A three-step reconstruction method for fluorescence molecular tomography based on compressive sensing.

Area of Science:

• Biomedical Imaging
• Optical Imaging
• Neuroscience

Background:

• Fluorescence molecular tomography (FMT) offers real-time in vivo neurotransmission (NT) quantification.
• FMT data are inherently noisy, and the reconstruction problem is ill-posed.
• Traditional compressive sensing methods are unsuitable for FMT due to a highly coherent system matrix.

Purpose of the Study:

• To develop and evaluate a robust three-step reconstruction method for FMT.
• To overcome limitations of existing methods in handling noisy data and ill-posed reconstruction problems.
• To improve the accuracy of in vivo neurotransmission quantification.

Main Methods:

• A three-step reconstruction process involving truncated singular value decomposition (SVD), homotopy-based ℓ1 regularization, and maximum-likelihood expectation maximization (MLEM).
• Truncated SVD reduces matrix coherence.
• Homotopy-based ℓ1 regularization exploits spatial sparsity, followed by MLEM for Poisson noise modeling and improved quantitation.

Main Results:

• The proposed method achieved a 20% lower root mean square error in fluorescent source intensity estimation compared to the pure-homotopy method.
• Substantially more accurate fluorescence distribution reconstruction was observed compared to conventional ℓ2 regularized algorithms.
• The method demonstrated effectiveness across various simulated fluorescent source sizes and intensities.

Conclusions:

• The proposed three-step reconstruction method significantly enhances the accuracy and reliability of FMT for in vivo neurotransmission quantification.
• This approach effectively addresses the challenges posed by noisy data and ill-posed reconstruction in FMT.
• The method shows considerable promise for future applications in neuroscience research and clinical settings.