Daijiworld Media Network - New Delhi
New Delhi, Oct 25: Artificial intelligence (AI) is emerging as a promising tool for improving the diagnosis, monitoring and prediction of disease outcomes in autoimmune disorders affecting the central nervous system (CNS), according to a review of current scientific evidence.
The review examined the use of AI in multiple sclerosis (MS), neuromyelitis optica spectrum disorder (NMOSD), autoimmune encephalitis and myelin oligodendrocyte glycoprotein antibody-associated disease (MOGAD). These conditions often pose diagnostic and treatment challenges because of overlapping symptoms, limitations in available biomarkers and the need for long-term patient monitoring.
Researchers found that AI has been explored for several clinical applications, including automated analysis of brain imaging, differential diagnosis, disease classification, relapse prediction, assessment of prognosis and risk-based care planning.
The technology is also being used to analyse laboratory results, electronic health records, molecular datasets and multiple sources of patient information to support clinical assessments.
However, the amount of available research varies considerably across these conditions. Multiple sclerosis accounts for the largest body of published evidence, particularly in neuroimaging applications such as identifying and measuring lesions, monitoring disease activity, assessing brain volume changes and predicting disability.
Some AI-based imaging tools have progressed to multicentre studies and prospective evaluations within clinical workflows, indicating progress towards their potential use in routine healthcare.
In comparison, research involving neuromyelitis optica spectrum disorder, autoimmune encephalitis and MOGAD remains more limited, with studies generally involving smaller patient groups and varying research methods.
Although several studies have reported encouraging results in diagnostic accuracy and prediction of disease outcomes, the review highlighted important limitations. Many studies relied on retrospective data and lacked external validation, raising questions about whether their findings can be consistently reproduced in different patient populations and healthcare settings.
The researchers also pointed to the limited representation of patients with uncertain diagnoses, negative antibody tests and overlapping clinical conditions. This gap could restrict the usefulness of AI models in some of the most challenging real-world diagnostic situations.
Emerging areas of research include federated learning, digital biomarkers, integration of multiple types of biological data, foundation models and large language models. These approaches could potentially improve the analysis of complex patient information and support more personalised healthcare.
However, the review cautioned that evidence for these advanced applications remains exploratory. Their actual benefits in clinical practice have yet to be firmly established.
A major limitation identified by the researchers was the absence of prospectively validated AI models capable of comparing treatment effectiveness or selecting the most appropriate treatment for individual patients across the diseases reviewed.
Most existing models focus on diagnosis, disease monitoring and risk prediction rather than directly guiding treatment decisions.
The researchers emphasised the need for larger multicentre datasets, stronger external validation and prospective clinical studies to determine whether AI tools can improve medical decision-making and patient outcomes in everyday practice.
In the near term, progress is expected to come primarily from combining different types of patient data, standardising imaging analysis and improving risk assessment. However, the researchers stressed that current evidence does not yet support the use of AI for autonomous diagnosis or treatment selection in these complex neurological conditions.