NeuroTech, an emerging industry at the crossroads of neuroscience and technology, holds promise for demystifying long-lingering questions about how the brain functions — and how human cognition and motion may degrade with aging and diseases. Artificial intelligence and machine learning (AI/ML) in particular, have an important role to play in this area of research. As recently reviewed by Yoshua Bengio et al.², research on artificial neural networks was motivated by the observation that human intelligence emerges from highly parallel networks of relatively simple, non-linear neurons that learn by adjusting the strengths of their connections. This observation leads to a central computational question: How is it possible for networks of this general kind to learn the complicated internal representations that are required for difficult tasks such as recognizing objects or understanding language?

The human brain likes to make empirical assumptions — it's how we move through the world without becoming overwhelmed by a constant onslaught of data. But when it comes to medical diagnoses, these assumptions can stand in the way of truly personalized care.

This is where AI/ML comes in. Let's take the example of a brain image analysis for a patient with Multiple Sclerosis (MS). Magnetic Resonance Imaging (MRI) is currently one of the primary vehicles for diagnosing and monitoring the progression of MS; it involves capturing 3D images of the brain's tissue through multiple acquisition modalities sensitive to distinct physical properties of the brain matter. Each high-resolution brain MRI image contains perhaps a million voxels (the 3D equivalent of a pixel). This means if you acquire for instance 6 different brain MRI sequences, there are around 6 million voxel-based data points in each person's scan.

Every voxel contains a certain grayscale in each acquisition sequence, which is quantifiable in signal intensity level. But the human eye cannot process voxel-level information (a voxel size being as tiny as 1 cubic millimeter) and seeks out lower-resolution boundaries, turning 3D high-resolution infinite grayscale data into gross arbitrary borders of 2D data. This means that when a neurologist sits down to analyze brain images and visually scan for lesions their findings are often purely empirical. Without the capacity to take in all of the millions of small continuous variations in the data, the human brain instinctually reduces the resolution — and therefore the quality/precision — of the information.

AI/ML enables to circumvent this issue and therefore improve an analysis by orders of magnitude. It can take in those millions of voxels each entailing continuous variables — and notice patterns — in a more comprehensive way. AI/ML allows researchers to identify areas of the brain that are "hiding" information from the naked human eye.

Much in the same way as each person has a unique way of thinking and making decisions, there is no single way that neurodegenerative disease manifests or progresses. It makes little sense to treat all patients as if their disease will follow a predetermined path presumably inferred from averaged population data.

There's now an evolving school of thought that certain neurodegenerative diseases have specific subject-level biomarker signatures or phenotypes — for example, gait and motor performance may be a critical biomarker of a disease like Parkinson's. Digitalizing these biomarkers by collecting them via connected sensors in active tasks and passive monitoring may enable earlier detection of disease onset, as well as alert patients when they should seek the advice of a healthcare professional. What's more, biomarker-driven phenotyping aided by AI/ML pattern recognition may enable researchers to model the heterogeneity of disease course and better understand them in the context of individual cases.

As one example of how AI/ML and deep learning may inform personalized, predictive analytics, Biogen collaborates with TheraPanacea, one of the world's preeminent healthcare companies using applied mathematics and AI in imaging and beyond for treatment implementation and prognosis. Combining TheraPanacea's AI/ML-based biomarker discovery and prognosis capabilities, and Biogen's unparalleled data and scientific expertise will focus on impacting clinical practice and R&D with transformational use of machine learning. As an example, in our collaboration to date, we have created an ensemble-based learning algorithm using single brain MRI scans to classify MS lesions - acute vs chronic ‘inactive’ or slowly expanding lesions - but also to predict where new lesions may form^3-5. The impact and applications can be tremendous and far-reaching for patients. With this development, we can envision a future where we can augment the "doctors eye" with patients’ unique “lesion fingerprints” that may lead to improved care through more personalized and data-driven clinical decisions.

Soon, AI/ML may enable us to create paradigms of research that could influence how drugs are developed — according to recognizable, quantifiable and predictable differences in the disease course of each patient.

The intersection of deep AI/ML capabilities and massive data repositories could improve our capacity to design more intelligent clinical trials, too. These technologies hold immense potential to optimize patient stratification — or the different "buckets" of risk level that allow healthcare providers to make more informed care management decisions. AI/ML may ultimately reduce clinical trial duration and sample size by enrolling better characterized study participants, allowing faster and more accurate readout, which in turn allows researchers to demonstrate a potential drug effect more rapidly.

While these technological advancements are certainly exciting, it's important to note that human input will remain a critical component of patient care. Machines are just that — machines — and while superior in their objectivity and capacity for parsing data, they have to be fed with expert domain knowledge and will always lack the human judgment and empathy that's a critical part of comprehensive and personalized healthcare.

That's not to say the professional landscape isn't evolving. In particular, there will be a rising need for mathematicians and computer scientists to specialize in medicine, and vice versa — for medical professionals and neurologists to develop competency in applies mathematics. The human brain is, after all, primarily a mathematical organ. Deciphering its elaborate code will involve cross disciplinary collaboration on a massive scale.

Passionate about digital health? Check out our open positions.