A BLIND COMPRESSIVE SENSING FRAMEWORK FOR ACCELERATED DYNAMIC MRI.

Area of Science:

• Image processing
• Signal recovery
• Computational imaging

Background:

• Dynamic imaging often requires high sampling rates, leading to large data acquisition burdens.
• Existing methods like low-rank techniques struggle with significant inter-frame motion in dynamic datasets.
• Compressive sensing offers a potential solution for acquiring dynamic images from under-sampled data.

Purpose of the Study:

• To develop a novel blind compressive sensing (BCS) framework for dynamic image reconstruction.
• To address limitations of current methods in handling inter-frame motion.
• To improve the accuracy and quality of reconstructed dynamic images from under-sampled measurements.

Main Methods:

• Modeling dynamic signals as sparse linear combinations of temporal basis functions from a large dictionary.
• Simultaneously estimating the dictionary and sparse coefficients from under-sampled measurements.
• Employing an efficient majorize-minimize algorithm with a continuation strategy for optimization.

Main Results:

• The proposed BCS framework demonstrates improved reconstruction performance compared to low-rank methods.
• Significant performance gains were observed in datasets with considerable inter-frame motion.
• The method effectively handles the simultaneous estimation of dictionary and sparse coefficients.

Conclusions:

• The novel BCS framework offers a superior approach for dynamic image recovery from under-sampled data.
• This method is particularly effective in scenarios with substantial inter-frame motion.
• The simultaneous estimation strategy and efficient algorithm contribute to enhanced reconstruction quality.