Research On High-quality Photoacoustic Image Reconstruction Method

Photoacoustic imaging(PAI)technology is an emerging non-invasive and in-vivo biomedical imaging method.It has the advantages of the high contrast of pure optical imaging and high resolution of pure ultrasound imaging.And it has been applied in the early diagnosis of cancer and the detecting of vulnerable plaques in cardiovascular diseases.PAI is essentially a thermos-acoustic imaging,which combines a certain image reconstruction algorithm to invert the received ultrasound signals into an initial photoacoustic pressure distribution image of the biological tissues.There are many methods have been explored to implement the reconstruction of photoacoustic images,but the images reconstructed by the traditional methods are generally the light-absorption cumulative images.However,what is really close-related to many diseases is the optical absorption coefficient of the biological tissue,and various parameters computed by it,such as oxygen saturation and oxygen metabolism.In practical applications,the methods to recover the tissue absorption parameter are called quantitative photoacoustic imaging.At present,most of the quantitative photoacoustic imaging methods are based on numerical simulation experiments or in-vivo quantitative imaging for shallow tissues.When quantitative imaging is performed on deep tissues in the limited view data acquisition mode,the photoacoustic images will have obvious reconstruction artifacts,which has a great influence on the quantitative reconstruction results and affects the correctness of the quantitative images directly.In addition,to obtain a high-quality photoacoustic image,the elements in the ultrasonic array are densely packed,which directly increases the system cost.Due to the low laser repetition rate used in PACT(usually 10-20 Hz),multiplexing techniques will be required,leading to longer image acquisition times.Sparse sampling is an effective way to conquer above problems.However,in the current reconstruction model for sparse-sampling photoacoustic computed tomography(PACT),the sparse transforms used are usually pre-defined and they cannot adequately capture the underlying features of a particular dataset,so it would limit high-quality recovery of photoacoustic images.To improve the two problems discussed above,this thesis proposed two methods correspondingly.The main works and contributions are as follows.(1)Aiming at the problem of quantitative imaging for deep biological tissues in the limited-view mode,a quantitative method based on improved fixed point iterative is proposed.Firstly,the original photoacoustic pressure image was filtered by Wiener filtering.Secondly,theoptical transmission model was used to solve the optical fluence of the target imaging region.Finally,iterative calculation was performed to obtain the optical absorption coefficient distribution of the target tissue.In addition,TOAST++ software was introduced to implement the forward computation of the optical transmission model,which can improve the efficiency and accuracy of quantitative imaging.Using the method proposed in this thesis,there were fewer artifacts in the deep quantitative reconstructed photoauoustic images.The optical absorption coefficient of the quantitatively reconstructed deep target tissue was very close to the optical absorption coefficient of the shallow target tissue.As a result,the quantitative reconstruction of the optical absorption coefficient of the deep biological tissue was implemented successfully.(2)To improve the performance of photoacoustic image reconstruction in sparse sampling mode,this thesis proposed a three dimensional photoacoustic image reconstruction method based on dictionary learning(DL).First,about 5% frames distributed homogeneously on the 3-D imaging area are selected for full sampling,these frames are reconstructed by the traditional back projection(BP)algorithm,and then combined into a training data.Secondly,using the training data,the dictionary is computed by running the K-SVD algorithm.Finally,DL based reconstruction for each B-Scan photoacoustic image was achieved by an iterative process,and a3-D photoacoustic images were constructed.In this thesis,the dictionary learning technique is incorporated into the iterative reconstruction of PACT,which effectively suppresses the sparse sampling artifacts in the reconstructed photoacoustic images.Using the method proposed in this thesis,the accuracy and contrast of the reconstructed photoacoustic image are improved dramatically.Further,the method developed in this thesis only requires sparse ultrasonic array,which leads to less amount of data acquisition and lower system cost.