Automating and accelerating magnetic resonance imaging
Abstract/Contents
- Abstract
- Magnetic resonance imaging (MRI) can provide high-quality multi-contrast diagnostic images. It is non-invasive and does not use ionizing radiation. Therefore, it is safe for young patients. MRI exams follow a procedure consisting of preparation, scan prescription, data collection (i.e., the actual scanning), image reconstruction, and checking the result. Unfortunately, MRI has significantly longer scan times compared to other modalities such as computed tomography (CT). These long scan times are especially challenging for children who may struggle to stay still. In this dissertation, we aim to expedite the whole MRI exam procedure by automating and accelerating four parts of the scanning process: prescription, data collection, reconstruction, and the after-scan check. This is done through a series of three projects. First, we present a method for region-of-interest (ROI) prediction and field-of-view (FOV) prescription. Manual prescription of the field of view by MRI technologists is variable and prolongs the scanning process. Often, the FOV is either too large or crops critical anatomy. We propose a deep-learning framework, trained with radiologists' supervision, for automating FOV prescription. An intra-stack shared feature extraction network and an attention network are used to process a stack of 2D image inputs to generate scalars defining the location of a rectangular ROI. The attention mechanism is used to make the model focus on a small number of informative slices in a stack. Then the smallest FOV that makes the neural network predicted an ROI free of aliasing is calculated by an algebraic operation derived from MR sampling theory. The framework's performance is examined quantitatively with intersection over union (IoU) and pixel error on position, and qualitatively with a reader study. The framework's prescription is clinically acceptable 92\% of the time as rated by an experienced radiologist. Second, we present a learning-based model for reconstructing undersampled data using unpaired adversarial training. The lack of ground-truth MR images impedes the common supervised training of neural networks for image reconstruction. To cope with this challenge, this work leverages unpaired adversarial training for reconstruction networks, where the inputs are undersampled k-space data and naively reconstructed images from one dataset, and the labels are high-quality images from another dataset. The reconstruction networks consist of a generator which suppresses the input image artifacts, and a discriminator using a pool of (unpaired) labels to adjust the reconstruction quality. The generator is an unrolled neural network -- a cascade of convolutional and data consistency layers. The discriminator is also a multilayer Convolutional Neural Network (CNN) that plays the role of a critic scoring the quality of reconstructed images based on the Wasserstein distance metric. Our experiments with knee MRI datasets demonstrate that the proposed unpaired training enables diagnostic-quality reconstruction when high-quality image labels are not available, or when the amount of label data is small. In addition, our adversarial training scheme can achieve better image quality (as rated by expert radiologists) compared with the paired training methods using pixel-wise loss. Finally, we present a no-reference image quality assessment (IQA) framework that checks the exam outcome. In clinical practice MR images are often first seen by radiologists long after the scan. If image quality is inadequate the patient may have to return for an additional scan, or a suboptimal interpretation is rendered. Automatic IQA would enable real-time remediation. Existing IQA methods for MRI give only a general quality score. These are agnostic to the cause of the low-quality scan and the solution for improvement. Furthermore, radiologists' image quality requirements vary with the scan type and diagnostic task. Therefore, the same score may have different implications for different scans. We propose a framework with a multi-task CNN model trained with calibrated labels and measured with image rulers. Labels calibrated by human inputs follow a well-defined and efficient labeling task. Image rulers address varying quality standards and provide a concrete way of interpreting raw scores from the CNN. The model supports assessments of perceptual noise level, rigid motion, and peristaltic motion. Our experiments show that label calibration, image rulers, and multi-task training improve the model's performance and ability to generalize.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2023; ©2023 |
| Publication date | 2023; 2023 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Lei, Ke |
|---|---|
| Degree supervisor | Vasanawala, Shreyas |
| Thesis advisor | Vasanawala, Shreyas |
| Thesis advisor | Nishimura, Dwight George |
| Thesis advisor | Pauly, John (John M.) |
| Degree committee member | Nishimura, Dwight George |
| Degree committee member | Pauly, John (John M.) |
| Associated with | Stanford University, School of Engineering |
| Associated with | Stanford University, Department of Electrical Engineering |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Ke Lei. |
|---|---|
| Note | Submitted to the Department of Electrical Engineering. |
| Thesis | Thesis Ph.D. Stanford University 2023. |
| Location | https://purl.stanford.edu/jj603nd4629 |
Access conditions
- Copyright
- © 2023 by Ke Lei
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...