Session Index

-

  Preview abstract

We have developed a multi-scale OCM imaging system at a 1.7-μm regime enabling multi-scale imaging of the ex vivo oral precancerous tissue. The light source had a sweep rate of 90 kHz. Objectives with three different magnifications are mounted to a power turret, enabling seamless change of the OCT imaging resolution via the software control with up to 5.3 μm transverse resolution. Most importantly, we demonstrated the ability of our custom-developed volumetric mosaic stitching and processing algorithm applicable for aforementioned multi-scale OCM datasets in an fully automatic fashion without requiring manual adjustment of individual parameters.

 

In this study, we have developed a catheter-based OCT system with a polarization diversity detection apparatus allowing volumetric imaging of the cervix, including the endocervical canal. The rotary junction and catheter are custom-designed to tailor the aforementioned biomedical applications. Preliminary imaging results of the human fingertip and palm skin were shown to validate the imaging capability of the developed imaging system. In order to suppress the polarization artifact present in the volumetric OCT images, a custom-designed polarization diversity detection apparatus was designed and implemented to assure the optimum detection of OCT signals on the circumference as well.

 

Recently, the functional extension of optical coherence tomography (OCT) with the OCT angiography (OCTA) technique provides volumetric imaging of the subsurface microvasculature of the imaging tissue without requiring exogenous contrast agents. In this study, we have developed a small-footprint imaging platform with the MEMS scanning technologies, allowing volumetric imaging of the mouse brain microvasculature at different physiological states, including anesthesia, waking, and movement. In addition, in order to better examine the changes of microvasculature among different aforementioned states, we have employed three different OCTA algorithms, including decorrelation (DM-OCTA), speckle variance (SV-OCTA), and complex differential variation (CDV-OCTA).

 

In vivo cellular imaging has been highly desired for research use and practical diagnosis. Ultrahigh resolution optical coherence tomography (UHR-OCT), generally refers to 1–2 μm or even higher spatial resolution in tissue, has been considered a critical role on in vivo cellular or subcellular level imaging. In this study, we have developed an UHR-OCT system with a mode-locked laser pumped supercontinuum laser (ML-SCL), achieving ~1.8 μm on both axial and lateral resolution in biological tissues. This system is then implemented on mouse cornea and human skin imaging, demonstrating its feasibility of in vivo studies and clinical applications.



 

Neural canal opening (NCO) points are important landmarks of the retinal pigment epithelium (RPE) layer in the optic nerve head (ONH) region. How to identify the NCO points is a challenging issue by means of conventional image processing. In this study, we proposed a deep learning-based scenario for two purposes: one is to check the existence of NCO points as the binary classification task, and the other is to find the NCO location as the detection task. Proposed algorithm was validated to yield more stable NCO points detection under a series of testing OCT images.

 

-

  Preview abstract

Quantitative differential phase contrast (qDPC) microscopy provides phase information with asymmetric illumination, which enables label-free imaging. However, typical qDPC microscopy requires twenty-four intensity measurements to reconstruct a single phase image, while motion artifacts and acquisition time are important issues for live cell imaging. We propose dual-color coded qDPC microscopy with patch-wised U-net to minimize required frames. To demonstrate the performance of the system, live cell time-lapse imaging is executed.

  Preview abstract

X-ray luminescence computed tomography (XLCT) provides new possibility for molecular imaging through X-rays with nano-luminescent particles which can excite visible light. In this article, Filtered Back Projection (FBP) method is employed to reconstruct the optical data of luminescent tomography through visible light at 24 angles in free space. Furthermore, the simultaneous algebraic reconstruction technique (SART) is applied in the reconstruction process as an iterative method. At the end of the article, we will actually reconstruct a tablet-shaped light-emitting prosthesis.

 

A non-invasive imaging technique, diffuse optical tomography (DOT), quantifies the optical properties of biological tissue. It is ill-posed and ill-conditioned due to the diffusive nature of light propagation in biological tissues. In this study, a 2D convolution neural network is employed to build up an image reconstruction network that tackles the above issues of DOT. To stabilize the distribution of input(radiance) to a given network layer during training, we introduce batch normalization layers to the network which also train our network quicker and more reliably.

 

It is desirable to optimize the imaging depth in biomedical imaging, which offers the possibilities of investigating the structural changes from diseases or pathologies hidden at deeper tissue locations. In this work, we qualitatively analyzed the depolarization effect and imaging depth as using different polarization states in second harmonic imaging.

  Preview abstract

Discriminating type I and II collagen is important as several diseases are known to alter the properties of these collagen structures in bone and cartilage. Dual-liquid crystal based polarization-resolved second harmonic (SHG) microscopy is utilized for quantitative characterization of collagen types I and II in fracture healing tissues via the pitch angle analysis according to 18 different polarization-SHG images. Furthermore, data reliability is ensured by using right and left-hand circular polarization imaging centered circular dichroism analysis. This approach presents a promising tool for differentiating collagen types at the molecular scale and for understanding the role of collagen in ECM structure.

 

In our cone-beam X-ray luminescence computed tomography (CB-XLCT) system, luminescent optical imaging was affected by lens distortion, which caused great difficulty in the direct interpretation of the image. When the magnification of the lens decreases with axial distance causing, radial distortion occurs, causing each image point to move radially towards the center of the image. Thus, it is essential to use Geometric calibration to correct the image. Therefore, Fitzgibbon’s division model will be employed to solve the initial values of distortion for presentation of real data.

 

We optimized the spectral coverage of self-phase-modulation-enabled femtosecond fiber sources by careful investigations into the influence of input pulse width, fiber length, and fiber damage, and we have demonstrated a widely tunable source ranging from 740-1250 nm for two-photon microscopy applications. In addition, tens of milliwatt tunable near-UV/visible spectrum is easily obtained by a frequency-doubled conversion, and the gap between the fundamental and frequency-doubled spectra can be filled with the nonlinear wave breaking around 650 nm. A multi-modality microscopy incorporating two-photon microscopy and confocal fluorescence microscopy was also demonstrated to prove the versatility of our development for biomedical imaging.

 

To image three-dimensional (3D) living organisms, optical sectioning and high-speed image acquisition are required. Using Bessel light sheet microscopy, we experimentally demonstrate 3D real-time fluorescence imaging of Caenorhabditis elegans (C. elegans). High resolution images obtained with our system can resolve fine structural details to perform developmental biological studies.

 

Isotropic quantitative differential phase contrast microscopy (iDPC) requires an optical system, electronic hardware to control illumination, and computational post-processing to reconstruct quantitative phase images. To make the system compact and portable, an efficient strategy is required. Here, we propose FPGA based iDPC microscopy which is designed and developed for biomedical imaging. The FPGA system is used to perform three basic operations: illumination control, image acquisition, and image reconstruction. Our system can provide high speed phase contrast images suitable for label-free live cell imaging.

  Preview abstract

The measurement of soft tissue in biological body is a known problem. To solve this problem, we developed a Cone-beam X-ray luminescence computed tomography (CB-XLCT) system, a dual modality technique, which can measure the hard and soft issue at the same time. To reconstruct the optical data of luminescent tomography, we employed Filtered Back Projection (FBP) method, and we made comparison between the constructions of soft tissue by three iterative reconstruction techniques: Algebraic Reconstruction Technique (ART), Simultaneous Algebraic Reconstruction Technique (SART) and Simultaneous Iterative Reconstruction Technique (SIRT).

 

We have achieved video-rate three-photon fluorescence microscopy based on a 1320-nm femtosecond source, driven by a single 24-MHz Cr:forsterite oscillator and fiber-optic nonlinear conversion. This demonstration provides a solution for robust a source system for deep-tissue GFP imaging. We optimized the laser oscillator to deliver 40-nJ output pulse energy at around 1260 nm, and the matching of the output wavelength and GFP excitation can be realized by precisely managing the interplay between self-phase modulation and fiber dispersion. We have obtained clear video-rate three-photon images from fluorescent beads, and further optimization toward deep-tissue GFP imaging will be discussed.

 

Without additional optical delay line and complicated dispersion compensation, we investigated the possible fiber-optic nonlinear process to realize simultaneous multicolor two-photon fluorescence imaging. Precise generation of the spectral broadening enabled by self-phase modulation and dispersive wave generation delivers two fully compressible femtosecond pulses, and the relative delay between the pulses can be controlled under different input conditions. Based on a 24MHz Cr:forsterite laser and proper fiber-optic nonlinear conversion, we obtained wide spectral coverages of 900-1200 nm and 700-1200 nm using different fibers. We also demonstrated simultaneous video-rate observation of cyan, green, and red fluorescent beads with controllable sum-frequency excitation.

 

In our system, temporal focusing multiphoton excitation microscopy (TFMPEM), which can excite a large area at once, also can provide high frame rates capability, and is faster than using point-scanning method. But with this characteristic, it cannot excite all area equally distributed, and the scattering problem in bio-tissue is the deeper the worse. So we registered and restored TFMPEM images by using 3D U-Net architecture via the point-scanning images as the ground truth. This study demonstrated 100 frame rates and improved image quality from 8.7 dB to 32.7 dB.

 

Digital image stitching/mosaicking is a basic requirement for a state-of-the-art nonlinear optical mesoscope (NLOM) while imaging a centimeter-scale tissue sample, especially when the NLOM field-of-view (FOV) is limited, owing to a moderate or high numerical aperture (NA) objective lens. For an enhanced effective scanning speed, such stitching operations are expected to be real-time so as to eradicate a need of post-acquisition data processing. We report a post-processing-free sub-minute multicolor gigapixel NLOM imaging technique capable of point-scanning an ultra-large tissue sample at a >30 M/s effective pixel rate while securing a submicron digital resolution.

 

Investigating hematoxylin and eosin (H&E) stained images of frozen-sectioned biopsy tissues is a state-of-the-art clinical diagnosis protocol to assess tumor borders during brain surgery. However, a single assessment usually takes at least 30 minutes and actual margins can be lost owing to the process of frozen sectioning. We report a method to rapidly assess glioma surgery borders, where a whole-mount H&E staining protocol is first applied prior to the slide-free nonlinear mesoscope imaging. The protocol greatly enhances the signal-to-noise ratio and image contrast, and helps maximize the effective scanning rate with much reduced imaging time for centimeter square area sizes.

 

Overcoming photodamage has been challenging in bio-imaging. We demonstrated a power-enhancement model in harmonic generation microscopy by deep learning. With the image acquired at low optical power, the model outputs an image at higher optical power close to the ground-truth image. Consequently, the risk of photo-damage can be reduced.

 

Concentration of hemoglobin is an index for healthcare. Complete Blood Count (CBC) is the main way to calculate the concentration of hemoglobin in hospitals. However, there are some disadvantages, such as time consuming, pain and infection. In this study, we develop a non-invasive and hand-held optical system with algorithm to calculate the concentration of hemoglobin of 24 subjects in hospital. The correlation coefficient (r) of hemoglobin between hospital lab and our device is 0.870. Bland-Altman analysis had a deviation of -0.006 g/dL with 95% of values within the limit of agreement of 1.62 g/dL to -1.62 g/dL.

  Preview abstract

A virtual micro-computed tomography subsystem is simulated to validate the ability of X-ray excited luminescence tomography by a Monte-Carlo tool. When nanophosphor is irradiated by X-rays, 500-700 nm luminescent light will be excited. We implemented an AI denoiser to improve CT image quality to be used for small animal imaging.

 

Functional oxygen saturation (SaO2) is an important physiological parameter which estimate a person carries enough oxygen in blood. Pulse oximeter has been broadly used as a non-invasive diagnostic tool in many conditions, e.g., surgical procedures, anemic and injuries. However, the accuracy of pulse oximeters can be affected by many factors, such as low perfusion. This study aims at analyzing the effect of low perfusion on pulse oximeter. The results shows that the optical signal collected from longer source-detector separation (SDS) is less affected by low perfusion; in contrast, shorter SDS is well affected.

  Preview abstract

This research emits near-infrared light into the human’s wrist and receives the light scattered inside the human’s body. Then we analyze the human body information that contains in it. And we use deep learning to analyze the data more deeply so that we can predict the bone density more accurately. This research uses a neural network called U-net for deep learning. After training, we compare the result with DXA to see whether the training result conforms to DXA. And use the obtain wrist bone density value to predict the bone density value of other parts of the body.

  Preview abstract

We demonstrated a compact, high-speed spectral-domain optical coherence microscopy (SD-OCM) system to observe the tumor cell spheroid. Volumetric OCM images of the cell spheroid were acquired using an in-house C++ interface and used a low-cost microcontroller for triggering to synchronize the galvanometer mirror to the detector array. We designed a hermetic incubator on the microscope stage to control temperature, humidity, CO2 concentration while performing long-term OCM imaging.

 

The detection of human mutated DNA fragments is the key to cancer therapy. Surface-Enhanced Raman spectroscopy (SERS) is merging as a powerful tool to detect DNA molecules. However, SERS DNA biosensors are still rarely found in clinical applications due to their insufficient active area and varying signal intensities.
We use indium gallium nitride quantum wells (InGaN QWs) grown by MOCVD to address the abovementioned issues. This finally results in reliable detection of DNA hybridization events of human mutated DNA segments.
The sensitivity of our substrate is further boosted by adjusting MOCVD conditions towards the clinical SERS DNA biosensor.

 

The of biological tissue and the laser processed material structure at micrometer resolution is hard to obtain the 3D information directly by optical bright field microscope. According to the tissue and material nonlinear optical properties, the multiphoton-excited fluorescence microscopy (MPEFM) could perform noninvasive inspection by reconstructing the high-resolution 3D fluorescence images that could directly see the details in specimen.

  Preview abstract

We use two kinds of deep learning structure to reconstruct image that reduces periodic noise in infrared image, and the two kinds of models are DNCNN, NIRCNN. By slowly decreasing the depth, the image PSNR will improve from 25.71 dB to 25.90 dB.

 

The optical power divider flatness in wavelengths for chip-based optical coherence tomography (OCT) is crucial to higher signal-to-noise ratios. A Mach-Zehnder directional coupler (MZDC) structure could be utilized to smoothly maximize the splitting ratio up to 40-nm wavelengths on a silicon platform with a 0.2 micrometer of decoupler optical path difference. However, a bidirectional coupler is necessary for spectral-domain OCT. Even though a MZDC structure is insensitive to the process variation, its bidirectional wavelength response is not flat. An array waveguide grating in spectral-domain OCT is proposed for a flatness compensation to maintain the axial resolution and sensitivity.

 

The Fano-resonance biosensor is formed through the phase variation of dual microring resonators coupled Mach-Zehnder interferometers. After windowed Fourier transform interrogating two ring interferograms, the biological phase sensitivity could be effectively enhanced and theoretically demonstrate the detection limit as 10^-7 refractive index units.

 

-

  Preview abstract

Zinc ion is the second largest transition metal ion in human body, and the concentration of zinc ion around axons is around 10-6 M. The concentration of zinc ion is a useful indicators for physiological states especially in brain and nerves. Therefore, it is helpful for clinicians’ medical treatment if the concentration of zinc ion in the injured brain and central nervous system can be detected in real time. Thus, we design a minimized fiber-based sensor coated with a hydrogel containing sensing molecules. The fiber-based sensor can be a candidate to detect the concentration of zinc ion in real-time applications.

  Preview abstract

In this work, the metal-oxide TFT (thin-film transistor) biosensor integrated with a microfluidic channel was developed to investigate biochemical reaction kinetics. Using the malate-aspartate shuttle as an example, we established the correlation between the molecular diffusion behaviors in the microfluidic channel and the TFT current responses. Mixtures of different reagent concentrations were characterized to extract the ratio between NAD+ and NADH, and thus determining the apparent equilibrium constant. Because the whole analysis was conducted by a TFT sensor fabricated using semiconductor process, it has the advantages of exploring biochemical reaction kinetics in the massively parallel manner.

  Preview abstract

In this work, we propose a new biosensing setup that includes a typical electrochemical(EC) biosensor, a TFT (thin-film transistor) and a constant current source. The integrated circuit provides a negative feedback control to mitigate current fluctuations induced during the voltammetry detection of the EC biosensor. Using levodopa (L-dopa) molecules as the target analyte, our new structure demonstrated a LOD an order of magnitude lower than the traditional EC sensor because of a lower standard deviation among the tested samples

  Preview abstract

By culturing different amounts of PC-3 cells on NbOx/n-si Light-addressable potentiometric sensor (LAPS), the cell metabolism responses can be and quantitatively studied on the basis of a biocompatible LAPS chip with self-developed LAPS LAPS measurement system. The real-time 2D chemical images and PC-V curves can be easily generated automatically for surface charge relative potential monitoring.

 

Since the neurological diseases are difficult to be cured with drugs. In this study,
we built a self-made neurofeedback system as a new physiotherapy and hoped to provide
effective diagnosis and treatment for Tourette syndrome children with near-infrared
spectroscopy (NIRS). The result shows that the cerebrovascular reactivity had improved.

  Preview abstract

Extracorporeal membrane oxygenation (ECMO) is a medical emergency device combining with respiratory and circulatory support. It is used for acute and severe patients with cardiopulmonary failure. However, the use of the ECMO device requires considerable labor costs and medical resources, so it is very important for clinicians to evaluate the condition of patients and how to use it. Therefore, in this study, Near-infrared spectroscopy (NIRS) was applied to real-time monitoring of the lower limb microcirculation of patients with ECMO, which is helpful to provide clinicians with diagnostic information.

  Preview abstract

We calculated the tissue chromophore values with our diffuse reflectance spectroscopy system to distinguish between healthy sites and gingivitis-infected sites. We found the machine learning method can effectively use for enhancing the classification of healthy and gingivitis-infected sites. As compared to the diagnostic performance of using only gingival tissue functional parameters derived from diffuse reflectance spectroscopy, we found that introducing the machine learning method can increase the accuracy, and specificity from 57.7% to 60%, and 68.6% to 81.3%, respectively, which outperformed the general dichotomy to distinguish between healthy sites and gingivitis-infected sites.

  Preview abstract

This study proposes a course of action and method for wound sterilization using ultraviolet (UV) light, specifically with the use of UV light-emitting diodes (LEDs), and with particular emphasis on multi-stage sterilization steps and methods to avoid surface damage caused by UV irradiation. During the irradiation process, we also found some tenacious bacteria will form a bio-membrane (Biofilm) above the wound to protect themselves from being destroyed by the UV rays. Therefore, a debridement step can be included to remove biofilm to ensure the effectiveness of the bactericidal UV-LED.

  Preview abstract

The study addresses the integration of previous in-house developed image reconstruction schemes for diffuse optical tomography. Two phantom designs were employed to verify the image reconstruction performance compared with former results.

 

This study is mainly used to detect pleural effusion through solar cell biosensor chips. It can simplify the complex techniques and reduce waiting time required for current detection, and use this element to detect the degree of carcinogenesis of lung cancer cells and analyze them.The study used a solar cell biosensor wafer and a serrated finger electrode prepared on the surface of the ITO substrate for measurement, thereby analyzing lung cancer cell types.

  Preview abstract

This study uses retinopathy caused by diabetes to analyze the oxygen content in blood vessels. The hyperspectral imaging technology is used to establish the spectral data, and then the principal component score map of the variable veins of retinopathy is obtained through principal component analysis, and the arteries and veins are identified.

  Preview abstract

This research mainly prepares molybdenum disulfide photochemical biosensor. We will use this sensor chip to detect and analyze the degree of canceration of lung cancer cells.In the research, molybdenum disulfide is grown on the light-absorbing layer of the silicon-based solar element, combined with the self-designed Serrated Interdigitated Elec-trode. The photoelectric flow measurement was performed on three different types of lung cancer cell clinical samples CL1 lung adenocarcinoma cell line NCI-H460 lung large cell carcinoma cell line NCI-H520 lung squamous cell carcinoma through the photo-generated charge carrier transport mechanism.


  Preview abstract

Combining hyperspectral imaging and deep learning to detect early esophageal cancer. Using visible light hyperspectral imaging technology to give hyperspectral information. Through the algorithm, the spectrum information of esophageal cancer is distinguished, and the ability to identify the lesion is achieved without relying on image changes.

  Preview abstract

This research has chosen to use the idea of semantic segmentation for more precise location. U-Net is used as the basic structure, and Resnet is used as the basic structure. The ability to extract feature maps to predict the classification and location of esophageal cancer.

  Preview abstract

This study uses the combination of hyperspectral imaging technology and band selection to simulate white light endoscopy images as narrowband endoscopy images. Such images are called hyperspectral images, which will make the blood vessel characteristics in the endoscopy images more obvious.

  Preview abstract

We demonstrate a novel proximal scanning mico-endoscope based on a rectangular fiber. In the novel mico-endoscope, a proximal scanner was used to reduce the complexity in packaging the distal end. The fiber-nonlinearity enhanced dispersion-induced pulse broadening was avoided. Finally, the wave-vectors can be continuous tuned to avoid pixelike images.

  Preview abstract

The changes of glucose concentrations are measured in the scattering pattern in space by using a single laser. We analyze the light intensity of each spectrum and select the light intensity range related to the glucose concentration in data processing.

 

An improved system calibration method for single photon emission microscope is developed. The method combines the data from the geometric calibration and grid-scan experiment to fit the system geometric parameters. Also, the imaging model adopts non-circularly symmetric point response functions to build the system matrix essential in the tomographic reconstruction.

 

Monitoring sleep and bed occupancy are essential for eldercare. A well-placed passive infrared motion sensor PIR provides additional information that can be useful in understanding differences between the other techniques.

 

The electric field Monte Carlo method can be used to include the phase and polarization state information of complex light in turbid media. Based on the EMC framework, we investigate the propagation properties of the Bessel beam over the penetration depth in a turbid medium for biomedical applications.

 

We report on using multimodal nonlinear optical (MNLO) microscopy with the image contrast of second harmonic generation (SHG) and coherent anti-Stokes Raman scattering (CARS) to investigate the structures of bone-engineered biomaterials. By analyzing the nonlinear signals on images, it is possible to characterize and quantify the structural variations in the prepared collagen scaffolds, which are processed with different crosslinking methods. The results show that the correlation between signals and mechanical properties can be established. And the comparative analysis of these two signals can be used to monitor the structural (triple-helix) change of collagen molecules during the crosslinking processes.

  Preview abstract

To improve work quality of indoor workers, physiological responses such as EEG and heart rate variability which were influenced by the polychromatic lighting with different compositions of blue light were analyzed in this study. In addition, the correlations between the physiological responses and characteristics of blue light are also discussed.

  Preview abstract

In this study, we developed a simple fluorescence sensing system to rapidly measure the total carbonyl stress in urine because carbonyls may form advanced glycation end products, leading to diseases. 20 µL rat urine samples added to a carrier were derivatized with 5,6-diamino-2,4-hydroxypyrimidine sulfate dihydrate for testing, and only 1 s was required. In addition, the correlation coefficient between the total carbonyl content detected by the simple fluorescence system and HPLC reached 0.8132. Collectively, our findings demonstrated that our system can rapidly measure total carbonyl stress in urine with a small sample volume.

 

We develop computer-code assisted segmentation and outlining of the sectional time-lapse images acquired from a beating heart of larval zebrafish for automatic determination of the cardiac function zebrafish at different temperature. The approach will facilitate using zebrafish models for fundamental cardiac research and screening drugs of cardiac toxicity.

  Preview abstract

This research is expected to use near-infrared spectroscopy during the mental stress task to measure the changes of the blood oxygen changes in the prefrontal cortex of 13 healthy subjects, 9 chronic migraine patients, and 12 medication-overuse headaches patients. The blood oxygen information was processed and then fed into machine learning. The sensitivity of chronic migraine and specificity of medication-overuse headache reached 100 %, and the specificity of chronic migraine and sensitivity of medication-overuse headache reached 75 %. The results prove that functional near-infrared spectroscopy combines with machine learning is feasible for the migraine classification.

  Preview abstract

n this research, we apply a NIRS imaging system to take images and a deep learning model with X-ray images to find the relation between thigh muscle and diagnosis of sarcopenia.The accuracy of our results prove the significant relation and the feasibility of diagnosing sarcopenia with deep learning models. Our NIRS system is able to take photos of thinner body parts such as wrists. This research explores X-ray images in sarcopenia as well as a method to take images through NIRS and will combine both as the future work.


  Preview abstract

We employed the pseudospectral time-domain (PSTD) method to simulate the macroscopic scattering phenomenon. A series of circular dielectric cylinders were used for modeling, and a wide range of light sources as input to operate. In this study, we modified the optical signal and medium to explore the light through the sclera medium.

  Preview abstract

The finite-difference time-domain (FDTD) method is employed to simulate the structure of the chameleon skin, which is composed of nanocrystals, that is, photonic crystals. By changing the arrangement of the iridophore cells, we analyze the influence of the structural material properties on the color of the biological surface. Furthermore, we compare the difference between the simulation results and the real biological spectrum and further discuss the error and accuracy.

  Preview abstract

The photon path length can be controlled thorough the spatial frequency domain approach. Based on this principle, we developed a dual wavelength Mueller matrix imaging system with structured illumination technique and demonstrated depth-resolved polarization feature images of lipoplus and pork loin.