Research On Photoacoustic Imaging Method Based On Dictionary Learning

Photoacoustic imaging(PAI)is a new noninvasive imaging modality.In medical field,this imaging method,with high contrast,high resolution and large imaging depth,has the advantages of both optical imaging and ultrasound imaging.In recent years,PAI has attracted more and more attention and has become one of the research hotspots in the field of biomedical imaging.In order to ensure the quality of imaging,ultrasonic array based photoacoustic tomography(PACT)needs arranging densely.Therefore,the large amount of data acquisition directly restricts the imaging speed of this kind of new imaging methods and limits its application in the field of higher requirements for imaging speed.In addition,the intensive array production process is difficult and expensive.Therefore,it has great significance to reduce the amount of data collection and improve data collection speed.However,there are a large number of streak artifacts in the photoacoustic images through the traditional reconstruction method with undersampled data,which greatly reduce the image quality.Undersampled noise characterized by inhomogeneous structure,and under the condition of the large scale undersampling,the size and shape of artifacts amplitude is similar with vascular signal,so the common denoising method is hardly to remove the noise.Therefore,in this paper,we study the method of image denoising based on the dictionary learning for unsampled photoacoustic image,and prove the superiority of the dictionary learning algorithm in the removal of the unsampled photoacoustic image reconstruction artifacts.In this thesis,there are three main research contents:(1)the image denoise method based on the traditional Dictionary Learning algorithm(DL).This method can effectively remove most of the noise and artifacts while retaining the clarity of the original signal.(2)Photoacoustic image denoise based on the spatial-domain Dual Dictionary Learning(DDL).This method produces a kind of double dictionary.Half of the dictionary is a featured dictionary with image feature information obtained from signal training.The other half is the noisy dictionary which contains image noise information through image noises and artifacts training.This process of denoising is realized by making zero of the sparse coefficient of noises.The double-dictionary learning algorithm can remove most of the noise and artifacts quickly and effectively,and retain the original signal clarity as much as possible.(3)Wavelet domain Dual Dictionary Learning algorithm(WDDL).In different subbands of wavelet domain,this method produces signal dictionary and noise dictionary,and the sparse representation based on their respective double dictionaries is carried out in each subband.Then the image is processed by the inverse wavelet transform.Compared with the previous two methods,the wavelet double dictionary can better remove the artifacts and noises while retaining the signal.In order to verify these methods of photoacoustic image noise reduction proposed in this paper,the imaging experiments of blood vessels in living mice and blood vessels of hand were carried out.The experimental results show that the double-dictionary learning method can effectively remove the artifacts in the undersampled photoacoustic image,and reduce the data sampling rate and improve the imaging speed while ensuring the image quality.The study provides the undersampling photoacoustic imaging algorithm support which can reduce the PACT data sampling rate and production cost of ultrasonic array,and effectively promote the photoacoustic imaging in a variety of applications in the field of biomedical research.