Comparison of sampling strategies and sparsifying transforms to improve compressed sensing diffusion spectrum imaging

Area of Science:

• Medical Imaging
• Neuroimaging
• Signal Processing

Background:

• Diffusion Spectrum Imaging (DSI) reconstructs the ensemble average propagator (EAP) but requires extensive measurements.
• Compressive sensing (CS) offers a method to reduce the number of required measurements for DSI.
• Optimizing sampling strategies and sparsifying transforms is crucial for accelerating CS-DSI.

Purpose of the Study:

• To experimentally compare three sampling strategies and six sparsifying transforms for accelerating CS-DSI.
• To evaluate the impact of these choices on the accurate reconstruction of the EAP.

Main Methods:

• A novel sampling scheme with uniform angular and random radial q-space samples was proposed.
• Six discrete sparse representations of the EAP were implemented and compared.
• Evaluation used synthetic and real data, assessing metrics from the EAP, kurtosis, and orientation distribution function.

Main Results:

• The discrete wavelet transform (DWT) using Cohen-Daubechies-Feauveau 9/7 wavelets demonstrated superior performance.
• Uniform angular and random radial sampling combined with DWT yielded more accurate EAP reconstruction.
• This combination outperformed other tested techniques in preserving EAP features.

Conclusions:

• Joint optimization of sampling schemes and sparsifying transforms is essential for accelerated CS-DSI.
• Robust recovery of radial and angular EAP features is achievable with significant undersampling (64 measurements, factor of 4).
• Findings were validated on synthetic and real human brain data.