Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TAI-GAN: A Temporally and Anatomically Informed Generative Adversarial Network for early-to-late frame conversion in dynamic cardiac PET inter-frame motion correction

Xueqi Guo, Luyao Shi, Xiongchao Chen, Qiong Liu, Bo Zhou, Huidong Xie, Yi-Hwa Liu, Richard Palyo, Edward J. Miller, Albert J. Sinusas, Lawrence H. Staib, Bruce Spottiswoode, Chi Liu, Nicha C. Dvornek

TL;DR

Dynamic cardiac PET with 82Rb faces inter-frame motion and rapid tracer kinetics that degrade MBF quantification. The authors propose TAI-GAN, a temporally and anatomically informed GAN that performs all-to-one conversion of early frames into late-frame appearance using a FiLM-based temporal encoder and rough anatomical masks as guidance. Across 5-fold cross-validation on a clinical dataset, TAI-GAN yields superior frame conversion quality and consistently improves both conventional and DL-based motion correction methods, translating to reduced MBF estimation errors. The approach demonstrates potential to streamline inter-frame motion correction and enhance diagnostic reliability in dynamic cardiac PET imaging.

Abstract

Inter-frame motion in dynamic cardiac positron emission tomography (PET) using rubidium-82 (82-Rb) myocardial perfusion imaging impacts myocardial blood flow (MBF) quantification and the diagnosis accuracy of coronary artery diseases. However, the high cross-frame distribution variation due to rapid tracer kinetics poses a considerable challenge for inter-frame motion correction, especially for early frames where intensity-based image registration techniques often fail. To address this issue, we propose a novel method called Temporally and Anatomically Informed Generative Adversarial Network (TAI-GAN) that utilizes an all-to-one mapping to convert early frames into those with tracer distribution similar to the last reference frame. The TAI-GAN consists of a feature-wise linear modulation layer that encodes channel-wise parameters generated from temporal information and rough cardiac segmentation masks with local shifts that serve as anatomical information. Our proposed method was evaluated on a clinical 82-Rb PET dataset, and the results show that our TAI-GAN can produce converted early frames with high image quality, comparable to the real reference frames. After TAI-GAN conversion, the motion estimation accuracy and subsequent myocardial blood flow (MBF) quantification with both conventional and deep learning-based motion correction methods were improved compared to using the original frames.

TAI-GAN: A Temporally and Anatomically Informed Generative Adversarial Network for early-to-late frame conversion in dynamic cardiac PET inter-frame motion correction

TL;DR

Dynamic cardiac PET with 82Rb faces inter-frame motion and rapid tracer kinetics that degrade MBF quantification. The authors propose TAI-GAN, a temporally and anatomically informed GAN that performs all-to-one conversion of early frames into late-frame appearance using a FiLM-based temporal encoder and rough anatomical masks as guidance. Across 5-fold cross-validation on a clinical dataset, TAI-GAN yields superior frame conversion quality and consistently improves both conventional and DL-based motion correction methods, translating to reduced MBF estimation errors. The approach demonstrates potential to streamline inter-frame motion correction and enhance diagnostic reliability in dynamic cardiac PET imaging.

Abstract

Inter-frame motion in dynamic cardiac positron emission tomography (PET) using rubidium-82 (82-Rb) myocardial perfusion imaging impacts myocardial blood flow (MBF) quantification and the diagnosis accuracy of coronary artery diseases. However, the high cross-frame distribution variation due to rapid tracer kinetics poses a considerable challenge for inter-frame motion correction, especially for early frames where intensity-based image registration techniques often fail. To address this issue, we propose a novel method called Temporally and Anatomically Informed Generative Adversarial Network (TAI-GAN) that utilizes an all-to-one mapping to convert early frames into those with tracer distribution similar to the last reference frame. The TAI-GAN consists of a feature-wise linear modulation layer that encodes channel-wise parameters generated from temporal information and rough cardiac segmentation masks with local shifts that serve as anatomical information. Our proposed method was evaluated on a clinical 82-Rb PET dataset, and the results show that our TAI-GAN can produce converted early frames with high image quality, comparable to the real reference frames. After TAI-GAN conversion, the motion estimation accuracy and subsequent myocardial blood flow (MBF) quantification with both conventional and deep learning-based motion correction methods were improved compared to using the original frames.
Paper Structure (16 sections, 15 equations, 8 figures, 9 tables)

This paper contains 16 sections, 15 equations, 8 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Clinical cardiac dataset and pre-processing
  4. Proposed network
  5. Implementation details
  6. Frame mapping evaluation
  7. Simulated and real-patient motion correction
  8. Myocardial blood flow analysis
  9. Results
  10. Frame conversion evaluation
  11. Motion simulation study
  12. Parametric fitting results
  13. Real-motion evaluation
  14. Ablation studies of temporal and anatomical information
  15. Discussion
  16. ...and 1 more sections

Figures (8)

  • Figure 1: The architecture of the proposed temporally and anatomically informed generative adversarial network (TAI-GAN).
  • Figure 2: Sample results of early-to-late frame conversion using each method with overlaid RVBP (red), LVBP (blue), and myocardium (green) segmentation contours, with arrows highlighting alignment or mismatch between the cardiac segmentations and structures.
  • Figure 3: Sample motion simulation and correction results with different methods of frame conversion, with overlaid RVBP (red), LVBP (blue), and myocardium (green) segmentation contours and arrows highlighting alignment or mismatch between the cardiac segmentations and structures.
  • Figure 4: Scatter plots of MBF results estimated from no motion frames vs. no motion correction and BIS motion correction after different conversion methods.
  • Figure 5: A comparison of LVBP and myocardium TACs of each conversion method before and after BIS motion correction.
  • ...and 3 more figures