Photo credit: www.sciencedaily.com
Breakthrough AI Tool Enhances Understanding of Multiple Sclerosis Treatment
Researchers at UCL have developed an innovative artificial intelligence (AI) tool designed to interpret and evaluate the efficacy of treatments for multiple sclerosis (MS), a condition affecting many in the UK.
This AI leverages mathematical models to analyze large datasets, enabling the computer to learn and tackle problems reminiscent of human capabilities, including intricate tasks such as image recognition.
The tool, named MindGlide, is capable of extracting crucial insights from MRI scans of the brain taken during the treatment of MS patients. It measures areas of damage and identifies subtle changes in the brain, such as shrinkage and plaque formation.
Multiple sclerosis is characterized by the immune system attacking the brain and spinal cord, leading to difficulties in movement, sensation, and cognitive function. In the UK alone, approximately 130,000 individuals are diagnosed with MS, which incurs an annual cost of over £2.9 billion to the National Health Service (NHS).
The measurement of MRI markers is essential for the investigation and evaluation of MS treatments. However, the need for specialized MRI scans for accurate assessment often restricts the utility of standard hospital imaging.
In a recent study published in Nature Communications, researchers evaluated MindGlide using over 14,000 images from more than 1,000 MS patients. Traditionally, this analysis required the expertise of neuro-radiologists to interpret complex scan results, often leading to weeks of delays in reporting due to heavy NHS workloads.
Significantly, MindGlide demonstrated the capability to identify how various treatments influenced the progression of the disease in both clinical trials and everyday medical practice. Remarkably, it processed each image in just five to ten seconds.
In performance comparisons, MindGlide outperformed two existing AI tools—SAMSEG, which focuses on outlining different brain components in MRI images, and WMH-SynthSeg, designed to detect bright spots in MRI scans indicative of MS. Its superiority was quantified as being 60% more effective than SAMSEG and 20% more effective than WMH-SynthSeg in locating plaques and monitoring treatment effects.
Dr. Philipp Goebl, a leading author from the UCL Queen Square Institute of Neurology, remarked, “With MindGlide, we are now able to utilize existing brain images stored in hospital records to deepen our understanding of multiple sclerosis and assess the impact of treatments on the brain.”
He emphasized the potential of the tool to reveal invaluable insights from millions of previously challenging brain images, anticipating that it will enable a more nuanced understanding of a patient’s condition within the next five to ten years.
The study outcomes affirm that MindGlide can effectively recognize and quantify significant brain tissues and lesions despite limitations in the MRI data, utilizing single scan types typically not employed for such assessments, like T2-weighted MRI scans lacking FLAIR technology.
Beyond its proficiency in detecting superficial brain alterations, MindGlide also managed to accurately analyze deeper brain regions. The findings presented were robust and reliable, both in isolated instances and over extended monitoring periods, such as annual patient scans.
Moreover, MindGlide verified prior high-quality research on the effectiveness of specific treatments, paving the way for its application to assess MS treatments in real-world scenarios while addressing the limitations of exclusive reliance on high-quality clinical trial outcomes that may not reflect the full spectrum of MS patient experiences.
Dr. Arman Eshaghi, the principal investigator and leader of the MS-PINPOINT group, emphasized, “Previously, we have neglected to analyze the vast majority of clinical brain images due to their perceived lower quality. AI, particularly through MindGlide, has the potential to unleash the hidden value of hospital records, offering unprecedented insights into the progression of MS and the ways treatments impact the brain.”
Study Limitations
The current version of MindGlide is specialized for brain scans and does not encompass spinal cord imaging, which is crucial for assessing disability levels in MS patients. Future investigations will need to broaden the focus to include comprehensive assessments involving both the brain and spinal cord.
Development of MindGlide
The development of MindGlide involved deep learning techniques applied to MRI images to pinpoint damage and alterations caused by MS. Initial training utilized a dataset of 4,247 MRI scans collected from 2,934 MS patients across 592 MRI scanners. Validation was conducted against three independent databases comprising 14,952 images from 1,001 patients.
Source
www.sciencedaily.com