Democratizing LED-Based Photoacoustic Imaging with Adaptive Beamforming and Deep Convolutional Neural Network

Abstract

The current standard photoacoustic (PA) imaging technology includes two hardware requirements: high power pulsed laser for light illumination and multi-channel data acquisition device for PA signal recording. These requirements have been limiting factors to democratize PA imaging because a laser is heavy, expensive and includes hazardous risk, and most parallel data acquisition technology is available only in specialized research systems. The goal of this chapter is to provide an overview of technologies that will enable safer and more accessible PA imaging, as well as introduce the use of safe and fast light emitting diode (LED) light sources in combination with clinical ultrasound machines. There are two limiting factors that prevent achieving this. First, clinical ultrasound machines typically only provide post-beamformed data based on an ultrasound delay function, which is not suitable for PA reconstruction. Second, a PA image based on the LED light source suffers from low signal-to-noise-ratio due to limited LED-power and requires a large number of averaging. To resolve these challenges, an adaptive synthetic aperture beamforming algorithm is applied to treat defocused data as a set of pre-beamformed data for PA reconstruction. An approach based on deep convolutional neural network trains and optimizes the network to enhance the SNR of low SNR images by guiding its feature extraction at different layers of the architecture. We will review and discuss these technologies that could be the key to advancing LED based PA imaging to become more accessible and easier to translate into clinical applications.