Biogen Digital Health

Seizing the inflection point - The potential for AI to advance personalized healthcare

By Steffanie Bristol, Head of Imaging and AI/ML at Biogen Digital Health
March 31, 2022

During the past decade, artificial intelligence (AI) and machine learning (ML) have transitioned from figments of science fiction to the status quo of everyday life. 

In industries ranging from agriculture to advertising to automotive, AI/ML is driving exponential progress and establishing a foundation for human/machine collaboration. Whether or not we realize it, AL/ML powers the outputs of smartphone map application driving route recommendations, personalized content for products on social media ads, and speech recognition that creates dictation text, emails, physician notes from voice-based devices.

Self-driving vehicles infused with AI capabilities are a reality, from Tesla cars for everyday driving to self-driving tractors1 created by manufacturer John Deere that navigate obstacles and plant crops autonomously while farmers observe the process remotely via an app.

While many fields have embraced AI/ML as a core part of business operations, the healthcare industry lags in applying this computer science field despite much hype. Impact today is most evident in areas related to standardization and automation of healthcare operations; AI/ML for predicting or confirming clinical diagnosis and informing patient treatment plans remains in its infancy.

Steffanie Bristol
Head of Imaging and AI/ML at Biogen Digital Health

The lag may be explained from various factors.  Access to adequate amounts of high-quality data to power accurate algorithm development is far more challenging in healthcare than collecting data about real-time driving patterns or product recommendations for new furniture. Foundational healthcare IT infrastructure is still modernizing. Regulatory uncertainty remains. Scepticism persists, resulting from lack of results reproducibility and rigor of validation studies. End users seek clarity from developers that algorithms are not biased and transparency in how algorithms work.

Despite these challenges, there has been great progress across various key enablers for AI/ML in healthcare to move beyond hype to impact. Progress is being propelled by more efficient and affordable cloud storage systems, more robust computational power, improved regulatory guidance from organizations like the Food and Drug Administration (FDA), international collaborations for standardization such as the image biomarker standardisation initiative (IBSI)2, and an influx of patient data from clinical records, blood test, imaging studies, genetic testing, and digital tools like sensors & wearables. Start-ups are working with big players in technology & healthcare to gain access to high-quality data necessary to inform highly accurate algorithms with adequate study design to demonstrate effectiveness. These “enablement layers” make up the backbone of successful and widespread AI/ML deployment - which can ultimately lead to better outcomes for patients and more personalized, humane care. 

As the Head of AI/ML and Imaging activities at Biogen Digital Health (BDH), I am excited about the profound implications of AI/ML to unlock the mysteries of the human brain, improve diagnostics for diseases, increase clinical trial efficiency to get better therapies to patients faster, and create more personalized medicine — a concept that’s long been near and dear to my heart.

AI/ML use cases in healthcare

Can you imagine a world where technology allows healthcare to be more humane versus disease focused? Where automated tools replace the time clinicians spend charting in electronic medical records and memorizing medical textbooks with patient conversations? Where automatic analysis of daily measurements using conventional sensors such as individual’s voice patterns could potentially detect conditions like depression or Parkinson’s Disease? Where AI/ML models trained with data from best clinical practices helps guarantee equal access to care for patients independent of their treating healthcare facility? Where the guess work of which therapy is best for a given patient is replaced with algorithm-driven personalized recommendations? AI and ML hold the potential to do so.

To the average consumer, technical mechanisms make something like a self-driving car seem complicated. Opaque terms like “neural networks”3 and “deep learning” offer little clarity. But you don’t need a Ph.D. to grasp the basics of how AI/ML works. The concepts that underpin the technology are quite simple4.

Objectives of artificial intelligence

AI has two primary, simple objectives. First, AI aims to reproduce human capacity to perform specific tasks.  To achieve this objective, AI responds to stimuli and makes decisions in much the same way as humans do, mimicking human capacity for contemplation, judgment, and intention. Given a set of inputs or data points, AI relies on algorithms to continuously improve their observations and subsequent reactions. Second, AI aims to extend human capabilities by processing and identifying correlations in multi-channel/multi-modal data in which the human brain is not able to process. In a healthcare context, the second objective is critical to achieve personalized medicine.

Machine learning5 is a subset of AI that focuses on this iterative process — it helps to identify patterns in massive data sets to make algorithms more accurate. Quality, quantity, and diversity of data are essential for training of AI/ML algorithms.  The more robust the training data, the more precise the resulting output. Similarly, the more you use a social media site, the more tailored the ads you receive seem to get.

In healthcare, instilling a quantitative, computational approach through AI/ML can help create highly personalized insights across a wide spectrum of use cases – from predicting an individual’s risk for health events like atrial fibrillation via smartwatch data to more accurate detection of breast cancer via medical imaging scans. For patient care, AI/ML has the potential to improve disease prognostication, enable a therapy risk-benefit profile, and predict safety events before they occur. This could lead to faster and more accurate diagnosis and reduce care costs by helping inform streamlined treatment plans.

For clinical development, AI/ML can help optimize the entire research and development pipeline to shorten therapy time to market and improve patient treatment response by identifying populations of patients for whom a therapy may be particularly effective.

AI/ML in neurology and diagnostics

By harnessing computer models that simulate human intelligence, AI/ML can help us unlock long-standing mysteries about how our brains work. Already, AI is making new discoveries about how the brain processes information6. And the better we understand the intricacies of this mysterious organ, the better we can grasp how to treat neurological diseases like Multiple Sclerosis (MS), Alzheimer’s disease (AD), depression or Parkinson’s disease - conditions that currently affect hundreds of millions of people.

In the context of neurology, AI/ML could prove truly revolutionary.  Radiology is an area of healthcare with some of the greatest progress in applying AI/ML in clinical care and drug discovery for Neurology patients. Today, in clinical practice, Radiologists typically take a rather abstract qualitative approach through visual assessment of medical images to detect, monitor, and characterize diseases to support clinical evaluation. Rather than relying on the human eye, AI/ML can automatically provide quantitative assessments of medical image characteristics through pattern recognition.

Radiomics – where features are abstracted for medical images like MRIs, X-ray, and CT scans – use AI algorithms to uncover patterns characteristics such as intensity of signal, texture, shape, and interrelationships between pixels that are otherwise invisible to the human eye.  The output of radiomics holds potential to provide quantitative local disease characterization and sub-phenotyping across clinical conditions.  In collaboration with Radiologists, Neurologists can then identify anomalies linked to clinical signs at the onset of or progression of disease to make more informed decision-making for an individual patients’ care plan.  Potential use cases where AI/ML is applied to medical imaging are broad: for Multiple Sclerosis patients, there could be automatic identification and quantification of demyelination of lesions to aid clinical diagnosis. For Alzheimer’s Disease patients, detection of brain cell degeneration BDH is actively working to hone use cases for AI/ML that may be applied in clinical trials and clinical care. A recent partnership7 with TheraPanacea8, for instance, aims to leverage AI/ML analysis to create algorithmic solutions that improve patient care, advance personalized medicine, improve benefit-risk of treatments, and further our understanding of the underlying pathologies and heterogeneity of neurological diseases.

Paired with Biogen’s data, TheraPanacea’s AI/ML-based biomarker discovery capability has already led to some major discoveries in our understanding of MS – which like cancer - is not a single disease.  MS encompasses various sub-disease types that affect patients differently. Together, we’ve created an ensemble-based learning algorithm using single brain MRI scans. The AI can classify MS lesions as acute, chronic ‘inactive’ or slowly expanding based on a single MRI scan. The technology can even predict where new lesions may form and has potential to eventually lead to better personalized treatment selection.

Obviously, the impact on patients is significant, particularly because early detection, diagnosis, and enhanced disease phenotyping are key9 for a condition like MS.  For optimal patient outcomes, we must arm clinicians with personalized rather than disease-level information to inform clinical decision making.

The development with TheraPanacea is just one example of a future in which data-driven augmented clinical decisions can pave the way for patients to receive truly personalized care as early on in their condition’s progression as possible.

AI/ML healthcare advancements on the horizon

While the full potential of AI/ML has yet to be realized in the medical realm, the gears are now in motion to move beyond the hype for a more tech-fuelled future.

While there is much to accomplish, working across industry stakeholders will overcome current hurdles to remove bias, break down data siloes, and improve transparency.

At BDH, we view the adoption of AI/ML as a vital component of the future of digital medicine. But no one company can transform such a complex industry alone. Getting to the point of real, impactful and ubiquitous AI/ML in healthcare will require alliances and partnerships10. That’s why collaborations with companies like TheraPanacea are so critical and why we’re constantly on the lookout for the best companies to collaborate with talent to join our growing Biogen Digital Health team.

Beyond the hype and jargon that surrounds AI/ML, there’s real potential for this technology to completely transform our healthcare system with greater humanity - and help people live longer, healthier, and happier lives.

You may also be interested in


Biogen Digital Health

Leveraging the Full Value of Digital Biomarkers Requires Payer Buy-In


Biogen Digital Health

How passive monitoring through technology could enable better care


Biogen Digital Health

Why co-creation of digital health solutions with users is the future


Biogen Digital Health

What clinical trials of the future might look like


Biogen Digital Health

Digital vision: a conversation with Martin Dubuc


Biogen Digital Health

The digital health shift – a look at France’s public-private efforts