Studies On High-performance Iterative Photoacoustic Image Reconstruction Methods

The photoacoustic (PA) imaging is a novel biomedical imaging technique. Compared with traditional imaging techniques, the PA imaging posseses many advantages. It reflects the laser absorption distribution of the tissue, thus it has a higher definition than the ultrasound imaging. It receives ultrasound signals, thus it has a deeper penertration than the optical imaging. It adopts laser as the energy source, which is nonironizing, thus when the energy is under control, it is noninvasive compared with ironizing imaging techoniques, such as X-ray imaging. Meanwhile, the PA imaging is able to provide functional images. Because of these advantages, the PA imaging is greatly potential in many anspects of biomedical application. The research of the PA imaging is significantly meaningful.The PA image reconstruction method is the key point in the imaging technique. The method’s performance directly influences the quality of the PA imagl and the speed in obtaining the image. From the current stage of this research, the iterative method is a widely-used and high-quality reconstruction method, however, there exsit some problems. In this dissertation, the research work is implemented mainly for the following difficulties, the decline of iterative methods’quality under the sparse-view sampling condition, the inadequacy of reconsturuction methods under multi-sampling conditions, and the heavy computation. Several novel reconstruction methods with the high performance have been peoposed, which includes the following points.A total variation-gradient descent based PA image reconstruction method is proposed. The total variation is involved in the PA image reconstruction. Utilizing the characteristic that an image is often with the small value of the total variation, we calculate the image with the smallest total variation value as the solution of the image reconstruction. Through adding the total variation value into the objective function of the interation, and using the gradient descent method, we can reconstruct the PA image with the total variation value minimized. It is revealed in the simulation and in vitro experiments results that the proposed method is less reliable to the amount of signals. Besides, the proposed method can reconstruct images with the high quality under sparse-view sampling conditions, and the accuracy is better than the other methods including the filtered back-projection method, the deconvolution method and etc..The proposed total variation-gradient descent method is expanded to multi-scanning modes. The image reconstructions under multi-scanning modes are brought in a uniform calculation frame. For different scanning modes, the sampling matrix can be calculated according to the scanning geometry, and the image can be obtained under the same reconstruction frame. The proposed method is verified under the full-view circular scanning, the limited-view circular scanning, the linear scanning, and the unregular scanning based on simulations and in vitro experiments. The performance of the proposed method is proved better than other methods including the algebraic reconstruction method, Norton method and etc..An image reconstruction method based on the polar coordinates is proposed. For the 2D circular scanning PA imaging, a forward model is established in the polar coordinates, based on which the image can be reconstructed. The Centro-symmetry of the scainning circle is fully used, thus the repeat computation of sampling matrices will be avoided, and the computation can be saved. This method is verified through simulations and in vitro experiments. Compared with the method in the Cartesian coordinates, the proposed method in the polar coordinates is available to reconstruct images with the similar quality, meanwhile the computation time can be reduced.An efficient image reconstruction based on the discrete cosine transform (DCT) is proposed. The forward model is established between the PA image and the DCT coefficients of PA signals, and then the image is reconstructed from the DCT coefficients. After setting a certain threshold, a small part of the DCT coefficients is picked out for the reconstruction, which contains the most information for the image reconstruction. Through the comparison of the theoretical calculation complexity, the proposed DCT based method is more efficient than the algebraic reconstruction method. The optional threshold value is fixed through simulations, and the calculation efficiency can be enhanced with the reconstructed images remainning the high quality. It is also revealed in the in virto experiments’results that the DCT based method is more efficient and more practical than the total variation-gradient descent method.As the foundamentation of the image reconstruction method research, the computer simulation method is investigated in this dissertation. A simulation method is proposed by combining the time-of-flight method and the finite-difference time-domain method. With the short-time Fourier transform, the high frequency part of time-of-flight signals and the low frequency part of finite-difference time-domain signasl are melted, and then the new simulation signal can be obtained. It is revealed in the simulation that the proposed simulation method is more suitable for the heterogeneous acoustical medium condition rather than the time-of flight method, and it is more accurate in the amplitude than the finite-difference time-domain method. This simulation method can provide a stable and reliable source of PA signals for the PA image reconstruction reseach, and it is the basis for the comparison and analysis of image reconstruction methods.