Research On High Performance Reconstruction Algorithm For Frequency Domain Photoacoustic Tomography

Photoacoustic imaging(PAI)combines the high reconstruction contrast,diverse imaging functions of optical imaging and the deep penetration depth of ultrasound imaging.As a result,it has become one of the most potential nondestructive biomedical imaging techniques in recent decades.PAI can be classified into diverse categories according to different classification criterion.Among them,frequency domain photoacoustic tomography employs the continuous wave laser as excitation source,which is modulated by different modulation mode in accord with different imaging modality.Utilized by various weak signal processing strategies,important information is extracted from the high level background noises.Unlike the high-energy pulse laser source which is used in time domain PAI,continuous wave laser is smaller,safer and less expensive.Nevertheless,lower photoacoustic signal(PAS)intensity come along with the much lower source power,which degrades the signal to noise ratio(SNR).In consequence,accurate extraction of important information from PAS becomes even harder,leads to more serious distortion in imaging result and demands algorithm of higher performance for reconstruction.Moreover,most frequency domain PAI methods requires huge amount of data that consumes a good deal of time and space for data collection and calculation.Those disadvantages confine the application of this technique.This dissertation focuses on the methods of weak signal extraction,noise reduction,high quality and high performance image reconstruction in photoacoustic tomography of two different modalities.Researches on signal preprocessing,important information extraction and regularization methods have been carried out in frequency domain PAI.Innovations have been made in the research of new filtering algorithms,the extraction of amplitude and phase information together with the high performance reconstruction algorithms.Aiming at the limitations of weak signal processing methods in frequency domain PAI,this dissertation focuses on the research of denoising reconstruction algorithm under two difference PAI modalities.(1)Under the filtered back-projection PAI modality,a new filtering algorithm based on empirical mode decomposition(EMD)has been proposed for the linear frequency modulation(LFM)source excited PAI under the principle of zero phase filtering.This algorithm greatly improves the SNR and increases the accuracy of filtering result via the “two stages” filtering scheme.By filtering the actual signal collected in the experiment,the background noise of the reconstruction result is reduced by about 50%.Also,the cross-correlation operation is used to verify the relationship between the target sample size and the center frequency of the ultrasonic transducer,which provides an experimental basis for their matching.(2)Under the model-based PAI modality,the appropriate discrete-frequency modulation mode for the light source is determined and a virtual lock-in amplifier is designed according to the characteristics of its observation matrix.By applying weak signal extraction technology,the virtual lock-in amplifier realizes the extraction of the amplitude and phase information from the PAS collected by experiment and generates the observation matrix automatically.Simulations are carried out to test the operation condition and precautions of the lock-in amplifier.With the application of this virtual amplifier,the model-based frequency domain PAI modality can realize the feasibility without relying on hardware,reduce the complexity and cost of the system,improves the portability of the system at the same time.(3)A self-adaptive filtering algorithm based on EMD for discrete frequency modulation mode is designed to solve for the serious information distortion in observation matrix under the model-based PAI modality.With fully consideration of the discrete frequency modulation mode and different working conditions of the virtual lock-in amplifier,the absolute value of the correlation coefficient is used as a reference for filtering effectiveness.By gradually denoising the signal the modal aliasing is eliminated to the utmost extent which increases the accuracy of filtering.SNR is further improved by narrowband segmentation for more accurate extraction results of the virtual lock-in amplifier.The effectiveness of the filtering algorithm is verified by both simulation and experiment thereafter.The background noise level is sharply declined and the peak signal-to-noise ratio(PSNR)is improved by about 10%.(4)Original reconstruction algorithm applied in the model-based PAI modality is the combination of the least square QR factorization and Tikhonov algorithm.Aiming at the problems of loose constrain,slow speed,poor anti-noise ability and more artifacts in reconstruction results of the original algorithm,truncated singular value decomposition (TSVD)and total variation augmented Lagrangian alternating direction algorithm(TVAL3)are applied into the into the imaging solving step of model-based frequency domain PAI and are analyzed in details.Based on the adaptive empirical mode decomposition filtering algorithm and virtual lock-in amplifier,a comprehensive high performance frequency domain photoacoustic imaging reconstruction algorithm under the model-based reconstruction modality is proposed based on the TVAL3 algorithm.Under experimental condition,not only the contrast of the image is improved,the artifact generated by the original algorithm is also eliminated.The PSNR is further increased by 16% and the reconstruction speed is increased by 4 times.