PET motion correction using PRESTO with ITK motion estimation

Melissa Raquel da Silva Botelho, Liliana Caldeira, J. Scheins, Nuno Matela, Elena Rota Kops, N. Jon ShahEJNMMI 2014 · DOI

TL;DR

This paper presents a method for correcting motion artifacts in PET images by combining the PRESTO library with ITK's motion estimation capabilities. This approach successfully corrects for simulated and real-world motion, improving image quality and making PET/MR imaging more robust.

What Problem Does This Paper Address?

Positron Emission Tomography (PET) is a powerful imaging technique, but its images are susceptible to motion artifacts. Patient movement during scans, even subtle shifts, can blur images, distort quantitative measurements, and obscure important diagnostic details. This is particularly challenging in hybrid PET/MRI systems where patients might be in the scanner for extended periods. Existing motion correction methods often require complex setups or specialized hardware. This paper aims to address this by leveraging readily available software tools to provide an effective and accessible motion correction solution.

Methods

The core of this work lies in integrating two key software components:

  • PRESTO (PET Reconstruction Engine for Spatio-Temporal Optimization): This library is designed to convert motion-corrected Lines of Response (LORs) into high-quality, generic projection data. Essentially, it takes corrected scan data and prepares it for reconstruction.
  • ITK (Insight Toolkit): This is a widely used open-source software system for image analysis. In this study, ITK's registration algorithms were employed to estimate motion. Specifically, the Mutual Information metric was used to align a fixed (reference) image with a moved image. The output of ITK's registration provides six parameters: three for translation and three for rotation, quantifying the patient's movement.

These estimated motion parameters from ITK are then fed into PRESTO to correct the PET data. The reconstructed images were generated using the Ordered Subset Expectation Maximization (OP-OSEM) algorithm with 41 iterations.

Datasets

The authors evaluated their method using three types of datasets:

  • Utah Phantom: A standard phantom used for PET imaging, simulated with varying levels of noise (statistics), scatter, and random events.
  • Brain Phantom: A more anatomically realistic phantom representing brain structures.
  • Clinical Images: Actual PET scans from patients administered with [18F]FDG.

Simulated motions included translations (1, 5, 10, and 30 mm) and rotations (degrees) along three axes. Additionally, EPI (Echo-Planar Imaging) sequence images, commonly acquired during PET/MRI scans, were also used in conjunction with PET data.

Key Insights and Ideas

  • Synergistic Power of PRESTO and ITK: The combination of ITK's robust motion estimation with PRESTO's efficient data correction proves to be a powerful approach for improving PET image quality.
  • Image-Based Correction is Viable: The study demonstrates that image-based motion correction, using readily available software, is a practical and effective strategy.
  • Impact of Statistics and Motion Severity: The accuracy of ITK's motion estimation is significantly affected by image statistics (lower statistics lead to more errors) and the severity of motion (larger movements introduce greater errors).
  • MR vs. PET Image Sensitivity: ITK's motion estimation errors tend to be smaller for MR images compared to PET images, which is an important consideration for hybrid imaging.

Key Results

  • PRESTO, when used with ITK-estimated motion parameters, successfully corrected moved PET images.
  • For simulated motions, ITK showed considerable errors when dealing with low statistics (1E6) and large movements (30mm/degrees).
  • Clinical data exhibited larger errors than phantom data, which is expected due to the inherent complexities of real patient scans.
  • The method is capable of correcting for both translational and rotational movements across three axes.

Limitations and Future Work

The authors acknowledge several limitations:

  • Low Statistics: Images with low signal-to-noise ratios (low statistics) can significantly hamper the accuracy of motion correction due to blurring.
  • Severe Motion: If the motion is too severe, the object of interest may move out of the Field of View (FOV), making correction impossible.

Future work could focus on developing strategies to mitigate the impact of low statistics and exploring methods for handling more extreme motion scenarios. The authors suggest that ITK, coupled with PRESTO, offers a viable and accessible solution for motion correction in both PET and MR imaging.

Summary updated

Abstract

The Siemens BrainPET scanner is a hybrid MRI/PET system. PET images are prone to motion artefacts which degrade the image quality. Therefore, motion correction is essential. The library PRESTO converts motion-corrected LORs into highly accurate generic projection data [1], providing high-resolution PET images. ITK is an open-source software used for registering multidimensional data [2]. ITK provides motion estimation necessary to PRESTO. For this study the Utah phantom, a brain phantom and clinical images were used. The data was simulated with different statistics and included scatter and random events. The data was reconstructed using OP-OSEM [3] with 41 iterations. The patient was administrated with [18]F-FDG. The acquisition took about 35min (96min p.i.). Different motions considering translations and rotations with dimensions of 1, 5, 10 and 30mm and degrees in three axes were simulated with PRESTO. In addition to PET images, EPI sequence [3] images were also used. The metric used in ITK registration algorithm was the Mutual Information. Given a fixed image and a moved image, the outputs are six parameters, which represent motion: three translations and three rotations. Then, the images were corrected in PRESTO using these parameters. For low statistics (1E6) and larger motion (30mm/degrees) ITK showed considerable errors (Figure ​(Figure1).1). PRESTO using ITK estimated parameters can correct well the moved images. Regarding the PET images, clinical data has bigger errors than phantom data as expected. For MR images ITK errors are smaller than for PET images (Figure ​(Figure11). Figure 1 Simulated images in the study with Utah phantom (with a hot and a cold cylinder). On left side of the figure, in the first row, are the acquired PET images of the phantom: first transversal view and then coronal view. The next row shows the simulated ... ITK has some limitations, such as, images with low statistics hamper the correction because of the blurring and if the motion is severe the object of study goes out the FOV and then correction is not possible. ITK is a viable image-based correction, which used with PRESTO makes it possible to easily use PET or MR images for motion correction.
BibTeX
@inproceedings{botelho2014pet,
  author = {Melissa Raquel da Silva Botelho and Liliana Caldeira and J. Scheins and Nuno Matela and Elena Rota Kops and N. Jon Shah},
  title = {{PET motion correction using PRESTO with ITK motion estimation}},
  booktitle = {EJNMMI},
  year = {2014},
  doi = {10.1186/2197-7364-1-s1-a59},
}
Figure