Sparse Denoising Technique for Enhanced Photoacoustic Tumor Vasculature Imaging

Photoacoustic (PA) imaging is a promising non-invasive bioimaging modality that captures optically induced acoustic waves to visualize deep tissue structures with high contrast. However, its practical implementation often restricts the number and arrangement of sensors, resulting in sparse or limited-view configurations. In addition to these hardware constraints, optical scattering, acoustic attenuation, and measurement noise substantially degrade the quality of reconstructed images, thereby limiting the reliable resolution of fine structural details. To address these challenges, this study proposes a Sequential Generalized k-Means (SGK)-based sparse representation framework integrated with an Adaptive Kalman Filter (AKF) and skeletonization algorithm to enhance the structural quality of reconstructed PA images. The SGK method effectively suppresses noise by representing PA images using a compact set of adaptively learned dictionary atoms. The AKF further refines the denoised output by enhancing structural continuity and suppressing residual artifacts. Skeletonization is incorporated to preserve vascular topology, maintain connectivity, and improve the morphological consistency of vessel structures. The proposed hybrid method is evaluated on clinically relevant datasets, including a microscopic tumor-affected vasculature PA dataset and synthetic normal and glaucoma-affected retinal PA datasets, where reliable tissue enhancement is critical for preserving pathological features and supporting clinical decision-making. Quantitative evaluation using F1-score, precision, sensitivity, length ratio, and intersection over union demonstrates the effectiveness of the proposed denoising strategy in enhancing tissue structures for photoacoustic imaging (PAI)-based diagnostic applications.