Advanced Methodologies For Time-resolved Optical Breast Imaging Based On Panel Detection Mode

The research on how to perform more accurate regular examination of breast has become one of the focused topics. Traditional X-ray mammography and ul-trasonic inspection have shown some deficiencies in sensitivity, specificity, security and comfortableness. As a new noninvasive optical measurement technique, op-tical mammography of diffuse optical tomography (DOT) has attracted a lot of attentions for providing a richer functional image. However, due to diffusive nature of NIR light in tissue, DOT has relatively poor spatial resolution and quantita-tive ness. In practice, an applicable breast imaging technology should detect down to I centimeter tumor in size. If not,the mortality of breast cancer will grow exponentially. Meanwhile, the precise quantification of the reconstructed opti-cal properties of breast tissue, is to determine the benign and malignant tumors. In view of this, how to improve spatial resolution and quantitative ness of opti-cal breast imaging technique is an important job in our research. In this thesis, based on the developed numerical model of diffusion equation, a series of novel approaches to improve the optical breast imaging technology are proposed here to improve the reconstructed image quality.To cope with the computation and storage problem for some applications, such as breast tumor diagnosis, it is preferable to develop a subdomain based parallel computation scheme. Schwarz-type domain decomposition for solving the forward model in parallel, splits the global model into several overlapping subdomains. Then all the sub-domain computing tasks are assigned to multiple processors in parallel computing, which greatly improves the computational speed.We introduce a multi-level scheme that uses discrete wavelet transform to reach different resolution level of wavelet domain decomposition and in some de-gree reduce the number of unknown optical parameters at coarse resolution levels and alleviate the ill-posedness of inverse problem. Compared with previous tra-ditional full spatial domain voxel based algorithm, this method can efficiently improve the reconstruction quality and spatial resolution.Because fiber-based single-photon detection system usually has a relatively low spatial sampling rate. And improving the hardware is very expensive. To deal with this problem, an expanding measurement space scheme is presented that uses cubic spline interpolation to obtain more measurements from less optodes to alleviate the problem that the number of boundary measurements data is generally far fewer than that of unknown parameters to be reconstructed in reconstructing process. Compared with traditional algorithm, this method can efficiently improve the reconstruction quality and spatial resolution.The image reconstruction in diffuse optical tomography is regarded as a large-scale, nonlinear optimization problem, whose solution strategy is usually based on a Newton-like method involving the construction and inversion of the Jacobian matrix. Whenever the scale of the inverse problem is large, it becomes intractable. To overcome this problem,algebraic reconstruction techniques (ART), an iterative solving method based on row action, is proposed to reduce computational effort but result in poor image quality. In this paper, conjugate gradient (CG) method is employed here, which requires only the computation of the objective function gradient to avoid calculating and storing the whole Jacobian matrix. Step length is determined by inexact line search method. Compared with ART, we find that the CG method performs better than ART at the image quality, convergence and the speed of it.The widely used ART ignores the link between the rows, making the solutions tend to converge to a local solution, which is usually not satisfactory reconstruc-tion results. To solve this problem, a Jacobian matrix decomposition algorithm is proposed to split the Jacobian matrix along the "column" direction into sev-eral small blocks, and in turn perform singular value decomposition (SVD) based Levenberg-Marquart optimization methods to solve it. This method can improve the image reconstruction quality.A set of experimental system of time-domain DOT is set up based on the time resolved panel-detection mode to validate the algorithms. A kind of solid optical phantom for breast imaging is developed and designed.