Automatic Assessment of Human Gastric Motility and Emptying from Dynamic 3D Magnetic Resonance Imaging

Kun‐Han Lu, Zhongming Liu, Deborah Jaffey, John M. Wo, Kristine M. Mosier, Jiayue Cao, Xiaokai Wang, Terry L. PowleybioRxiv 2020 · DOI

TL;DR

This paper introduces a novel, automated pipeline for analyzing dynamic 3D MRI scans of the stomach. This system can simultaneously assess gastric emptying and motility, overcoming limitations of previous manual and semi-automated methods and paving the way for more widespread clinical and research applications of gastric MRI.

What Problem Does This Paper Address?

Assessing gastric emptying and motility is crucial for understanding digestion and diagnosing related disorders. Current methods are often invasive, cumbersome, indirect, or lack the necessary spatial and temporal resolution. While MRI offers a non-invasive solution with excellent soft-tissue contrast, analyzing dynamic 4D gastric MRI data has been hampered by the lack of standardized, automated software. The complex anatomy of the stomach, respiratory motion, and variations in contrast make automated analysis challenging. This paper aims to bridge this gap by developing an integrated, automated pipeline for comprehensive gastric function assessment.

Methods

The core of this work is an automated image processing pipeline designed for dynamic 3D (4D) gastric MRI data. Key steps include:

  • MRI Protocol: A specific MRI protocol was developed to acquire high spatiotemporal isotropic resolution images of the stomach. This involved using a naturalistic contrast meal (blended ingredients rich in manganese) to enhance luminal contrast in T1-weighted images, enabling better visualization.
  • Respiratory Motion Correction: A non-rigid registration scheme using a multi-resolution, fast free-form deformation (FFD) approach based on cubic B-splines was employed to mitigate motion artifacts caused by breathing.
  • Atlas-Based Segmentation: An automated atlas-based segmentation method was developed. This involved creating a stomach volume atlas from multiple subjects' MRI data, then propagating this atlas to segment and compartmentalize the stomach (fundus, corpus, and antrum) in target images using non-rigid registration and a 3D deformable model.
  • Volumetric Analysis: Gastric emptying was quantified by calculating the volume of the segmented stomach and its compartments over time, normalized to the initial volume.
  • Surface-Based Motility Analysis: A wire-frame mesh model of the segmented stomach surface was created. By tracking the motion of nodes on this mesh using non-rigid surface registration, the frequency, amplitude, and coordination of peristaltic contractions were quantified and visualized.

Datasets

The study utilized dynamic 3D contrast-enhanced MRI images acquired from seventeen healthy human volunteers. For evaluating the segmentation method, 30 3D MRI images (2 per subject, acquired at 0 and 30 minutes post-meal) were manually segmented to serve as ground truth.

Key Insights and Ideas

  • Naturalistic Contrast Meal: Using a meal rich in manganese (like a smoothie with tofu and fruits) proved effective for enhancing luminal contrast in MRI, simplifying segmentation without relying on potentially problematic exogenous contrast agents.
  • Integrated Pipeline: The true innovation lies in the integration of motion correction, segmentation, and motility analysis into a single, automated workflow, making 4D gastric MRI analysis more accessible.
  • Surface-Based Motility Visualization: Moving beyond simple volume measurements, the surface-based analysis allows for direct visualization of peristaltic wave propagation, offering a more intuitive understanding of gastric motility.
  • Quantifying Peristaltic Waves: The pipeline successfully quantifies key aspects of peristalsis, including wave frequency, amplitude, and coordination, providing objective metrics for gastric function.

Key Results

  • Segmentation Accuracy: The atlas-based segmentation achieved an accuracy of 91.10±0.43%, with Type-I error of 0.11±0.01% and Type-II error of 0.22±0.01%.
  • Gastric Emptying: Total gastric volume decreased by 34.64±2.8% over 1 hour, following a linear-exponential pattern. Fundus, corpus, and antrum emptying rates were also quantified.
  • Gastric Motility: Peristaltic patterns were observed with a median of 4 wave-fronts (range 3-6) and a mean frequency of 3.09±0.07 cycles per minute (CPM). The contractile amplitude was significantly stronger in the antrum (5.18±0.24 mm) compared to the corpus (3.30±0.16 mm).
  • Motion Correction: The respiratory motion correction significantly reduced image misalignment, as indicated by reductions in DVARS and SAD metrics.

Limitations and Future Work

  • Computational Demands: The automated algorithms, while efficient, still require significant computation time (approximately 2 hours per session on a workstation PC), which is noted as a potential limitation.
  • Future Directions: The authors suggest that advancements in deep-learning for computer vision could further accelerate the registration and segmentation processes. They also envision similar analysis frameworks being developed for lower gastrointestinal motility, aiming to create a complete methodological framework for assessing the entire GI system. The study establishes a baseline for future gastric MRI studies in disease states.

Summary updated

Abstract

Abstract Background Time-sequenced magnetic resonance imaging (MRI) of the stomach is an emerging technique for non-invasive assessment of gastric emptying and motility. However, an automated and systematic image processing pipeline for analyzing dynamic 3D (i.e., 4D) gastric MRI data is not yet available. This study introduces an MRI protocol for imaging the stomach with high spatiotemporal isotropic resolution and provides an integrated pipeline for assessing gastric emptying and motility simultaneously. Methods Diet contrast-enhanced MRI images were acquired from seventeen healthy humans after they consumed a naturalistic contrast meal. An automated image processing pipeline was developed to correct for respiratory motion, to segment and compartmentalize the lumen-enhanced stomach, to quantify total gastric and compartmental emptying, and to compute and visualize gastric motility on the surface of the stomach. Key Results The gastric segmentation reached an accuracy of 91.10±0.43% with the Type-I error and Type-II error being 0.11±0.01% and 0.22±0.01%, respectively. Gastric volume decreased 34.64±2.8% over 1 hour where the emptying followed a linear-exponential pattern. The gastric motility showed peristaltic patterns with a median = 4 wave-fronts (range 3 - 6) and a mean frequency of 3.09±0.07 cycles per minute (CPM). Further, the contractile amplitude was stronger in the antrum than in the corpus (antrum vs. corpus: 5.18±0.24 vs. 3.30±0.16 mm; p < .001). Conclusions & Inferences The automated, streamlined software can process dynamic 3D MRI images and produce comprehensive and personalized profiles of gastric motility and emptying. This software will facilitate the application of MRI for monitoring gastric dynamics in research and clinical settings.
BibTeX
@misc{lu2020automatic,
  author = {Kun‐Han Lu and Zhongming Liu and Deborah Jaffey and John M. Wo and Kristine M. Mosier and Jiayue Cao and Xiaokai Wang and Terry L. Powley},
  title = {{Automatic Assessment of Human Gastric Motility and Emptying from Dynamic 3D Magnetic Resonance Imaging}},
  year = {2020},
  doi = {10.1101/2020.12.11.421743},
  howpublished = {bioRxiv},
}

Loading related content...

Figure