Sparse optoacoustic sensing with convolutional dictionary learning

Sparse optoacoustic sensing (SOS) enhances tomographic imaging by enabling high frame rates and reducing system complexity through partial data acquisition. However, its performance depends on advanced algorithms that compensate for under-sampled data. This study introduces a novel multi-layer convolutional dictionary-learning algorithm for SOS to improve image reconstruction accuracy. Methods: We propose a multi-layer convolutional dictionary-learning approach that eliminates the need for pursuit algorithms and dictionary-wise parameters. Unlike traditional patch-based methods, our model enforces slice-wise communication to achieve a globally consistent solution from sparse data. The algorithm was validated on both synthetic and experimental in-vivo datasets. Results: The proposed method demonstrated superior recovery accuracy compared to existing dictionary-learning techniques, yielding higher-fidelity reconstructions in under-sampled optoacoustic imaging scenarios. Conclusion: Our algorithm significantly improves image reconstruction in SOS, offering a robust computational solution for sparse data acquisition. Significance: This advancement enhances optoacoustic imaging performance and can be extended to other modalities relying on sparsely sampled data, with broad implications for biomedical imaging.