The image features a research poster from Nanyang Technological University, specifically from the Rapid-Rich Object Search (ROSE) Lab. The poster details a study focused on Advanced MRI acquisition techniques. It introduces a new method called Progressive Divide-and-Conquer (PDAC) for reconstructing clean MRI images from their undersampled k-space measurements. 

The "Background" section explains the goals and challenges of reconstructing MRI images using a Deep Unfolding Network (DUN) approach, highlighting the difficulty in recovering missing data during each iteration. The "Motivation & Contributions" section compares the proposed PDAC method with existing reconstruction methods, demonstrating its superiority in image quality through visual iterations.

An illustrated flowchart under "Progressive Divide-and-Conquer (PDAC)" explains the iterative process of the PDAC method, detailing components such as Severity Conditioning, Degradation Prediction, and Predicting Mask. The methodology includes multiple iterations that involve data consistency, severity embedding, and degradation prediction to progressively refine the MRI image reconstruction, ultimately aiming for a high-quality final image.
Text transcribed from the image:
NANYANG
TECHNOLOGICAL
UNIVERSITY
SINGAPORE
ROSE
Rapid-Rich Object Search Lab
博云搜索实验室
Progressive Divide-
Background
Accelerated MRI acquisition:
y= Ax+e=DFx + €
⚫y: zero-filled undersampled measurements in k-space
• x: underlying spatial MR images
• D: zero-filled subsampling matrix
F: 2-D Fourier transform, €: measurement noise
Goal: Reconstruct clean MR images from its undersampled
k-space measurements.
Challenges: Existing DUNS recover the missing data in the
entire null space during each iteration, which could be chal-
lenging when dealing with highly ill-posed degradation.
Progressive Divide-and-Conquer (PDAC)
Regularization on each decomposed degradation:
Motivation & Contributions
Comparsion with existing methods:
(a) Existing Deep Unfolding Network
Undersampled
k-space
Iteration t
(b) Progressive Divide-and-Conquer (Ours)
Iteration t+1
Final
reconstructio
Undersampled
k-space
Iteration t
Iteration t+1
Final
reconstruction
Solving a severe MRI reconstruction problem with a single regularizer can be challenging when the measurement y
introducing a series of regularizers according to the decomposed degradation, as: mina ||y – Aox|| +1 \t¥t (At
z0 = y
Y
ข
ZT
Y
Zt-1
PDAC
PDAC
PDAC
Mo
(Mt-
Mt
MT
Reconstructe
image
PDAC Iteration t
Zt-1
Y
Not
Zt
Data
Consistency
Severity Conditioning
Pt-1
Zt
Network Not
Pot
Degradation
Prediction
Pe.
Degradation Predicto
Ee.
Pt
Eet
Degradation Severity
Embedding
Predicting
Mask
Mt
Mt-1
Severity Conditioning: During each iteration of PDAC,
the network aims to reconstruct an intermediate mea-
surement from a specific degradation severity, i.e., zt E
R(AH). The degradation pattern m+ Opt in the k-space
is embedded via a severity conditioning module Eet.
Element-wise
Multiplication
Degradation Prediction: A degradation predictor Pet
duced to estimate a probability pt indicating the location
quency columns to sample, i.e., pt = Pot (t). Thus, the
be obtained via on the previous mask mt-1 according to
[i]
mt
[i]
= m 1+I(i Є idx) Vi.