Multi-center Study of Artificial Intelligence Model for Gadolinium-based Contrast Agent Reduction in Brain MRI (MAGNET)

Brain Diseases Clinical Trial

— MAGNET  

Official title:

Multi-center and Prospective Cohort Study of Artificial Intelligence Model for Gadolinium-based Contrast Agent Reduction in Brain MRI (MAGNET)

Clinical Trial Details — Status: Recruiting

Administrative data

• NCT number: NCT05754476
• Other study ID #: KY-2021-184-04
• Status: Recruiting
• Start date: March 29, 2019
• Est. completion date: December 31, 2024

Study information

• Verified date: March 2023
• Source: Beijing Tiantan Hospital
• Contact: Yaou Liu, PhD
• Phone: +86 1059975396
• Email: yaouliu80[@]163.com
• Is FDA regulated: No
• Study type: Observational

Clinical Trial Summary

MAGNET is a multi-center and prospective study to minimize Gadolinium-based Contrast Agent (GBCA) combining novel artificial intelligence (AI) methods with pre-contrast images and/or low-dose images to synthesize virtual contrast-enhanced T1 (vir-T1c) images, based on a large clinical and MRI database and subsequently validated for its clinical value. MRI examinations for patients included T1-weighted images (T1WI) before and after contrast agent administration and at two dose levels: low-dose (10% or 25%) and full-dose (100%), T2-weighted images (T2WI), fluid-attenuated inversion recovery (FLAIR), and diffusion-weighted imaging sequences (DWI) and the computed apparent diffusion coefficient (ADC), all either acquired three dimensional [3D] or two dimensional [2D]). The standard dose of intravenous gadolinium contrast agent was 0.1mmol/kg(body weight) by manual injection or automatic injection with a high-pressure syringe at a flow rate of 4mL/s.The sequence parameters used for the 3DT1WI scans must be consistent, and the standard for intravenous injection of gadolinium contrast agent is 0.1mmol/kg (body weight), administered either manually or automatically with a high-pressure syringe at a rate of 4mL/s.

Additionally, arterial spin labeling (ASL), amide-proton transfer chemical exchange saturation transfer (APT-CEST), susceptibility-weighted imaging (SWI), or quantitative susceptibility mapping (QSM) can be acquired at the same time if the conditions permit.

Description:

MRI with GBCA is an indispensable part of imaging exams for brain disease diagnosis. Generally, GBCA is safe, with a few mild side effects since GBCAs received FDA approval in 1989. There are numerous issues that challenge the current practice of widespread use of GBCA. GBCA can trigger nephrogenic systemic fibrosis(NSF) under particular circumstances, cause allergic reactions, may increase the risk of fetal death, and accumulate in the brain such as the dentate nucleus and globus pallidus. Efforts need to be made to reduce dose while still maintaining diagnostic capabilities. Artificial intelligence (AI) techniques have shown great potential in medical fields. Deep learning (DL), a branch of AI, has been applied to image segmentation, computer-aided diagnosis, and reduce GBCA dose.

This study intends to build a prospective brain MRI dataset including patients with suspected or known brain abnormalities to minimize the use of GBCA. Then train DL models to process pre-contrast images and/or low-dose T1 images to predict virtual contrast-enhanced T1 (vir-T1c) images, taking the full-dose images as the reference standard. Later quantitatively and qualitatively evaluating and comparing vir-T1c images from DL models about clinical diagnostic performance, focusing on lesion detection, diagnosis, and therapy, to explore a DL model universal, provide enhanced images faster and more convenient in clinical practice. To minimize the use of GBCA, we will:

1. Use novel artificial intelligence (AI) methods with pre-contrast images including conventional (T1WI, T2WI, FLAIR, DWI/ADC), new physiological MRI techniques (ASL, APT-CEST, SWI/QSM) by adding physiological information from perfusion as well as metabolic and susceptibility imaging, and/or low-dose images (10% or 25%) to synthesize vir-T1c images;

2. Quantify when (in which patients and at what follow-up times) GBCA can be omitted or minimized without influencing brain disease diagnosis and treatment evaluation for doctor raters and therefore patient prognosis.

This study does not limit manufacturers including 1.5T and 3.0T scanners， or kinds of GBCAs.

Recruitment information / eligibility

• Status: Recruiting
• Enrollment: 3000
• Est. completion date: December 31, 2024
• Est. primary completion date: December 31, 2023
• Accepts healthy volunteers: Accepts Healthy Volunteers
• Gender: All
• Age group: N/A and older

Eligibility

Inclusion Criteria:

1. Patients with suspected or known brain diseases including tumors, vascular disorders, inflammatory diseases, neurodegenerative diseases and trauma, follow-up, routine brain, and others requiring MRI exams with GBCAs.

2. Informed written consent obtained from the patient, and/or patient's parent(s), and/or legal representative.

Exclusion Criteria:

1. Patients with contraindications to MR examination.

2. Patients with incomplete MRI data and obvious image artifacts.

Related Conditions & MeSH terms

• Brain Diseases

Intervention

Other:
• Low-dose GBCA levels
MRI examinations for patients at two dose levels: low-dose (10% or 25%)can be chosen.

Locations

Country	Name	City	State
China	Beijing Tiantan Hospital	Beijing	Beijing

Sponsors (1)

Lead Sponsor	Collaborator
Beijing Tiantan Hospital

Country where clinical trial is conducted

China, 

References & Publications (3)

Gong E, Pauly JM, Wintermark M, Zaharchuk G. Deep learning enables reduced gadolinium dose for contrast-enhanced brain MRI. J Magn Reson Imaging. 2018 Aug;48(2):330-340. doi: 10.1002/jmri.25970. Epub 2018 Feb 13. — View Citation

Jayachandran Preetha C, Meredig H, Brugnara G, Mahmutoglu MA, Foltyn M, Isensee F, Kessler T, Pfluger I, Schell M, Neuberger U, Petersen J, Wick A, Heiland S, Debus J, Platten M, Idbaih A, Brandes AA, Winkler F, van den Bent MJ, Nabors B, Stupp R, Maier-Hein KH, Gorlia T, Tonn JC, Weller M, Wick W, Bendszus M, Vollmuth P. Deep-learning-based synthesis of post-contrast T1-weighted MRI for tumour response assessment in neuro-oncology: a multicentre, retrospective cohort study. Lancet Digit Health. 2021 Dec;3(12):e784-e794. doi: 10.1016/S2589-7500(21)00205-3. Epub 2021 Oct 20. — View Citation

Luo H, Zhang T, Gong NJ, Tamir J, Venkata SP, Xu C, Duan Y, Zhou T, Zhou F, Zaharchuk G, Xue J, Liu Y. Deep learning-based methods may minimize GBCA dosage in brain MRI. Eur Radiol. 2021 Sep;31(9):6419-6428. doi: 10.1007/s00330-021-07848-3. Epub 2021 Mar 18. — View Citation

Outcome

Type	Measure	Description	Time frame	Safety issue
Primary	quantitative metrics	To quantitatively describe the discrepancies between the vir-T1c and the full-dose images by measuring the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). The PSNR measures the voxelwise difference and the PSNR range is between -1 and 1. The SSIM compares nonlocal structural similarity and the minimum value of PSNR is 0. The metrics will be reported in separate(e.g.,SSIM, 0.90; PSNR,42 in vir-T1c, SSIM, 0.94; PSNR,45 in full-dose images).	after training and applying of the proposed deep learning model, an average of 1 year
Primary	qualitative assessments	To qualitatively describe discrepancies between the vir-T1c and full-dose images by evaluating enhancement degree, homogeneity, and pattern.; Firstly, zero indicates no intracranial or non-enhancing lesion. For enhancement degree, 1 indicates mild enhancement, 2 indicates moderate enhancement, and 3 indicates clear enhancement.; For enhancement homogeneity, 1 indicates heterogeneous enhancement, 2 indicates mildly heterogeneous enhancement, and 3 indicates homogeneous enhancement.; For enhancement pattern, 1 indicates mass enhancement(proportion enhancement more than 50%), 2 indicates nodular enhancement (proportion enhancement less than or equal to 50%), 3 indicates ring enhancement, 4 indicates linear enhancement, and 5 indicates other enhancement.	after training and applying of the proposed deep learning model, an average of 15 months
Secondary	clinical effects	To describe whether vir-T1c images combing other sequences affect diagnosis or treatment according to evaluation of neuroradiologist and neurologist from 1 to 5 scores.; Zero indicates enhancement error and can not be used. 1 indicates non-diagnostic. 2 indicates affecting diagnosis or treatment significantly. 3 indicates affecting diagnosis or treatment moderately. 4 indicates no affecting diagnosis or treatment almost. 5 indicates no affecting diagnosis or treatment completely.	after training and applying of the proposed deep learning model, an average of 18 month