| Optical coherence tomography(OCT)has been widely used in the biological imaging field since it was developed in the early 1990 s due to its high resolution,high imaging speed,and noninvasive imaging modality.In more and more clinical applications,such as photoreceptors imaging,tumor margin assessment,and some other fields,cellular imaging has become a new trend to achieve a more accurate diagnosis and effective treatment.At the present stage,the main improvement methods for both axial and transverse resolutions are based on hardware systems,which will increase the cost of the system making the system more complex at the main time,hindering the wider applications of high-resolution OCT system.Therefore,on the premise of not replacing light sources or adding additional hardwares,an in-depth study on the computational image quality enhancement method of OCT is carried out and the auxiliary diagnosis method in clinical application is explored.The main research contents and innovation points of this thesis are as follows:Aiming at the computational improvement of axial resolution,a precise assembly scheme of the galvanometer-based scanner in the sample arm of SD-OCT is first developed.Through geometric analyses and model building in Zemax,the optical path difference(OPD)shift is taken as the criterion to assist the alignment of the galvo mirror,and an iterative process is proposed for the alignment of the galvanometer-based scanner to realize the precise localization of galvo.The dispersion mismatch caused by the OPD shift can be reduced so that the actual resolution of the system can approach the theoretical one across the scanning field.On this basis,an axial resolution improvement method based on the multiple signal classification(MUSIC)algorithm is proposed to further improve the axial resolution without changing the light source of the SD-OCT.An air wedge composed of cover glasses is used to verify the proposed method.The experimental results showed that the MUSIC can achieve a 2.4-fold increase in axial resolution without changing the light source.Aiming at the computational improvement of transverse resolution,an automated fast computational aberration correction method based on a stochastic parallel gradient descent(SPGD)algorithm is proposed for aberration-corrected imaging without adopting extra adaptive optics hardware components.The SPGD-based computational aberration correction updates the phase filter in each iteration by stochastically choosing the number and order of Zernike terms to keep the image entropy decreasing until it converges.The performance of the proposed method is first demonstrated by a simulated resolution target image.Importantly,the computing time for 12 aberration modes will only increase slightly compared to correction with7 terms,which indicates the high speed for high-order aberration.The feasibility of the SPGD algorithm is demonstrated by correcting the images of samples with different scattering properties.The images after aberration correction all showed a significant improvement in contrast and detailed structure information.The proposed SPGD method shows good run-time performance due to the parallel computation contributing to great potential for future real-time aberration-corrected imaging in clinical applications.In order to achieve high-quality imaging and accurate diagnosis of human breast tissue,the computational improvement methods for imaging quality described above are applied to the spectral domain polarization-sensitive OCT system,and an automated differentiation model for classifying breast tissue was proposed.The differentiation model is trained for classifying malignant tumor,fibro-adipose,and stroma by the features extracted from both intensity metric and four polarization metrics.Taking the hematoxylin and eosin(H&E)histology-confirmed diagnoses as the gold standard ground-truth,the robustness of the model is further validated after further being optimized by forward searching and validated by the leave-one-site-outcross-validation(LOSOCV)method.Since the intensity information and the polarization information are used in the diagnostic model,the overall accuracy of the proposed model can reach 93.5%.This demonstrates that with additional polarization information,more information about the tissue can be acquired and utilized by multiple numerical metrics,illustrating the potential of polarization OCT for enhanced breast cancer detection. |
