Listen on
Overview
Host Ross Katz speaks with Annemie Ribbens, VP of Science, Evidence, and Trials at icometrix. Precision medicine in neurology faces a critical measurement gap. While new treatments emerge for complex neurological disorders like Alzheimer’s and MS, clinicians and drug developers struggle to track subtle disease progression and treatment impact with sufficient granularity. Standard visual assessment of brain scans often misses critical changes—a 1% brain volume loss or two new lesions can signal significant shifts, yet remain imperceptible to the human eye. This lack of precise, quantifiable data slows drug development, complicates market access strategies, and hinders effective patient care.
For biotech and pharma companies, this means unclear endpoints in clinical trials, suboptimal drug positioning, and delayed time-to-market for effective therapies. For healthcare providers, it translates to delayed diagnoses and less personalized treatment paths, ultimately impacting patient outcomes and operational efficiency. Overcoming this data deficit is essential for driving ROI in R&D and improving the quality and cost-effectiveness of care delivery.
In this episode of Data in Biotech, Annemie Ribbens, VP of Science, Evidence, and Trials at icometrix, shares how her company addresses this challenge. icometrix builds regulatory-cleared AI tools that quantify minute changes in brain MRIs and integrate patient-reported outcomes to provide a clearer picture of disease activity. Annemie discusses the technical architecture supporting compliant data transfer, the operational realities of ensuring data quality across hundreds of clinical sites, and how icometrix partners with pharmaceutical companies to redefine trial endpoints and accelerate the path to true precision medicine.
Key Takeaways
Achieving real-world impact with medical AI requires managing regulatory pathways and securing reimbursement.
For AI tools like icometrix’s to move beyond research and meaningfully influence clinical practice, they must be regulatory-cleared (CE, FDA) and reimbursed. This ensures trust and adoption, enabling the collection of real-world evidence crucial for pharmaceutical market access and post-market surveillance. The process involves rigorous technical validation and demonstrating health economic benefit.
Data quality, not processing scale, often presents the greatest challenge for AI in medical imaging.
While cloud infrastructure can handle large volumes of high-resolution MRI data, inconsistent acquisition practices at clinical sites introduce significant quality issues. Only a small percentage of sites adhere to imaging guidelines, leading to missed information. Automated quality assessment tools, combined with direct consultation, are essential for improving data consistency and reliability.
Comparing medical images over time demands specialized time-series models to reduce measurement error.
Analyzing two follow-up scans together, rather than processing them separately and comparing outputs, dramatically lowers measurement error and reduces bias from scanner differences. This pairwise (or multi-scan) approach allows for the detection of more subtle, clinically relevant changes in disease progression. This is a key differentiator for accurate long-term patient monitoring and trial endpoints.
Neurology is positioned to achieve precision medicine by adopting lessons from oncology’s data-driven evolution.
The future of neurological care lies in moving beyond broad treatment labels to highly specific, biomarker-driven therapies. This requires combining diverse data streams—imaging, blood biomarkers, patient outcomes—into unified models to predict disease course and treatment response. This shift, mirroring oncology’s progress a decade ago, promises a closed system of aligned providers, payers, and pharmaceutical companies.
Related: CorrDyn helps organizations with complex data challenges, from developing a reliable AI strategy to ensuring high data quality and data reliability in regulated environments. Our expertise in the biotech and life sciences sector helps clients gain value from their most critical data assets.
Full Transcript
Jason: Welcome to Data in Biotech, a podcast from CorrDyn where we explore how companies leverage data to drive innovation in life sciences. Every two weeks we sit down with an expert from the world of biotechnology to understand how they are using data science to solve technical challenges, streamline operations, and further innovation in their business. This week, we’re excited to be joined by Annemie Ribbens, VP of Science, Evidence and Trials at icometrix, a company that offers a portfolio of AI solutions to assist healthcare with various challenges in neurological disorders, such as brain trauma, epilepsy, strokes, dementia, and Alzheimer’s disease. During the interview, Ross and Annemie discuss the mission of icometrix in analyzing and treating neurological disorders, the application of machine learning enhancements in their tools to better understand the data that’s being collected from patients and clinicians, what challenges exist in developing the data infrastructure necessary for the compliant transfer of patient data across the platform, and how machine learning will influence the application of precision medicine in biotech over the next five years. Here we go.
Ross Katz: Annemie Ribbens, welcome to the Data in Biotech podcast.
Annemie Ribbens: Thanks for having me, pleasure to be here.
Ross Katz: Awesome. So just to get us started, could you give us a brief introduction to your background and what led you here?
Annemie Ribbens: I hold a master’s degree in applied mathematics from the University of Leuven in Belgium. Afterwards I started a PhD in the engineering department with a focus on brain MRI analysis. I combined mathematical models for disease subtyping based on these brain MRI models to get a different kind of subtyping as compared to the clinical subtyping, to have a better understanding. After this more fundamental PhD, I went to University College London, and there I worked more on the translation towards bringing these kinds of techniques towards clinical practice. That’s also how I got involved with icometrix, which is a company that brings automatic brain MRI quantification towards clinical routine. I started here as a researcher, systematically became responsible for all our research collaborations as well as for pharmaceutical collaborations, and as such I first worked as a clinical project and program manager and then became the VP of clinical trials. Now we actually have a full visualized imaging CRO or life science unit, depending on how you want to call it. I’m responsible for that, and my full role is now VP clinical trials, science, and evidence. Beyond the pharmaceutical collaborations, I’m also responsible for running the evidence studies for our medical devices, working towards reimbursement, as well as working towards better adoption by physicians of our technologies.
Ross Katz: Interesting. So before we get to icometrix, could you just tell us a little bit about why you do what you do? What motivated you to get in this field?
Annemie Ribbens: Yeah, sure. As you can hear, I’m a data scientist at heart, still really a mathematician, but always had a core interest in healthcare and in the medical field, and in neurology in particular. Also very much into the pharmaceutical sector. What drives me is really this combination of AI and data insights and how those can be used to improve drug development, improve care pathways, and as such bringing the right treatment to the right patient at the right time and really improving patient outcome. That’s the core of it all.
Ross Katz: With that in mind, talk to us about icometrix. What’s the mission of the company and how does it accomplish that mission?
Annemie Ribbens: Yeah, sure. The mission is quite closely related with my personal ambitions, which is really nice. The mission of icometrix is to significantly improve the daily lives of people with brain disorders through AI-driven precision medicine. That’s quite a big statement. How does it work in practice? We have a set of AI tools that we developed with a specific focus on brain MRI images, where we help to quantify subtle changes in the brain that you could not visually assess directly. As such you’re capable of detecting earlier disease progression or detecting earlier side effects of a treatment and as such assess drug safety. This allows you to bring the patient faster to the right treatment at the right timing.
Ross Katz: Interesting. There are a variety of MR tools out there for radiologists and biotech companies. Could you talk to me a little bit about the icometrix portfolio and what differentiates it from the other tools that are out there?
Annemie Ribbens: Yeah, sure. Perhaps I start first by explaining what the icometrix portfolio is before diving into depth. icometrix has developed an AI-driven care management platform. It basically consists of two tools. One tool is our quantitative brain MRI images assessment, which is called icobrain. The other one is icompanion, which is a patient app that monitors what’s happening between two neurological visits. All this information is then brought together in one care management platform, which is made available for healthcare practitioners — they have a lot more information, quantitative information, and more subtle information about disease progression and as such they can make better treatment decisions. We have this information available for multiple sclerosis, but also for dementia, Alzheimer’s disease, traumatic brain injury, epilepsy and stroke. What differentiates us from other companies that offer this for radiologists and biotech companies — first of all, as you’ve heard, our goal is in the first place to bring these solutions towards clinical practice. That means there is a need for regulatory clearance. Our solutions are CE marked according to MDR and FDA cleared. Furthermore, for the icobrain solution, we also worked a lot towards reimbursement. We actually applied for a CPT code last year, that was issued beginning of this year by the American Medical Association, and is now officially active. So there is reimbursement now according to CMS for icobrain solutions, which is a big step forward in using these tools in daily clinical routine. Second aspect is more at the technological level, where if you try to bring these solutions towards clinical practice, they have to work for an individual patient. You really have to look at the individual level at what is changing, and the measurement error of your tools has to be very small. There’s a lot more validation, a lot more technical work to bring it to that level. You also need to integrate it properly in clinical routine, so it’s focused on PACS and EMR integration. All these aspects help with our work towards the biotech industry as well, since using these regulatory-cleared aspects could translate things from drug development phases towards clinical routine. We can also really look at individual cases — who is the responder, who is the non-responder, and so on. One final thing: in the field of Alzheimer’s disease, there are now amyloid targeting drugs that have a very specific side effect unique to the drug, which is called ARIA, and that’s visible on the brain MRI scan. icometrix is actually the only company in the world that has developed a software solution to automatically detect this ARIA side effect on the brain MRI scan. That also really differentiates us from other companies.
Ross Katz: Interesting. I think what would help ground us in the conversation about data science as it’s applied to this problem and how these tools are used in a clinical setting is if we understand it from the patient perspective. If I have a neurological disorder and I’m already seeing a specialist who’s helping me with that disorder, how do I encounter the icometrix ecosystem? How are you gathering data about the disorder and how are you helping the physician improve the treatment or understanding of it?
Annemie Ribbens: Yeah, very good questions. We work most of the time directly with providers, which means we go to clinical sites first. Our patient app is publicly available — every patient worldwide can just access the app and start using it. But the connection with the healthcare provider is made via the healthcare provider. The icobrain solutions, which work on the brain MRI scans, are connected to the imaging center you’re going to, or through an academic center if that’s where you’re seen. Once a patient comes in, they have the information they filled out over the year in the patient app, and that becomes visible in the portal. So the provider can already see, for instance, that the patient indicates they have a lot of problems on the cognitive aspects of their disease. They then get a brain MRI scan. On the scan, they might see atrophy in certain regions — we do a precise quantification of those assessments, like there’s this much brain volume loss in the hippocampus. This combination of information helps make better treatment decisions. That’s an example in multiple sclerosis. Other examples are in Alzheimer’s disease, where atrophy patterns and cognition play a crucial role. We also have examples where brain structural patterns can help with a better differential diagnosis, distinguishing Alzheimer’s from frontotemporal dementia. All this information is made available for the healthcare practitioner, and we have quite a few very compelling case studies from providers who say: if we hadn’t had this information, we would have thought the patient was stable when in fact they were not, or we originally thought it was an Alzheimer’s patient when it turned out to be frontotemporal dementia. That really improves diagnosis and monitoring of patients.
Ross Katz: So inside the tool itself that the provider is interacting with, there are AI/ML-driven quantitative assessments happening that are giving them insight into the image they would not otherwise have. Or is there something in the images themselves or in the data collected that is somehow richer or more insightful than it would be with another tool?
Annemie Ribbens: It’s a bit of both. In the first place, what we’re providing for daily clinical routine is something that is very difficult to visually assess. For instance, the brain is a 3D shape. If it shrinks 1%, which is already a lot, it is very, very hard to visually see that — let alone quantify it. If you compare an MS patient versus a normal healthy person, a normal healthy person loses 0.2% of their brain per year, an MS patient loses 0.7% per year. Making a differentiation between both would become very hard visually. Second, in multiple sclerosis there are lesions you can visually see, but the radiologist has a limited amount of time. They have to scroll quickly through the brain MRI scans, and sometimes there are over 100 lesions present. To quickly see whether there is one or two new lesions is actually quite hard, while those lesions can have quite a big impact. We also have technologies that analyze more the microstructure of the brain — this is simply not visual at all. You really need AI software to quantify those.
Ross Katz: That’s really interesting, because it sounds like the features you’re trying to pick up on are built on an understanding of the nature of the diseases you’re trying to detect — the symptoms you would expect to see and where the human eye would fall short. The examples you gave — a volumetric assessment of the brain and a comparison to the previous time the patient was in, or a count of lesions relative to the previous time — the algorithms are not just classifying or diagnosing a particular disease. They’re really giving the provider the insights they need to better understand the progression of the disease. Am I thinking about that right?
Annemie Ribbens: Yeah, that’s correct. Based on this information, providers can better monitor whether there is disease activity or not, and that can be very subtle. Clinically, nothing may appear visual or nothing may seem to be happening, but in the brain changes might be occurring that actually indicate early progression — and due to this additional information, healthcare practitioners can act on that. Same for drug development: if you know very early on that something is changing, it can change your drug development process as well.
Ross Katz: I have a lot of follow-up questions on this but before I get too far, the patient-facing app — how does that play into the data that feeds the entire ecosystem the provider is using to understand the disease? Is it really patient-reported outcomes or is there some way you’re able to measure things about the neurological disorder directly from somebody’s phone that you wouldn’t otherwise get to see?
Annemie Ribbens: It’s definitely a combination of those. Our patient app — the first important thing is that it’s built from the patient out. We have quite good adherence. Everybody is already a little bit sick of apps, I think, nowadays. But because it is built from the patient perspective, around what they wanted to share with their physician, we have quite good adherence. That’s definitely the first part. There’s also a big educational component around improving health literacy with patients. We also look at what the healthcare practitioner actually needs to make decisions. There are patient-reported outcome questionnaires in the app, but there’s also passive monitoring — for instance we link it to step count, which is quite an important marker in multiple sclerosis. We also take sleep patterns into account. And we have some active tests concerning cognition to have an objective assessment rather than only the more subjective questionnaire.
Ross Katz: Interesting. So you’re going beyond patient-reported outcomes and into the real-world data ecosystem as well, directly monitoring what’s happening with the patient and feeding that information back to the provider. Given that you’re moving with the field in terms of what types of diagnostic tools are needed to give the provider better information about disease progression, how do you decide what types of machine learning enhancements to build on the data you’re already capturing?
Annemie Ribbens: There are two answers to this question. One is a technological question — which AI models we’re using. When we started the company they were image-driven, intensity modeling combined with prior information and so on. But once deep learning models and their training became more established, our company moved completely towards using convolutional neural networks to build our algorithms. That’s on the technical aspect. There’s also a clinical aspect: which direction are you moving? Are you doing segmentations of brain MRIs or going towards predictions as well? And looking at how the field is moving — what is relevant at a given point in time? In multiple sclerosis, we know that a lot of treatment decisions are made based on lesion load. Lesion load was of course the first aspect to tackle. But then more and more evidence emerged about silent progression in MS, so we started developing algorithms for silent progression. Same in Alzheimer’s disease — for a long time there was actually no treatment, so what’s the benefit of better monitoring if you cannot act on it? Nowadays treatments are coming to the market, but these treatments have safety effects. A first question is: how can we measure these safety effects faster? The second aspect is treating patients early, so we also have to look towards biomarkers and prediction models — how do we find these patients in the early stages? Our development is of course going in that direction as well, to find an optimal combination of biomarkers towards finding patients as early as possible in their disease course.
Ross Katz: Very interesting. You mentioned convolutional neural networks and this idea that you were originally treating it as an image classification problem or an object detection or semantic segmentation problem, but it also seems like there’s a time series aspect to what you’re doing — you’re trying to understand how these images compare to each other over time. How does that time series component come into the way you’re giving information back to providers? Is it built directly into the models themselves or are you comparing model outputs at different points in time?
Annemie Ribbens: That’s very nice that you notice it — you’re actually coming to one of the key differentiators of icometrix. icometrix was one of the first companies that handled information together. If we have two follow-up scans of the same patient and handle them separately, measurement errors will be a lot higher. If you handle one image and then subsequently the other and just compare outcomes, a lot of bias can be introduced, for instance due to different image contrast. If you handle both scans at the same time and really assess what the difference is between them and try to model that difference, you find a lot more subtleties and your measurement error becomes a lot smaller. That’s exactly where icometrix focused, and it’s one of our key strengths.
Jason: Are you a biotechnology company looking to unlock the potential of your business data? CorrDyn can help. We’re an enterprise data specialist that helps companies working in life sciences make smarter, strategic decisions. From developing the right data strategy that starts with our data maturity assessment to building and delivering bespoke technical solutions, we are equipped to tackle the most complex data challenges. We have partnered with dozens of high-growth organizations, from manufacturers of custom oligonucleotides to molecular diagnostic companies to achieve data competency. Whether you need to supplement existing technology teams with specialist expertise or launch a data program that lays the groundwork for future internal hires, you can partner with CorrDyn to unlock the potential of your business data today. Simply visit connect.corrdyn.com/biotech to learn more. Now, back to the show.
Ross Katz: Interesting. So are you doing pairwise comparisons of the images, or as new images come in are you incorporating them into the model in some way?
Annemie Ribbens: It depends on the application. There’s definitely a lot of pairwise, while in theory we would love to handle all images together. The reason for pairwise comparison often comes down to the clinical workflow — when scans are coming in, are follow-up scans acquired on the same scanner, different scanners, and so on. That’s why most of the time we stick to pairwise comparisons. If we go to trials where there’s more standardization, then we can combine more scans over time and have more of a true time series approach.
Ross Katz: Interesting. The flexibility of deep learning models is both a great strength and a great weakness — the potential architectures you could explore for a problem like this, where you have a sequence of images and something you’re trying to detect, you could spend years going down that path. But you see the opportunity in pairwise comparison, it’s logical, and it fits with the way the provider thinks about the problem, so surfacing the insights back to the provider is so much easier. You have to think about how you operationalize the model as well. Am I thinking about that right?
Annemie Ribbens: Yeah, exactly. What we do in the end is bundle all information together for the provider. If there are providers where we’ve received scans from the last 10 years and they’re doing a very good job scanning patients on the same scanner, we’re capable of really providing what the trends are. That’s really nice to see, because you can clearly see how the lesions evolved over time or how the brain volume evolved over time. And then if you see all of a sudden a strong drop or a strong increase in lesion load, you know that something is going on and you have even more information to make better treatment decisions.
Ross Katz: Right. You’ve got this portfolio of tools and also this portfolio of analytical methods and machine learning models that you apply to the data coming out of those tools. I want to shift a little bit into how you partner with biotech and pharma companies to use both the tools and the data and the skills to help their use case. Could you give us some insight into how those partnerships tend to work?
Annemie Ribbens: Yeah, sure. It’s quite extensive. We come from the real-world setting, and pharma has always recognized that — icometrix is well-connected, integrated with a lot of hospitals, with insights from clinical practice. In that sense they really look to us for real-world evidence. Very logically, that means doing real-world evidence studies, phase four studies, post-market surveillance of their drugs in clinical practice. But also: if these digital tools are in the market and helping put patients on the right treatment and drawing attention to certain biomarkers linked to certain mechanisms of action, how can we leverage those tools to explore how clinical decision making is done in clinical practice and better strategize how a drug is positioned in the market? That’s where we can help with market access strategies. Another aspect — from the real-world setting — is that we have a very large network of integrated hospitals and sites that also serve as a kind of data collection platform, depending on data agreements. In that context we’ve also worked extensively with pharmaceutical companies towards registries, integrating with registries, and so on. A final very classical aspect of our collaborations with pharmaceutical industry: we also function somewhat like a specialist imaging CRO. We help during drug development phases — phase two, phase three studies — where icometrix measurements are used as trial endpoints. Instead of having a radiologist reader manually annotating everything, automated software is used. It can be in combination with radiologist supervision, but using the software first helps build in more consistency and can make the whole flow faster. As the tools are regulatory cleared, they’re very accurate as well, so they can increase the statistical power of the clinical trial.
Ross Katz: Awesome. Just starting with the clinical trial — I’m imagining a pharmaceutical company studying a neurological disorder, starting to write up the protocol, thinking they want to gather specific types of data to measure the impact of their therapeutic approach, and then they reach out to you. Is that the way things start, and how does the relationship grow from there?
Annemie Ribbens: Yeah, both are possible. There are pharmaceutical companies who say: we have this mechanism of action of our drug, we think it might have an effect on remyelination, or on a certain volume of a mass lesion in traumatic brain injury, or we think there might be an inflammation component we want to better estimate — and they ask us what the ideal biomarker is for that. Sometimes we really refine the biomarker in natural history cohort studies to see whether we’re really measuring something relevant to the mechanism of action. In other studies, particularly typical MS studies, there are fixed endpoints — new FLAIR lesions or brain atrophy patterns are quite typical endpoints in phase two and phase three multiple sclerosis studies. These are assessments we’ve already brought towards clinical practice, and we use the same tools in the clinical trial. The major difference is that in clinical practice our tool is used and the radiologist makes the final decision, whereas in clinical trials the radiologist’s supervision is even more strict — they can make small adaptations if necessary to really ensure the highest accuracy for the trial endpoints.
Ross Katz: Yeah, interesting. So your role as research partner to pharmaceutical companies applies when a new type of biomarker is being explored that’s not already being measured in the field. But if the impact they’re trying to have is already measurable with icometrix tools available on provider sites, it’s a matter of making sure all the trial sites have your software and that data is getting transferred back in the correct format and abiding by all the regulations. Am I thinking about that right?
Annemie Ribbens: Yes, more or less. We work most of the time software as a service, so all data is always sent to icometrix. We analyze it in-house, and because our platform is very interoperable, on one hand we have the connection with the clinical sites where they can directly send data from the PACS to icometrix. On the other hand, once we have all the results, we can send them back to the site, but also integrate with data platforms of the pharmaceutical industry such that the trial endpoints we calculate go directly into the databases for the trial itself and do the proper database analysis there. As mentioned, we are sometimes developing new biomarkers, but sometimes just acting as an imaging CRO with the software as a backbone to provide these services in a smoother way.
Ross Katz: Yeah, interesting. Obviously you’ve developed these tools for the measurement and understanding and diagnosis of different neurological disorders, but you’ve also got to have this back-end data infrastructure to make sure that patient data is transferred in compliance with all the regulations that govern it. Can you give us any insight into what that journey was like — developing the data infrastructure to put that in place?
Annemie Ribbens: It’s definitely changed a lot over time. We started with a public data infrastructure but quite fast we created our own in-house data infrastructure. Core to our approach is definitely the development of the icoBridge, which is the integration with the PACS system. It allows us to bring data directly from the PACS to the icometrix cloud, pseudonymize it on the way, and then it sits in the icocontainer, which is our own data infrastructure. There it’s automatically analyzed and then the results can be sent back to the same PACS using the encryption key, placing them in the right location. That’s definitely one core part of our infrastructure. We also have EMR integration via Smart on FHIR, which helps further integration in the clinical workflow, particularly with neurologists. The icocontainer — the database infrastructure itself — is built with microservices and is already very interoperable. We follow the OMOP Common Data elements to capture data in a standardized way, and the FAIR principles to have data that is findable, accessible, interoperable, and reusable. The full system is ISO 27001 certified for information security, to make sure all data is properly protected.
Ross Katz: Do you have a sense of how many providers you’re supporting right now?
Annemie Ribbens: It’s always a difficult question. I think we are connected with around 150 centers now, but what is tricky about that is we’re connected with large imaging networks — for instance, RadNet and Shields in the US — and they in turn are connected with a lot of private practices and neurology centers. The total number is a lot larger than 150. But 150 is really the direct full integrations we have. On top of that, through pharmaceutical industry we’re also working with around 400 sites, so on top of the 150 there’s actually a lot more. It’s a difficult number to give, but we’re a still-young company and already have quite a nice integration in the clinical workflow.
Ross Katz: And do you have a sense of the order of magnitude of the number of patients whose data you’ve analyzed?
Annemie Ribbens: Good question. I think we analyze around 50,000 scans per year — but don’t quote me on that. I think it’s around 50,000 a year.
Ross Katz: What comes to mind for me with the types of data you’re working with is that these are large datasets — the scale for a single scan is really large. What are some of the challenges you’ve faced in processing and analyzing huge quantities of these very large datasets and trying to scale out the infrastructure?
Annemie Ribbens: Scaling out the infrastructure was quite a ride. We work cloud-based and we have very good IT people; we work with microservices so we can deploy on a new server very easily. In terms of scalability in terms of numbers, that’s not the biggest issue. I think the biggest issue is data quality. Although there are guidelines nowadays in terms of what type of imaging you acquire and what quality it should be — there are quite a lot of institute foundations and consortia that provide guidelines, like CMSC and MAGNIMS on doing a proper 3D T1-weighted scan — we still see that in clinical practice only 5% of sites actually adhere to these guidelines, which is very low. I think it’s increasing, but still very low. That poses a lot more problems: scanner switches, or scans with insufficient quality. If you have very thick slices and you’re trying to measure a very small lesion, it can just fall in between the slices and you cannot see it anymore. A lot of information can be missed due to poor-quality scans, and I think that’s a bigger hurdle than the scalability of the analysis itself.
Ross Katz: That’s so interesting. How do you work with providers in that situation? I would imagine that the providers have an interest in providing you the quality of data you need to give the correct diagnosis.
Annemie Ribbens: Yeah, that’s indeed true. Luckily most of the providers working with us try to adapt. We have a special tool also developed, called QualityMetrics. QualityMetrics allows us to benchmark the scans we’re receiving from the site against a database we have in-house. As such we can basically say: your data is of sufficient quality, or actually there’s something wrong — field of view is insufficient, there’s a lot of movement artifacts, the contrast-to-noise is not sufficient, things like that. We can truly benchmark and advise on optimal protocols. But even with an optimal protocol, by using QualityMetrics we can still see whether there might be factors influencing it. A scanner is a magnetic coil — there can be small things that influence quality, and problems can exist like leakages in the coil, which can result in very poor image quality. Using this QualityMetrics tool we can at least know where problems are coming from or what the key problem is, and then we can advise on better protocols or help with improving acquisition or pointing towards aspects of the scanner where updates are probably required.
Ross Katz: So you have automated tools giving you baseline metrics for quality, but there also needs to be a human in the loop to help assess what exactly is going on with this image or why the quality is lower than it might otherwise be. Am I thinking about that right?
Annemie Ribbens: Yes — the first aspect is our automated assessment, the QualityMetrics tool, which is an AI-based methodology to assess quality. But there’s always human interaction. When looking at why quality is not good, what should we advise, what might be the underlying cause — that’s often discussed with the center as well.
Ross Katz: Interesting. As you encounter different neurological disorders that biotech companies are researching and the different biomarkers they’re searching for, how much of the work you’ve done to develop the pipeline for analysis of MR and CT scans is reusable for those problems, and how much do you find your organization exploring new pathways, building new infrastructure, trying to think differently about the problem space?
Annemie Ribbens: Good question, broad answer. If you’re looking at different neurological disorders, there’s definitely a generic framework for quantitative analysis of brain images — that framework is kind of the same. But in terms of properly training and validating it, we still do that for every disease separately. The reason is that every disease has its own distinct problems and presentation. MS lesions look quite different from brain tumors, quite different from traumatic brain injury, or even from these safety effects in Alzheimer’s disease. So we really retrain our algorithms per disease, but the generic framework stays the same. In validation, we first look at technical validation focused on accuracy, reproducibility, and sensitivity. After that, we look at clinical validation — what is the actual clinical impact of this tool? And then a health economic and outcome assessment: what is the real impact on patient outcome and health economic benefit, in terms of quality-adjusted life years and cost-efficacy? We try to evaluate that in each disease space separately. The second aspect is that I’m talking mostly about quantitative image analysis, but we try to go beyond that. We have multiple projects starting where we try to combine multiple biomarkers — evoked potentials, OCT, blood biomarkers — and combine this with the imaging information as well as patient-reported outcomes, putting this in unified models to see how we can optimally address the true disease course.
Ross Katz: At the baseline level, the interpretability of the models is really important because the core of what you’re doing is not diagnosing the disease — it’s providing insights to the providers so they can better diagnose it. That strikes me as a core reason why you have a model class per disease category: because interpretability depends very much on the logic underlying the way we understand that disease. Is that right?
Annemie Ribbens: Yeah, that’s correct. That’s one of our core reasons why we act in different ways for different diseases.
Ross Katz: The second thing I thought I heard was that over time, as we understand a disease better and have this entire ecosystem of data surrounding it, there’s also the opportunity to combine all the pieces of data together and have a less interpretable model that comes up with a probability that this person has a particular neurological disorder. Is that right?
Annemie Ribbens: Yeah, exactly. These technologies are still in the development phase. We have multiple models ongoing, working on large databases to get the right insights. The second challenging aspect will be: how do we bring this towards clinical practice? What are the regulatory aspects around this kind of device, since they are more like decision tools rather than just providing information? That’s still a big challenge, but at this point in time they’re already providing a tremendous amount of insights — which patients are fast progressors, which are slow progressors, what’s the underlying reason, what disease insights do we get from that, what is silent progression, which progression in MS is caused by relapses versus independent of relapses. These insights are of key importance for developing and bringing new drugs to the market. Although all these models are not yet in clinical practice, they’re already providing a lot of insights at this point in time.
Ross Katz: It sounds like you’re involved on the commercial side with biotech and pharma companies who are trying to understand how diagnosis occurs inside a provider setting and how they position their therapeutic approaches given how these diseases are diagnosed — and also that you’re using an understanding of the commercial landscape for these diseases to prioritize where you’re putting your attention from a modeling perspective. Am I hearing that correctly?
Annemie Ribbens: Yeah, definitely. That’s definitely one of the key aspects of what icometrix is doing. We are also heavily involved on the medical side — in the drug development aspect and in post-market surveillance studies. But indeed the commercial part, getting insight, and also positioning a certain drug based on certain biomarkers — that is definitely also a core part of our business.
Ross Katz: It sounds like you’re spanning the entire scope of biotech activity and, as you mentioned, you’re a relatively small company so there’s only so many places you can focus. How do you think about prioritization when you’re working with a data science toolkit that is really flexible and can be applied to very different areas? How do you determine where to focus the energy of the organization?
Annemie Ribbens: Focus is important. We are trying to cover the full pharma landscape end-to-end. We truly believe in that — if you start developing a drug, you should already at that point in time be thinking about which patients you’re going to target and how you’re going to reach that target population. We truly believe in this end-to-end approach. But it makes things sometimes difficult because pharma itself is often quite fragmented, so you have to work with all these different divisions over time and also bring these solutions and this vision towards the full company. The advantage is twofold. First, we come from clinical practice, so they know us in this field and know that we can bring things to clinical practice — that’s something they’ve seen and trust us for. The second advantage is that we really try to focus on certain neurological conditions. Our first focus was definitely multiple sclerosis, as there were already disease-modifying treatments. Now with treatments for Alzheimer’s disease coming to the market, there’s definitely an opportunity there as well, and we see MS as a bit of a blueprint for Alzheimer’s disease. Our focus, although we do other neurological conditions as well, is these two core areas, because the story is the most clear for taking this end-to-end approach.
Ross Katz: Do you have an example project that you’re most proud of in terms of partnering with biotech companies to do either medical research or impact clinical practice?
Annemie Ribbens: There are still a lot ongoing where I cannot yet disclose the results. There’s one that is a little bit longer ago, but I think it tells the story of icometrix. A lot has evolved since, and I think we approach things in a more fundamental way now, but it brings a little bit the story. It was an MS drug that was launched, and at the time there was not much known about brain atrophy — definitely not with the day-to-day physicians. Most treatment decisions were made, and still are made, on lesion load. But the drug itself had quite a big impact on brain atrophy. The company wanted to create awareness with physicians that impacting brain atrophy is also important because it’s a form of progression and can ultimately indicate disease worsening. We worked together with the pharma company to bring our icobrain tools towards clinical routine and towards a lot of physicians. Due to that, we created awareness around brain atrophy because they were now capable of quantifying it. The pharma company was better capable of product differentiation because brain atrophy became more common knowledge — whereas in the past it was mainly known by key medical experts and not really taken into account by the general neurologist. At least awareness was created. Not everything is clear about brain atrophy yet, but it had quite a big impact on product differentiation and for us it was also important because it helped with bringing our product towards clinical routine.
Ross Katz: It just comes back to: if you can’t measure it, you can’t really influence it or even understand your influence on it. Having that measurement there in front of the provider created the platform from which that influence could occur. Very interesting. As we come to a close, I want you to look into the future — how you see icometrix evolving, how you see machine learning in biotech in your arena evolving, and give us some insight into where you see things going over the next three to five years.
Annemie Ribbens: There’s a lot to do with AI algorithms in general — in terms of trial optimization or drug discovery — but in general, and also in the vision of icometrix, what we really hope for is a true vision of precision medicine, of population health. In that sense, I think we are now at a point where oncology was almost 10 years ago. Oncology developed the next generation of genomics — a scalable, sensitive, data-driven detection platform with the ultimate tumor genomic profile — and based on that, it became very specific for certain cancers: this cancer you can treat with this kind of treatment, because you did a certain risk prediction. If you look at neurology today and DMT usage — the disease-modifying treatments — the labels are very broad. You can treat everybody with almost every treatment for neurological conditions. While in fact we don’t know enough about who has a favorable disease course versus who has a poor prognosis or who is more prone to side effects of certain drugs. We strongly believe in having better measurements available to really move towards proper precision medicine. In this context, we also believe in the collaboration between provider, payer, and pharmaceutical companies to really look at the landscape as one — treat the right patient at the right time, reimburse those patients, reimburse the tools, and have a closed system. That’s our vision, and I think oncology was there 10 years ago, so I believe neurology can be there in 5 years.
Ross Katz: Wonderful. Well, as we draw to a close, can you let us know where people can find you or learn more about icometrix?
Annemie Ribbens: Yeah, sure. We have a website at www.icometrix.com. There are also our social media channels, particularly LinkedIn. We also often present at conferences — mainly radiology and neurology conferences — and we’d be pleased to see people in person there. Some conferences by name: RSNA, ASNR, AAN, ECTRIMS, AAIC, and ECR. It would be a pleasure to meet people there.
Ross Katz: Thank you. Annemie, it’s been a pleasure. Thanks for coming on and I look forward to connecting down the line.
Annemie Ribbens: Yeah, thank you so much for your time. It was a pleasure to be here.
Jason: And that’s it for this episode of Data in Biotech. If you enjoyed the episode, please subscribe, rate, or leave a review in your podcast platform of choice. See you next time.
