Grayscale brain MRI, CNN architecture schematic, and colorful activation map

Deep Learning for Neuroimaging Analysis

  • by Youngkyoo Jung
Grayscale brain MRI, CNN architecture schematic, and colorful activation map
The structure of (A) standard CNN and (B) H-CNN that estimate CBF and ATT maps with reduced input data. [from D Kim, et al., Magn Reson Med, 2024]

The IRIS Lab actively develops and applies deep learning methods to solve challenging problems in neuroimaging, from image reconstruction to anatomical segmentation and parametric mapping. Notable contributions include a 3D convolutional neural network (CNN) that simultaneously estimates CBF and ATT from multi-delay ASL data, bypassing traditional fitting approaches that are sensitive to noise. 

Grid of nine brain MRI slices with red brain tissue segmentations labeled AFN, FSL, Nobrainer
Brain segmentation of a rhesus macaques using different algorithms. Comparison between brain masks obtained from AFNI (left), FSL (middle), and machine learning based Nobrainer (right) [A Zhang, et al., ISMRM 2022]

Additionally, the lab has developed a U-Net-based framework for automated brain extraction from T1-weighted MRI of non-human primates (rhesus macaques), addressing the scarcity of tools tailored to non-human primate neuroimaging. These AI-driven pipelines accelerate processing speed, improve robustness under challenging signal conditions, and open pathways for integrating advanced quantitative MRI into routine clinical workflows. 

Primary Category

Machine Learning

Secondary Categories

Animal ASL Human MRI Structural Imaging