introduction results conclusions references methods

3 downloads 0 Views 701KB Size Report
Intravoxel incoherent motion (IVIM) is an emerging non- invasive diffusion-weighted MRI technique to quantify placental perfusion [1]. Since the fraction of ...
Semi-automatic segmentation of the placenta into fetal and maternal compartments using intravoxel incoherent motion MRI Wonsang You, Nickie Andescavage, Zungho Zun, Catherine Limperopoulos Developing Brain Research Laboratory, Children’s National Medical Center

INTRODUCTION

Placental segmentation using GrowCut

RESULTS

1. Initialization • Initial regions of fetal and maternal compartments were

Intravoxel incoherent motion (IVIM) is an emerging non-

invasive diffusion-weighted MRI technique to quantify placental perfusion [1]. Since the fraction of moving blood in the total volume -so called perfusion fraction- is known to be heterogeneous over the highly complex structure of placenta,

manually defined. 30% voxels of those regions were

randomly chosen and used as seed voxels. • The state of each voxel was represented as (L,W,F) consisting of label L, strength W, and feature vector F. L

{‘unlabeled’,’fetal’,’maternal’}

the accurate placental segmentation into fetal and maternal

W

Initialized to be 0.3

compartments is important to analyze regional differences in

F

Given as S0 intensity

As shown in Fig 2, the initial fetal compartment was roughly defined based on T2-weighted image, and improved using the GrowCut. Segmented fetal and maternal compartments occupied 31.5±7.0% and 68.2±7.0% of the placenta. As shown in Fig 3, the proposed method was effective in removing non-placental tissues from the seed region.

placental perfusion fraction. We first propose a semi-automatic method for segmenting the placenta into fetal and maternal compartments from IVIM data, using the GrowCut algorithm [2].

Fig 2. The semi-automatic process of placental segmentation

METHODS

CONCLUSIONS

Fig 1. Evolution of cell states through GrowCut algorithm

Semi-automatic placental segmentation was achieved using 2. State transition by iterative competition

Image acquisition and preprocessing

• Rule 1: The label of each seed voxel is ‘conquered’,

DWI Data were acquired from 16 healthy pregnant women

and changed into that of the neighbor voxel with higher

between 21-37 weeks of gestation. Echo planar imaging (EPI)

strength.

scanner.

on a larger dataset, our preliminary results show that the

• Rule 2: The voxel state becomes less vulnerable to the

differences in perfusion between fetal and maternal compartments of the placenta using diffusion-weighted MRI.

attack of neighborhood as its intensity is more hetero-

Item

Value

TR

8000 ms

TE

53.8 ms

FOV

420x420 mm2

Data matrix

96x96

#Slices

40-50

Thickness

regions. Although the tool needs to be improved and validated

proposed method is a promising tool to study functional

sequences were obtained with pulsed gradients of b-values

{0,25,50,114,243,500,543,800,900} sec/mm2 on 1.5T GE

the GrowCut algorithm given the manually created seed

4 mm

geneous compared to nearby voxels

REFERENCES

• Each seed region grew or shrank to encompass fetal or maternal compartment until convergence.

B0 volume

1. Moore et al. Placenta 21(7), 726-732 (2000). 2. Vezhnevets et al. Graphicon, 150-156 (2005). 3. Klein et al. IEEE Trans. Med. Imaging 29(1), 196-205 (2010).

Seed region including non-placental tissue

Segmented region excluding non-placental tissue

After the whole placenta mask was manually created, nonrigid image registration was performed using Elaxtix to correct placental motion between volumes [3]. The IVIM intensity was modeled as follows.

D D*

Diffusion coefficient Pseudo-diffusion coefficient

F Perfusion fraction S0 Signal intensity at b-value=0

Fig 3. The effect of GrowCut on removing non-placental tissues from segmentation.

Copyright © 2017 Wonsang You

Contact : [email protected]

Suggest Documents