Related Experiment Video
Updated: Jul 21, 2025
High-resolution Confocal Imaging of the Blood-brain Barrier: Imaging, 3D Reconstruction, and Quantification of Transcytosis
Published on: November 16, 2017
Improving measurement of blood-brain barrier permeability with reduced scan time using deep-learning-derived
Jonghyun Bae1, Chenyang Li2, Arjun Masurkar3
1Vilcek Institute of Graduate Biomedical Science, New York University School of Medicine; Center for Biomedical Imaging, Radiology, New York University School of Medicine; Center for Advanced Imaging Innovation and Research, Radiology, New York University School of Medicine; Department of Radiology, Weill Cornell Medical College.
A deep learning network estimates capillary input function (CIF) for accurate blood-brain barrier (BBB) permeability assessment. This method reduces scan time and uncertainty, enabling detection of age-related BBB changes.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence in Medicine
Background:
- Dynamic contrast-enhanced MRI (DCE-MRI) is crucial for kinetic parameter estimation.
- Arterial Input Function (AIF) introduces uncertainty in DCE-MRI parameter estimation.
- Accurate assessment of blood-brain barrier (BBB) permeability is vital for diagnosing neurological conditions.
Purpose of the Study:
- To evaluate a deep learning network for estimating Capillary Input Function (CIF) in DCE-MRI.
- To assess the feasibility of using CIF to estimate BBB permeability with reduced scan times.
- To minimize uncertainty associated with AIF in kinetic parameter estimation.
Main Methods:
- Developed and trained a deep learning network to predict CIF from DCE-MRI data.
- Simulated reduced scan times by truncating 25-min scans to 10 min.
- Compared BBB permeability (PS) estimates using AIF (Ca) and CIF (Cp) with reference values.
Main Results:
- The CIF-based method (Cp-10min) reduced PS overestimation by 81% compared to the AIF-based method (Ca-10min).
- Bland Altman analysis showed significantly improved agreement with reference values using CIF (mean difference 1.63 ± 2.25 x10-4 min-1).
- Detected significant age-related increases in BBB permeability (75% gray matter, 35% white matter).
Conclusions:
- Deep learning enables feasible CIF estimation for DCE-MRI.
- This approach accurately assesses BBB permeability with reduced scan times.
- The method automatically selects CIF, reducing user-dependent uncertainty and improving age-related change detection.
