Research On Quantitative Phase Microscopy Of Human Blood Cells Based On Digital Holography

In blood analysis,the quantitative measurement of morphological and dynamic parameters related to human blood cells holds significant importance in biological research and clinical medicine.Traditional imaging methods have several drawbacks,such as bright-field microscopy only allowing for qualitative examination of blood cells,and fluorescence microscopy techniques potentially causing some damage to blood cells due to staining.Quantitative phase microscopy based on digital holography has become an important research tool for quantitative measurement of blood cells morphology and parameters due to its capabilities of label-free,non-damage,fast quantitative imaging.However,it still faces challenges such as the reliance on specialized optical components,limited imaging accuracy,and time-consuming phase unwrapping and reconstruction.This dissertation focuses on the requirements of dynamic imaging of live blood cells and conducts research on quantitative phase microscopy with good real-time performance,particularly synchronous phase-shifting and common-path off-axis digital holography systems.By optimizing the optical setup of the synchronous phase-shifting digital holography system and the phase unwrapping algorithm,simplifying the phase reconstruction process of common-path off-axis digital holography,we have effectively improved the accuracy and efficiency of quantitative phase microscopy.Furthermore,we have established corresponding digital holography systems for experimental validation using standard samples and live blood cells.The research contents and main achievements of this dissertation are as follows:1.By integrating polarized light and Jones matrix theory and employing common optical components,a novel method for four-phase synchronous phase-shift digital holography is proposed.This approach can effectively address issues present in synchronous phase-shifting digital holography,such as reduced image accuracy due to inconsistencies in the optical path between object and reference waves,as well as the reliance on specialized optical elements that compromise a portion of the camera’s field of view and resolution.This method is built upon a Mach-Zehnder interferometer and utilizes readily available optical components,including beam splitters,waveplates,and polarizers,to simultaneously capture four phase-shifting digital holograms with a wide field of view and high resolution.And by placing a quarter-wave plate in the object wave path identical to that in the reference wave path to make the optical path of the object wave and reference wave the same,avoid introducing additional phase changes in the light field and ensure the phase image accuracy.Experimental results demonstrate that the method proposed in this dissertation exhibits a high level of sensitivity and stability.The system has a spatial noise of 2.42 nm and a temporal noise of 12.67 nm.Moreover,it enables real-time high-precision dynamic phase imaging and quantitative measurement of standard samples with a diameter of 2 μm and human blood cells.The standard samples thickness measurement error is 0.04 μm.2.A convolutional neural network VDE-Net based on the weighted jump-edge and mask attention mechanism is proposed,which can effectively address the challenges associated with the traditional phase unwrapping methods in four-phase synchronous phase-shift digital holography.These challenges include time-consuming phase reconstruction,low efficiency,and poor generalization ability.VDE-Net incorporates a mask into the penultimate feature maps of the network,formed by multiplying the wrapped phase image with the weighted jump-edge image.This integration suppresses features from background regions during the feature extraction process,thereby directing the focus of the convolutional neural network towards the jump-edge within the wrapped phase image,enhancing its feature extraction capability.Experimental results demonstrate that VDE-Net achieves an average time of 8.02 ms for unwrapping a single wrapped phase image,which is one fifty-nine of the time required by the traditional phase unwrapping method.This represents a significant enhancement in phase unwrapping speed.Additionally,VDE-Net achieves an average structural similarity of 0.9975 and an average phase unwrapping accuracy of 92.74%on a simulated data test set,which is 1.02% higher than the accuracy of the mainstream model combined with the attention mechanism.At the same time,this model exhibits strong versatility and can be directly applied to phase unwrapping of four-phase synchronous phase-shift digital holograms of human blood cells.It achieves an average structural similarity of 0.9571,which is 20.19% higher than the mainstream model combined with the attention mechanism.3.A deep learning-based white-light diffraction phase microscopy is proposed.This method belongs to the common-path off-axis digital holography and has higher sensitivity and stability than the four-phase synchronous phase-shift digital holography.However,the phase reconstruction process in common-path off-axis digital holography typically involves numerous steps,substantial computational requirements,and extended processing times,resulting in inefficient phase reconstruction.Based on the traditional white-light diffraction phase microscopy,this dissertation designs a convolutional neural network VY-Net,which can quickly reconstruct blood cells phase image from a single digital hologram of human blood cells without the need for phase unwrapping and phase compensation,simplifying phase reconstruction process.VY-Net consists of two improved image generation networks: V-Net that generates background calibration images and Y-Net that generates phase images.The background information provided by the images generated by V-Net significantly enhances the quality of phase reconstruction performed by Y-Net.Experimental results demonstrate that the white-light diffraction phase microscopy system achieves a spatial noise of 0.88 nm and a temporal noise of 1.24 nm,reducing noise levels by 1.54 nm and 11.43 nm,respectively,compared to four-phase synchronous phase-shift digital holography system.Additionally,the average phase reconstruction time is shortened to43.9 ms,meeting the requirements for real-time high-precision quantitative imaging of human blood cells.Moreover,VY-Net exhibits an average structural similarity of0.9309 and an average reconstruction accuracy of 91.54% on the blood cells image test set,which is 2.41% higher than the accuracy of the mainstream phase reconstruction model.The average structural similarity of phase reconstruction of leukocyte images alone is 0.8639,which is 36.58% higher than the mainstream phase reconstruction model.In summary,the imaging and phase reconstruction methods of digital holography proposed in this dissertation have significantly improved imaging accuracy and speed.They provide more precise and real-time techniques and means for quantitative imaging of human blood cells,holding significant potential for biomedical applications.