Autoencoder based blind source separation for photoacoustic resolution enhancement

Abstract

Photoacoustics is a promising technique for in-depth imaging of biological tissues. However, the lateral resolution of photoacoustic imaging is limited by size of the optical excitation spot, and therefore by light diffraction and scattering. Several super-resolution approaches, among which methods based on localization of labels and particles, have been suggested, presenting promising but limited solutions. This work demonstrates a novel concept for extended-resolution imaging based on separation and localization of multiple sub-pixel absorbers, each characterized by a distinct acoustic response. Sparse autoencoder algorithm is used to blindly decompose the acoustic signal into its various sources and resolve sub-pixel features. This method can be used independently or as a combination with other super-resolution techniques to gain further resolution enhancement and may also be extended to other imaging schemes. In this paper, the general idea is presented in details and experimentally demonstrated.

Blind Hyperspectral Unmixing Using Autoencoders: A Critical Comparison

“…It factors the observed HSI into the product of two nonnegative matrices, namely the nonnegative matrix of abundance fractions and the mixing matrix having the endmembers as its columns. Such autoencoders are widely used in problems that can be solved through NMF, such as various blind separation problems [62,63].…”

Secondary Pulmonary Tuberculosis Identification Via pseudo-Zernike Moment and Deep Stacked Sparse Autoencoder

“…SAE only allows a small fraction of the hidden neurons to be active at the same time [29]. This sparsity forces SAE to respond to unique statistical features of the training data [30].…”

Image Enhancement Method for Photoacoustic Imaging of Deep Brain Tissue

“…Liu et al [14] introduced the photoacoustic imaging vasculature enhancement filter (PAIVEF) algorithm to boost micro-vessel imaging in rat eyes, enhancing signals while effectively suppressing noise and extending the vascular imaging depth range. Matan Benyamin et al [15,16] employed a sparse autoencoder enhancement algorithm combined with contrast agents to improve PAI resolution. Gao et al [17] utilized an empirical mode decomposition (EMD) algorithm with a conditional mutual information Photonics 2024, 11, 31 2 of 12 de-noising algorithm to enhance imaging of mouse ear vascular.…”