Sparse image reconstruction for molecular imaging.

Area of Science:

• Atomic-level molecular imaging
• Sparse signal reconstruction
• Magnetic Resonance Force Microscopy (MRFM)

Background:

• Molecular imaging at the atomic level presents challenges due to sparse data at subatomic distances.
• Magnetic Resonance Force Microscopy (MRFM) is an emerging technology capable of nanometer resolution imaging.
• Traditional sparse estimators often assume low-coherence system matrices, which is not suitable for MRFM applications.

Purpose of the Study:

• To develop novel sparse estimators for high-coherence systems relevant to MRFM.
• To address the limitations of existing sparse estimation techniques in atomic-level imaging.
• To introduce a generalized thresholding rule and a corresponding estimator for sparse signal recovery.

Main Methods:

• Formulation of a discrete-continuous Laplacian and atom at zero (LAZE) probability density function (p.d.f.).
• Derivation of two sparse estimators by maximizing the joint p.d.f. of observation and image, conditioned on hyperparameters.
• Development of a hybrid thresholding rule, generalizing hard and soft thresholding.
• Utilizing Stein's Unbiased Risk Estimate (SURE) for hyperparameter estimation.

Main Results:

• A novel hybrid thresholding rule was derived, generalizing existing methods.
• A hybrid estimator was developed, which is a generalization of the lasso estimator.
• Numerical studies demonstrated that the hybrid estimator outperforms the lasso estimator in sparse image reconstruction with a Gaussian point spread function (psf).

Conclusions:

• The hybrid estimator offers improved performance for sparse signal reconstruction in high-coherence systems, such as those encountered in MRFM.
• The developed methods are applicable to atomic-level molecular imaging challenges.
• The findings advance sparse estimation techniques for advanced imaging technologies.