Reduce MS surveillance MRI scan time

Purpose	Patients with Multiple Sclerosis (MS) undergo multiple surveillance MRIs, many with contrast, throughout their lifetime. Real-time AI should be able to reduce scan time without compromising necessary information for treatment.
Tag(s)	multiple sclerosis, MS, MRI, imaging, contrast, gadolinium
Panel	Neuroradiology Panel
Define-AI ID	19020007
Originator	Julie Bykowski
Lead	Julie Bykowski
Panel Chair	Alexander Norbash
Panel Reviewers	Neuroradiology Panel
License	Creative Commons 4.0
Status	Public Commenting
RadElement Set	RDES83

                

Clinical Implementation

Value Proposition

Real-time comparison ensuring no significant change between prior and current exam would allow total scan time to be curtailed and contrast avoided. Benefits of this automation include improving patient comfort, reducing contrast exposure, and freeing up scanner time for other patients.

Narrative(s)

62 yr old woman with MS has had 13 brain MRIs (11 with contrast) since her diagnosis 14 years ago. She has had stable disease on the past 4 exams, however continues surveillance while on disease modifying medication.

Workflow Description

At time of patient intake, most recent prior brain MRI for MS is pushed to system.
MS MRI protocol is initiated with localizer and volumetric FLAIR as 1st exam.
(If other sequences are needed for AI in this effort, they would then be performed next.)
Data is sent from the scanner to the AI processing site for comparison between exam points, while the remainder of the non-contrast portion of the current MRI is being completed. An alert message is sent back to the technologist (via the scanner platform or other vendor-neutral means for real-time alert), with the AI decision (below). A set of summary DICOM images (e.g. subtraction or heat map overlay) should be provided back to PACS to be archived with the current exam, clearly demarcating the two exam dates with the images as a verification of the AI decision for the Radiologist to review.

AI decision/alert:
(1) No change = no new lesion(s) are present; exam may be truncated without contrast
(2) Change = new lesions identified or increase in lesion size; current MRI exam would proceed with post-contrast images (site would have capacity to set threshold for change in size e.g. >10% volume)
(3) Indeterminate = likely artifact or co-registration error. Sites may have capacity/desire to have real-time Radiologist review if AI provides this alert, otherwise current MRI exam would proceed with post-contrast images.

The ability to have this be a real-time application is essential. This will not be as useful if the data transfer and AI analysis/reporting takes more than 5-7 minutes.

While the ability to segment supra- and infra-tentorial demyelinating lesions on volumetric FLAIR images may be helpful, the main concern is assessing change between exams.

Considerations for Dataset Development

Procedures	MRI brain without contrast
View(s)	volumetric FLAIR sequence; Ability to co-register MR images across scan types and time points.
Sex at Birth	{Male, Female}
Age	[0,100]
Comorbidities	“black holes” from MS (compared to demyelination), PML (progressive multifocal leukoencephalopathy), volume loss, chronic white matter/leukoaraiosis, midline shift
Tissue Involvement	cortical/juxtacortical, subcortical, periventricular, involving corpus callosum, infratentorial

Technical Specifications

Inputs

 

DICOM Study

Primary Outputs

Lesion Count

New Lesion Detection

Lesion Size

Lesion Extent Change

Secondary Outputs

Comparison Overlay

Future Development Ideas

• AI engine suggests artifacts or co-registration error

Related Datasets

No known related public datasets at this time,  please alert us if you know of any.