will enable earlier treatment across a range of lung diseases beyond NSCLC. The multi-modal AI integration of radiomics and clinical data will converge with advances in minimally invasive devices and molecular diagnostics, to drive more patients toward the latest targeted therapies and immunotherapies.

“…I’m particularly excited about the potential of digital tools powered by radiomics to enable precision medicine in curative settings, such as early-stage non-small cell lung cancer. In these settings, traditional liquid biomarkers are limited, but radiomics can help drive label expansion, for example, by identifying high-risk Stage I-II patients. This area is evolving rapidly, with new therapies emerging every few months…”

‍

Awesome, we are looking forward to future developments! There are lots of exciting things that could be coming down the line, with potentially great benefits. Thank you for your time.

‍