A Real-World Clinical Validation for AI-based MRI Monitoring in Multiple Sclerosis

In the realm of modern multiple sclerosis (MS) management, the primary objective is achieving No Evidence of Disease Activity (NEDA). This encompasses the absence of clinical relapses, magnetic resonance imaging (MRI) indications of disease activity, and any progression in disability. While MRI serves as a crucial diagnostic tool for neurologists to monitor silent MS disease activity and determine treatment adjustments as needed, conventional radiology reports often fall short in providing comprehensive insights, as they tend to be qualitative and may not always detect emerging or expanding lesions accurately.

Moreover, the available quantitative neuroimaging tools lack robust clinical validation. In a study involving 397 multi-center MRI scan pairs conducted in routine clinical practice, we have demonstrated the remarkable sensitivity of an AI-based tool seamlessly integrated into the clinical workflow, outperforming standard radiology reports. The AI tool achieved a case-level sensitivity of 93.3% as opposed to the 58.3% sensitivity of traditional reports, as verified against a consensus ground truth, while maintaining a high level of specificity.

Additionally, we have established the equivalency of this AI tool with a core clinical trial imaging laboratory for assessing lesion activity and quantitative measures of brain volume. This includes the crucial biomarker of neurodegeneration in MS, known as percentage brain volume loss (PBVC). The results showed an impressive alignment between the AI tool and the clinical lab, with mean PBVC values of -0.32% and -0.36%, respectively. Even in cases of severe brain atrophy (exceeding 0.8% loss), which often goes unnoticed in traditional radiology reports, the AI tool proved its mettle.

To further enhance its utility, the AI tool also offers a valuable comparative feature, presenting MS patients with a relevant centile ranking based on lesion burden. In our cohort, this revealed inconsistencies in the qualitative descriptors used in radiology reports, emphasizing the significance of adopting AI-based image quantification.

In summary, the incorporation of AI-based image quantification significantly augments the accuracy and value of qualitative radiology reporting in the management of MS. Scaling the deployment of these tools promises to usher in a new era of precision management for MS patients.


Source Medrxicv

Author: Neurologica