Enabling fast and high quality LED photoacoustic imaging: a recurrent neural networks based approach.

Photoacoustic (PA) techniques have shown promise in the imaging of tissue chromophores and exogenous contrast agents in various clinical applications. However, the key drawback of current PA technology is its dependence on a complex and hazardous laser system for the excitation of a tissue sample. Although light-emitting diodes (LED) have the potential to replace the laser, the image quality of an LED-based system is severely corrupted due to the low output power of LED elements. The current standard way to improve the quality is to increase the scanning time, which leads to a reduction in the imaging speed and makes the images prone to motion artifacts. To address the challenges of longer scanning time and poor image quality, in this work we present a deep neural networks based approach that exploits the temporal information in PA images using a recurrent neural network. We train our network using 32 phantom experiments; on the test set of 30 phantom experiments, we achieve a gain in the frame rate of 8 times with a mean peak-signal-to-noise-ratio of 35.4 dB compared to the standard technique.