As hot topics go, artificial intelligence (AI) is up there with the most talked about advances in modern technology. In recent years, we have seen the approach transform many parts of our everyday lives, and now see the same techniques being applied to healthcare.
In multiple sclerosis (MS), for example, machine learning (ML) and deep learning (DL) techniques are being developed to aid diagnosis, inform prognosis, and even guide treatment plans. But what is AI, how can it help with MS management, and is it ready for the primetime of routine care?
AI and healthcare
While AI has become a ubiquitous term, used describe everything from robots to algorithms, it is a sprawling discipline. It can be defined as using computers to mimic human behaviour – with machine learning (ML) and deep learning (DL) being two techniques used to achieve that aim.
ML and DL models are particularly suited to healthcare. Their ability to quickly spot patterns and learn from patterns in large amounts of data means they can often reveal insights that would have otherwise remained invisible to the human eye.
Many believe the technology has the potential to help healthcare systems, which are increasingly being asked to do more with less, meet the challenges of rising demands. By automating routine tasks, the argument goes, ML and DL models can augment skilled professionals, freeing them up to focus on more value added tasks and patient care.
To date, these models have already been used to identify promising drug candidates [1] , to speed up microbiological testing [2], and to boost adherence to medication.
AI and MS
Imaging is an important area in which ML and DL are being put to work, and it is an application of particular interest to the MS healthcare community.
MRI has always played a role in MS care, but greater understanding of the disease, its course and treatments, as well as the modalities increased technical capabilities mean it is being used more and more. Now, MS teams are being asked to carry out more scans, requiring more time, resources, and expertise, than ever before. Numerous solutions, right the way across the care pathway, are currently in development.
They include a Convolutional Neural Network (CNN) DL model that can classify an MRI FLAIR image as being “with” or “without” white matter lesions, an important criteria in MS diagnosis [3] , with high accuracy and consistency. Another uses ML [4] to predict clinically defined MS using only baseline MRI lesion features derived from FLAIR and T1-weighted sequences. Research is also underway in areas such as phenotyping, where an ML algorithm has used diffusion tensor MRI data to discriminate between relapsing-remitting and progressive MS with an accuracy of between 60% and 92%. [5]
Is AI ready for routine MS care?
Such tools could help clinicians to make earlier, more specific diagnoses [6] , aid prognostication, and inform clinical decisions, and have demonstrated increasingly promising results. But according to a narrative review published last year, they still have major limitations.
One, they say, is the models’ inability to “show their workings”. “AI methods may detect subtle imaging abnormalities not detected by the human eye, which might reflect important pathophysiological mechanisms yet to be discovered. To the opposite, AI methods might follow unintended ‘shortcut’ strategies, which, while superficially successful, typically fail under slightly different circumstances,” write the authors.
The paper also cites a number of practical barriers. The availability of large datasets, for example, is a prerequisite to model development, yet obtaining the required volume and veracity of MRI scans is challenging due to a lack of system availability, cost constraints, and pathology-related variability.
“AI techniques are very promising for many clinical applications in the field of MS, including diagnosis, differential diagnosis, prognosis, disease and treatment monitoring, MRI protocol improvement, automated lesion and tissue segmentation.
“Future challenges are a better understanding of the information selected by AI algorithms, appropriate multi-centre and longitudinal validations of existing software, and practical aspects regarding hardware and software integration,” they conclude.
Learn about this topic at MSMilan2023
Scientific Session 3: Artificial intelligence in MS: From diagnosis to prediction and treatment monitoring.
Wednesday, 11 October 2023, 14:30 to 16:00 CEST
Talks include:
- AI, MS diagnosis and differential diagnosis
- AI, MS prognosis and monitoring
- Towards a more precise rating of neurological disability in multiple sclerosis: A new automatic and linear quantification of the neurological exam
- Predicting disease course in multiple sclerosis using deep learning on clinical data
- Prediction of progression independent of relapse activity at the first demyelinating attack with a deep learning image-based survival model
- Identification of cognitive phenotypes in paediatric multiple sclerosis using unsupervised machine learning
View the MSMilan2023 programme and secure your online or in-person spot at MSMilan2023 >> here.
[1] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10302890/
[2] https://www.europeanpharmaceuticalreview.com/article/166302/enhancing-rapid-microbiology-methods-how-ai-is-shaping-microbiology/
[3] https://www.sciencedirect.com/science/article/abs/pii/S2210537922000464
[4] https://www.sciencedirect.com/science/article/abs/pii/S1877750318305763
[5] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5078266/; https://pubmed.ncbi.nlm.nih.gov/31244599/
[6] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9163993/