Unsupervised Sparse Cooperative Learning for Limited View Optoacoustic Tomography

Limited-view optoacoustic tomography presents significant challenges in image reconstruction, especially in scenarios where full tomographic coverage is unavailable. To address this, we propose a sparse cooperative network that leverages twin convolutional sub-networks to cooperatively reconstruct high-quality images from complementary and sparsely sampled data. The network employs a self-supervised loss function, which eliminates the need for ground-truth supervision. Our approach effectively reconstructs high-quality images from sparsely distributed data using only 4 transducers, achieving substantial angular expansion to full tomographic coverage with 512 transducers in the frequency domain, enhancing the visibility of structures in all orientations. The model exhibits stable and efficiency convergence for optoacoustic imaging. Additionally, through a detailed analysis of convolutional configurations, we identify that the input noise and the kernel setup maximize reconstruction performance. The results confirm that the proposed method allows for effective image reconstruction even without the availability of full-view data.