Why might those living with multiple sclerosis benefit from AI support?
Artificial intelligence (AI) seems to be everywhere at the moment, leading us all to wonder about its potential benefits, including within the realm of healthcare. While it might seem like there’s a long way to go before AI can support us in managing medical conditions, I have discovered that there are already some clear ways it may be able to help those living with multiple sclerosis (MS).
MS is an autoimmune disease, affecting around 2.9 million people worldwide.1 In MS, the immune system attacks the myelin sheath on nerve fibres in the central nervous system. This myelin sheath acts like insulation on an electrical wire, so damaging it disrupts nerve communication.2 This commonly causes fatigue, movement and coordination issues, cognition and emotional changes, and visual problems2,3. However, everyone experiences a different set of symptoms, depending on where in the brain or spinal cord communication is disrupted.3,4 As well as varying between people, MS varies within the same person over time, and this can be unpredictable.4
This heterogeneity, alongside the volume of data generated through diagnosis and monitoring, makes MS a prime example of a condition that could benefit from AI support. There is no ‘one size fits all’ approach to managing MS; each person has a unique experience living with this condition and will benefit from a personalised approach that can adapt with their MS over time. AI could be a useful tool in this – through classifying diagnoses, streamlining monitoring, and generating prognostic predictions.
How could AI help diagnose MS?
Diagnosing MS can be complicated, as there are many other conditions to be ruled out, requiring several tests and examinations.5,6 To help with this differentiation, the pattern recognition and classification abilities of machine learning algorithms can be applied to magnetic resonance imaging (MRI) brain scans, or quantitative blood biomarker data, to detect diagnostic patterns.7 This can also be extended to characterise MS by subtype.7 This would decrease the manual analysis workload, freeing up more of clinicians’ time to spend with their patients. If trained on the right data, machine learning models may even be able to detect MS risk earlier than current diagnostic processes. One model was trained to predict risk for MS based on blood markers up to three years before diagnosis, and achieved this with consistently high accuracy across those three years.8
How might AI streamline MS monitoring?
People with MS are monitored over time, to assess myelin damage and any changes in areas such as movement or cognition. Clinicians use MRI to assess myelin damage and AI can be applied here to help quantify and analyse these scans.7,9 This could improve sensitivity for detecting new or expanding areas of damage.10 Increasingly, movement can be measured at home via wearable devices, such as smartwatches. Machine learning algorithms can classify this data (e.g., as running, walking, or standing), and subclassify by gait – reducing the manual analysis required.7 A classifying algorithm, applied to data from people with different levels of movement disability, has been used to pick out movement features that are most predictive of fall risk.7 This helps clinicians prioritise what to monitor – with further understanding, such features could alert doctors to when someone might have increased fall risk, and help them access the right support as early as possible. Researchers have also explored the use of machine learning in cognitive assessments, for example to rate peoples’ drawings and identify optical recognition deficits, in a digitised version of a visuospatial memory test.11
Could AI help predict this unpredictable disease?
Several factors are considered when planning MS treatment, including genetics, health, and lifestyle factors, alongside clinical measurements, MRI scans, and blood tests, and of course patient preferences.6,7 Clinicians quantify many of these factors as part of diagnosis and monitoring, generating a huge volume of multidimensional data to be factored into care planning decisions. There have already been several initial machine learning approaches to predicting MS prognosis, with a view to supporting treatment decisions.6,7 One approach identified predictive subtypes based on MRI data, while another project made predictions based on demographic data and clinical scores.12,13
An aspirational suggestion is to generate ‘digital twins’ for individuals with MS. Truly holistic and personalised, this would be a virtual representation of the individual, based on their own data, and on a large, representative, MS data set. This digital twin would be updated constantly using both data sources, to continuously model the person in almost real time and provide data-based recommendations for care and treatment.6
Too good to be true?
There are many challenges to overcome before the full potential of these tools can be realised. For example, large quantities of standardised data must be collected, and it must be determined which measurements are essential. Training datasets must be representative of all the people they aim to benefit, minimising the risk of bias and inequality. We will also need to establish processes to manage data access, consent, ownership, privacy, and security.6
Modelling prognosis in this way also comes with ethical questions. For instance, if a model predicts that someone’s mobility will rapidly deteriorate, how do we tell them? And do they have a right to not know? How do we ensure that patients’ preferences and clinicians’ judgements still lead care planning, when there are algorithms and data that suggest ‘optimal’ approaches, yet without human caution and sensitivity?6
Conclusion
All of these AI applications in MS come with challenges, from the practicalities of large-scale data sharing to transparency and trust of ‘black box’ algorithms that we may struggle to understand. If we can overcome these challenges, and validate AI processes in real-world settings, their integration into clinical practice could have huge benefits for people with MS, as well as for clinicians.
AI algorithms could support early, consistent diagnosis and classification; streamline monitoring and analysis; and even generate personalised models to predict MS progression, helping to inform personalised care planning to benefit all individuals with their unique experiences of MS. At TVF, we are exploring how AI tools can support us in our work, to ultimately benefit people living with a range of conditions, from cancers to autoimmune diseases.
By Sophie Scott
REFERENCES
