Brain tumor segmentation with multimodal magnetic resonance imaging

Wang, Bo

Permalink

https://hdl.handle.net/2142/110723 Copy

Description

Title

Brain tumor segmentation with multimodal magnetic resonance imaging

Author(s)

Wang, Bo

Issue Date

2021-04-26

Director of Research (if dissertation) or Advisor (if thesis)

Liang, Zhi-Pei

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Date of Ingest

2021-09-17T02:34:43Z

Keyword(s)

• MRI
• image segmentation
• brain tumor
• CNN
• deep-learning

Abstract

A brain tumor is an abnormal cell population that occurs in the brain. Identifying the abnormal region is one of the significant steps for patients with a brain tumor. However, manual labeling is subjective and time-consuming. For the patients to get an accurate diagnosis and timely surgery, an accurate and efficient automatic brain tumor detection and segmentation algorithm is necessary. For medical images, with limited data and a high requirement for accuracy, the traditional learning algorithms are currently insufficient. We follow the idea of most state-of-the-art approaches with multimodal brain tumor image segmentation, taking advantage of different modalities from magnetic resonance imaging (MRI) which can provide different texture features of the same brain image sample. However, most open-source brain tumor MRI images are recorded by modalities T1, T2, and FLAIR. In this thesis, we include a new modality of MRI imaging into existing modalities: Magnetization-Prepared Rapid Gradient Echo (MPRAGE). By including additional MPRAGE modality and applying corresponding input-level multimodal segmentation fusion strategy, we can enrich the image information for each single case and provide more intensity or texture information on specific regions of abnormal tissues to corresponding computing systems. With a comprehensive comparison of different segmentation models and different modalities usage, we can prove the improvement in efficiency and performance brought by MPRAGE modality in different multimodal brain tumor segmentation approaches.

Graduation Semester

2021-05

Type of Resource

Thesis

Permalink

http://hdl.handle.net/2142/110723

Copyright and License Information

Copyright 2021 Bo Wang

Owning Collections