A Wavelet‐Based Multiresolution Reconstruction Method for Fluorescent Molecular Tomography

Wei Zou, Jiajun Wang, Kongpei Wu, Dagan FengInternational Journal of Biomedical Imaging 2009 · DOI

TL;DR

This paper introduces a novel wavelet-based multiresolution reconstruction method for Fluorescent Molecular Tomography (FMT). By solving both the forward and inverse problems in the wavelet domain and employing a parallel computing strategy, the method significantly speeds up reconstruction and improves image quality, particularly in scenarios with large variations in tissue optical properties.

What Problem Does This Paper Address?

Fluorescent Molecular Tomography (FMT) is a powerful technique for visualizing biological processes at the molecular and cellular levels in vivo. It works by injecting fluorescent markers into a biological system, illuminating the tissue with light, and then detecting the emitted fluorescence. Reconstructing a 3D image of the fluorescent source distribution from these measurements is crucial. However, this reconstruction process typically involves solving computationally intensive, large-dimensional matrix equations. This computational burden becomes especially problematic when the optical properties of the target tissue deviate significantly from those of a reference medium used in some reconstruction algorithms. The paper aims to address this computational bottleneck and improve reconstruction accuracy, particularly in these challenging scenarios.

Methods

The core of the proposed method lies in two key innovations:

  • Wavelet-Based Multiresolution Reconstruction: Instead of working directly with the raw data and full-resolution images, the authors leverage the properties of wavelet transforms. Wavelets allow for the decomposition of signals into different frequency components at various resolutions. This multiresolution approach enables the reconstruction to be performed iteratively, starting with a coarse representation and progressively refining it. This is particularly beneficial because most of the important information in a signal is often concentrated in a few significant wavelet coefficients, allowing for the neglect of smaller ones and thus reducing computational complexity.
  • Parallel Forward Computing Strategy: The forward problem in FMT (predicting detector readings for a given source and medium) is coupled with the inverse problem. The authors propose a strategy to decouple the excitation and emission light diffusion equations. This decoupling allows the forward problem to be solved in parallel, significantly speeding up the computation. This is crucial because, in cases of large optical property deviations, the forward problem needs to be solved repeatedly during the reconstruction process.

The authors combine these strategies: the forward problem for excitation light is solved using the multiresolution wavelet approach, while the forward problem for emission light is handled using the parallel computing strategy. The inverse problem is also formulated and solved in the wavelet domain.

Datasets

The paper primarily uses simulated data for evaluation.

  • 2D Simulated Phantom: A 2D phantom with two anomalies was used to test the algorithm. This phantom included parameters for background and anomaly regions, with specified optical and fluorescent properties. Gaussian noise (15 dB SNR) was added to simulate realistic measurement conditions, and background fluorescence was also included.
  • 3D Simulated Phantom: A cylindrical phantom with a uniform background and a small cylindrical anomaly was used for 3D reconstruction. This phantom also had specified optical properties and measurement planes at different z-coordinates.

The paper does not introduce new benchmark datasets but rather uses simulated phantoms to evaluate its proposed method against existing approaches.

Key Insights and Ideas

  • Wavelets for Efficiency: The fundamental insight is that wavelets naturally lend themselves to multiresolution analysis, which can dramatically reduce the computational cost of solving large systems of equations by allowing iterative refinement and focusing on significant coefficients.
  • Decoupling for Parallelism: The key to accelerating the forward problem, especially when it needs to be solved repeatedly, is to decouple the coupled equations. This allows for parallel processing, a standard technique for speeding up computations.
  • Handling Large Optical Deviations: The proposed method is specifically designed to be effective even when the optical properties of the target tissue differ significantly from a reference medium. This is a critical limitation of some existing methods.
  • Data Correction is Crucial: The paper highlights the importance of correcting for background fluorescence, demonstrating that this step significantly improves the quality of the reconstructed images.

Key Results

  • Speedup: The wavelet-based multiresolution approach combined with parallel computing significantly reduces reconstruction time. For example, in the 3D case, the proposed algorithm achieved a computation time of 185 seconds, compared to 2279 seconds for a single-resolution method, representing a substantial speedup.
  • Accuracy: The proposed method demonstrates improved reconstruction accuracy, as measured by error functions (E and NRMS). For instance, in a 2D comparison, the proposed algorithm achieved a lower error (6.573 × 10⁻⁸) compared to a method in [9] (2.312 × 10⁻⁸) when dealing with large optical property deviations.
  • Image Quality: Visual inspection of the reconstructed images (Figures 3, 4, 7, 10, 11) shows that the proposed method produces clearer images with better localization and definition of anomalies, especially after data correction.
  • Adaptivity: The paper explores using adaptively refined meshes based on prior information, which further reduces computational requirements while maintaining reconstruction accuracy.

Limitations and Future Work

  • Limitations: The paper primarily relies on simulated data. While the simulations are designed to be realistic (e.g., including noise and background fluorescence), validation with real experimental data would be a crucial next step. The paper also mentions that the parallel implementation is particularly beneficial for 3D cases, suggesting that the benefits might be less pronounced for simpler 2D scenarios.
  • Future Work: The authors suggest that incorporating prior information from other imaging modalities (multimodality imaging) could further improve reconstruction accuracy and reduce computational demands. The exploration of adaptively refined meshes is also a promising avenue for future research. The paper also implicitly suggests that further optimization of the parallel computing strategy could lead to even greater speedups.

Summary updated

Abstract

Image reconstruction of fluorescent molecular tomography (FMT) often involves repeatedly solving large-dimensional matrix equations, which are computationally expensive, especially for the case where there are large deviations in the optical properties between the target and the reference medium. In this paper, a wavelet-based multiresolution reconstruction approach is proposed for the FMT reconstruction in combination with a parallel forward computing strategy, in which both the forward and the inverse problems of FMT are solved in the wavelet domain. Simulation results demonstrate that the proposed approach can significantly speed up the reconstruction process and improve the image quality of FMT.
BibTeX
@article{zou2009waveletbased,
  author = {Wei Zou and Jiajun Wang and Kongpei Wu and Dagan Feng},
  title = {{A Wavelet‐Based Multiresolution Reconstruction Method for Fluorescent Molecular Tomography}},
  journal = {International Journal of Biomedical Imaging},
  year = {2009},
  doi = {10.1155/2009/294545},
}
Figure