Nuclear Techniques ›› 2019, Vol. 42 ›› Issue (10): 100301-100301.doi: 10.11889/j.0253-3219.2019.hjs.42.100301


Research on atlas-based brain segmentation for HD integrated PET/MRI image

Zaisheng LI1,2,3,Lingzhi HU4,Qun CHEN1,4()   

  1. 1. School of Information Science and Technology, ShanghaiTech University, Shanghai 201210, China
    2. Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, Shanghai 200050, China
    3. University of Chinese Academy of Sciences, Beijing 100049, China
    4. Shanghai United Imaging Healthcare Co. , Ltd. , Shanghai 201807, China
  • Received:2019-08-12 Revised:2019-09-17 Online:2019-10-13 Published:2019-10-16
  • Contact: Qun CHEN
  • About author:LI Zaisheng, male, born in 1994, graduated from Nanjing University of Aeronautics and Astronautics in 2016, master student, focusing on medical imaging processing
  • Supported by:
    National Key Research and Development Program Digital Diagnostic Equipment R&D Pilot(2016YFC0103900)

Abstract: Background

Integrated positron emission tomography (PET) / magnetic resonance imaging (MRI) is a multimodal imaging system which can acquire PET and MRI images simultaneously. Due to the unique advantages in reflecting the anatomical structure and physiological function, PET/MRI has been widely commonly used in diagnosis of many brain diseases. Brain segmentation is of great significance to the quantitative study of brain images and the common method used in clinical diagnosis is atlas-based brain segmentation, which can be applied to both MRI images and PET images. Brain segmentation based on atlas for integrated PET/MRI image system only needs one modelity as the segmentation results can be mapped to another modelity.


This study aims to determine which modality should be used for atlas-based segmentation for high definition (HD) integrated PET/MRI image.


Comparative experiments with two image groups were designed and performed. In the first group, 150 PET images were registered to PET brain template to obtain brain segmentation. In the second group, 150 MRI images were registered to MRI brain template to obtain brain segmentation. Six regions were selected to calculate the dice value. Comparing the segmentation results of the two sets of images with their results based on first, the modality with a higher average dice value would be more suitable for atlas-based brain segmentation.


In PET group, the dice values of six regions were 0.62, 0.55, 0.63, 0.62, 0.71 and 0.69, respectively. In MRI group, the dice values of six regions were 0.68, 0.64, 0.79, 0.81, 0.77 and 0.79. The error coming from registering PET images to PET brain template was larger than that of registering MRI images to MRI brain template.


PET-based segmentation accuracy is lower than MRI-based segmentation precision, hence the MRI image is more suitable for atlas-based brain segmentation.

Key words: PET/MRI, Registration, Segmentation, Brain template, Brain atlas

CLC Number: 

  • R445