Pulsed laser diodes (PLD) are preferred as excitation sources in photoacoustic tomography due to their low cost, compact size, and high pulse repetition rate. When PLD is used in conjunction with multiple single-element ultrasound transducers (SUT), the imaging speed can be improved. However, during PAT image reconstruction, the exact radius of each SUT is required for accurate reconstruction. Herein, we propose a novel deep learning approach to alleviate the need for radius calibration. We developed a convolutional neural network (fully dense U-Net) with a convolutional long short-term memory (LSTM) block as the bridge to reconstruct the PAT images. In vivo imaging was used to verify the performance of the network. Our results and analysis demonstrate that the proposed network eliminates the need for radius calibration without sacrificing the reconstructed PAT image quality.
|