Accurate and efficient cardiac motion estimation

Yu, Hanchao

Permalink

https://hdl.handle.net/2142/115340 Copy

Description

Title

Accurate and efficient cardiac motion estimation

Author(s)

Yu, Hanchao

Issue Date

2022-02-25

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

Shi, Humphrey

Doctoral Committee Chair(s)

Shi, Humphrey

Committee Member(s)

• Allan Hasegawa-Johnson, Mark
• Liang, Zhi-Pei
• Sun, Shanhui

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• cardiac motion estimation
• optical flow
• cardiac MR

Language

eng

Abstract

Cardiac motion estimation plays a key role in MRI cardiac feature tracking and function assessment such as myocardium strain. Recent research shows promising results with deep learning-based methods. However, in clinical deployment, previous methods suffer from several issues: (a) Significant performance drops due to mismatched distributions between training and testing datasets, commonly encountered in the clinical environment. It is difficult to collect all representative datasets and to train a universal tracker before deployment. (b) The searching space is large and the optimal is not unique due to the lack of ground truth motion field. (c) Existing deep learning-based methods are 2D models while the cardiac motion is 3D. In this thesis, we proposed a series of approaches to improve the efficiency and accuracy of cardiac motion estimation, along with new evaluation metrics: (a) We propose motion pyramid networks (MPN), a novel deep learning-based approach for accurate and efficient cardiac motion estimation. We predict and fuse a pyramid of motion fields from multiple scales of feature representations to generate a refined motion field. Progress motion compensation is proposed to improve the accuracy through multiple inferences. We then use a novel cyclic teacher-student training strategy to learn the compensation in a single inference step. New evaluation metrics are also proposed to represent errors in a clinically meaningful manner. (b) On top of MPN, we extend it to a novel model for 3D cardiac motion estimation. (c) We proposed a novel fast online adaptive learning (FOAL) framework for better performance on unseen data. It is an online gradient descent-based optimizer that is optimized by a meta-learner. The meta-learner enables the online optimizer to perform a fast and robust adaptation. Our proposed methods outperform strong baseline models on two public available clinical datasets, evaluated by a variety of metrics. The proposed methods also demonstrate time efficiency in inference and online adaptation.

Graduation Semester

2022-05

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/115340

Copyright and License Information

Copyright 2022 Hanchao Yu

Owning Collections