Projects

Year 2024

Ultrasound Emitter Driven by CW Laser

Mentor: Prof. Chen Yang
Member: Donghao Guo

Efficient ultrasound emission technique has found widespread applications in areas such as industrial cleaning, nondestructive testing, biomedical imaging, and tissue treatment. Optical ultrasound emission technique based on the photoacoustic (PA) effect has been exploited to develop efficient ultrasound transmitters with a small form factor and broad bandwidth. However, conventional techniques based on the photoacoustic effect inevitably required an expensive high-energy short pulsed laser and dedicated preparation of the optically absorptive composite film. So we try to use the much cheaper CW laser to design a ultrasound emitter.

Cavitation, the formation of cavities inside liquids, is what we use to generate ultrasound. We use the fiber to heat the liquid to produce bubbles to generate ultrasound. However, the heating will lead to the convection of the liquid, which may reduce the effective use of the liquid. So we want to explore whether this would affect the strength of the generated signal and so on.

Tapered fiber-based volumetric photoacoustic emitter for neural stimulation

Mentor: Prof. Chen Yang
Member: Hongru Zeng

In the photoacoustic process, absorbers optically absorb laser pulses, generating transient heating and thermal expansion and compression in the surrounding medium, resulting in mechanical waves at the ultrasound frequency. An often-used method to generate PA signal is fiber-based optoacoustic emitter.the diameter of FOE is smaller than the traditional ultrasound transducers, and it can avoid ultrasound heating. So, it is Implantable and can be used for behavior study in live free-running animals.And candle soot is a very efficient material that can used as a coating of FOE, it is 10x higher conversion efficiency compared with FOC. the CSFOE can well perform optoacoustic activation of neurons.

However, current fiber emitters are all fixed single-site stimulation devices, we want to develop a programable volumetric PA emitter with stimulation at tunable locations.Different incident angles will change the number of modes that fiber carries, so I design an optical system that can change the incident angle of the fiber. Besides, the number of modes is also related to the diameter of the fiber, so I choose to use tapered fiber so that the light can be emitted at different sites of fiber.

I successfully realized laser emission at tunable locations by our system, but the PA signal is not ideal enough, we haven’t gotten a tunable single-site emission separately.Next, I will use other coating methods, like the mixture of candle soot and PDMS, and then the PA signal can be generated by the laser with shorter wavelengths. We will also use our TFOE to do PA neurons or brain stimulation.

Short-wave Infrared Photothermal Optical Coherence Tomography (SWIP-OCT)

Mentor: Prof. Jixin Cheng
Member: Jinnan Chen

A powerful non-contact imaging technology is quite important, especially in the medical field. And OCT provides such an effective solution. By measuring the interference signals at different depths, OCT can generate cross-sectional images of different depths within the sample. However, current OCT still have limitations, it focuses on the optical properties of the sample but has no chemical information.

In order to fill this gap, we come up with the idea of SWIP-OCT. By combining SWIP and OCT together, we can get a high-speed epi-detected deep-tissue vibrational imaging with chemical composition information.

The main task in the past several weeks of my project was to build the entire SWIP-OCT system and carry out debugging with Hongli Ni. Specific tasks included setting up and collimating the optical path, optimizing the beam shape, synchronizing different devices, and identifying signals. Currently, the hardware setup is almost completed, and the feasibility of the solution has been verified.

Then we plan to determine various characteristics of the system such as resolution and sensitivity by imaging specific samples like 3D scattering phantom made with PMMA beads, agar and intralipids, or some tissue from mice.

Protein Assembly by Fluorescence Detected Mid-infrared Photothermal Imaging

Mentor: Prof. Jixin Cheng
Member: Rong Tang

Protein assemblies range from stable and solid aggregates (amyloid fibrils), to liquid and dynamic condensates. The transition from condensate to amyloid fibrils has been hypothesized to be the underlying mechanism in various neurodegenerative diseases. However, existing methods either do not provide structural information, or are not applicable to both aggregates and condensates. Here, we utilize a fluorescence detected mid-infrared photothermal (F-MIP) microscope to map the secondary structure of protein assemblies in live cells by detecting the infrared modulated fluorescence intensity change. By comparing F-MIP spectra of GFP-tagged Huntingtin inclusions (as aggregates) and YFP-tagged α-synuclein (as condensates) in live yeasts, we demonstrate that aggregates and condensates could be differentiated by protein secondary structures: aggregates like huntingtin inclusions contain a β-sheet-rich structure, while condensates like α-Synuclein does not show high β-sheet content. Collectively, we showcase the secondary structure information on protein assemblies that could be acquired in live cells by F-MIP, and it holds the potential to unveil the aggregation mechanism in neurodegenerative diseases.

The primary goal of my summer project is to demonstrate the feasibility of FMIP to detect the protein secondary structure difference between aggregates and condensates using model systems in live yeasts. To achieve this, I first conducted literature review on protein assemblies and existing methods, and identified the unique advantage of our approach. To prepare samples for imaging, I built up skills in yeast cultivation and induction of protein expression. I learned the F-MIP system and was able to operate it independently. To overcome challenges of low signal-to-noise ratio and fast photobleaching, I explored different experimental conditions for optimization. I performed data analysis on collected hyperspectral F-MIP imaging stacks to extract and calibrate spectra for aggregates and condensates, followed by the quantification of different protein secondary structures by decomposition.

Next, we will perform further quantification of protein secondary structures using more biological replicas, to gain the statistical power for differentiating between protein aggregates and condensates. Based on the spectral difference, we will continue to explore the feasibility of differentiation using signature wavenumbers to further improve the speed and throughput of our approach.

Video-rate volumetric imaging with Z-splitter

Mentor: Prof. Jerome Mertz
Member: Ruize Zhang

Optical microscopy has long been essential for studying complex 3D biological systems. However, many volumetric imaging techniques are hampered by prolonged acquisition times, which limits their application to high-dynamic samples. Z-splitter-based multifocus widefield microscopy offers a rapid and high-contrast imaging solution by capturing 3D information in a single shot. Despite this advantage, this method still struggles with blurring caused by instantaneous motion in dynamic samples, leading to image degradation. To enhance imaging quality, we propose using LED flickering to effectively ‘freeze’ motion during high-dynamic activities. This approach would significantly improve the clarity of recordings, enabling more precise observation of dynamic biological processes.

Background Removal Algorithm in MIP using Penalized Least Square Smoothing(PLS)

Mentor: Prof. Jixin Cheng
Member: Tianyu Xia

Chemical imaging techniques such as CARS, SRS, MIP are facing the issue of background signal interference, which significantly affects image quality and the accuracy of data analysis. However, Traditional background removal methods, such as Wavelet transform, require human intervention and complicated parameter settings, which leads to low efficiency and accuracy.

In this project, We adopted arPLS(Asymmetrically reweighted penalized least square smoothing) to remove the background. arPLS effectively calculates the signal background by adding penalty weights to the baseline, preventing it from following peak values. We have successfully applied arPLS to extract backgrounds from 3D data, enabling the identification of peaks at different wavenumbers for subsequent statistical comparison.

To enhance performance, we have incorporated additional techniques such as denoising algorithm(BM4D) and drift correction. It is a versatile algorithm that can be applied to various cell types and systems, offering a universal solution for background removal in chemical imaging.

Wide-Field, high resolution image reconstruction through coordinate aware CNN

Mentor: Prof. Lei Tian
Member: Yongkang Cheng

My project focus on image reconstruction problem with great shift-variance due to its wide FOV. I developed patch-wise reconstruction network with coordinate encoding to handle large-scale image reconstruction task with strong shift variance. The new network outperforms conventional deep learning framework in SSIM and PSNR and is highly scalable due to its patch-wise design.

Year 2023

Aqueous Non-thermal (ANT) Mid-infrared Photothermal (MIP) Spectroscopy

Mentor: Prof. Jixin Cheng
Member: Zixian Yang

Mid-infrared photothermal spectroscopy can separate molecules based on their absorption spectra. However, when targeting molecules in an aqueous solution, the technique faces limitations due to the strong water absorption at infrared wavelengths, thereby reducing the detection sensitivity.
To address this challenge, we can use Aqueous Non-thermal (ANT) Mid-infrared Photothermal (MIP) microscopy, which leverages the dependency of the MIP signal on the temperature of the probed medium. By implementing this approach, we can effectively suppress water contributions and significantly enhance the contrast in the imaging process.

The primary focus of my summer research project is to develop methods to suppress the background signal caused by water absorption through precise temperature control. To achieve this, I have constructed a counter-propagation mid-infrared photothermal microscope with temperature control capabilities. During my experiments, I measured the signal level of the infrared spot while continuously lowering the temperature from room temperature to 6 degrees Celsius, observing a 50% reduction in the water signal. Additionally, I utilized fluorescence measurements to obtain a temperature profile within the sample dish, ensuring that the temperature gradient is accurately determined. To prevent heat accumulation, I employed a single pulse of infrared radiation for more precise temperature control.

Next, I will employ the Aqueous Non-thermal MIP system to conduct imaging of cell samples and beads, aiming to demonstrate its effectiveness in enhancing the contrast of targets with absorption peaks in proximity to the water absorption region. Furthermore, I will optimize the temperature control system to achieve even greater precision and stability in maintaining the desired temperature conditions.

A Highly Efficient 3-layer CS/PDMS PA Film for Neuromodulation

Mentor: Prof. Jixin Cheng
Member: Yujia Sun

Photoacoustic(PA) neuromodulation is an emerging technique with non-genetic, non-invasive and high-precision attributes, showing great promise in neuroscience research and clinical applications. However, the current conversion efficiency of optoacoustic composite materials ranges from several thousandths to several hundredth, raising concerns about excessive heat accumulation in biological tissue. Here, we present an approach to enhance PA conversion efficiency using a three-layer structure comprising a mixture of candle soot and PDMS as an absorber layer, a top layer of pure PDMS (PDMS Ⅰ layer) and a backing layer of pure PDMS (PDMS Ⅱ layer). PDMS Ⅰ layer facilitates improved thermal expansion for enhanced PA signal generation, while PDMS Ⅱ layer is designed to contribute to the superposition of ultrasound waves. To optimize performance, COMSOL simulation was employed to determine the optimal thickness of PDMS Ⅰ layer, considering the balance between heat absorption and optoacoustic attenuation. The thickness of PDMS Ⅱ layer was calculated to maximize the superposition effect of forward and backward waves, resulting in the formation of unipolar waves.

The theoretical framework was validated experimentally, demonstrating that the PA conversion efficiency of the three-layer structured CS-PDMS film is twice as high as that achieved with a two-layer structure. This advancement holds potential benefits for applications in neuromodulation research.

Enhancing SRP Signal-to-Noise Ratio Through Dynamic Matched Filtering for Improved Sensitivist

Mentor: Prof. Jixin Cheng
Member: Shanyu Wu

Since invention in 2008, Stimulated Raman Scattering (SRS) has been developed into a very advanced chemical imaging modality with huge potential in applications. However, the sensitivity of SRS is fundamentally limited by the intrinsically small Raman scattering cross-section, bringing bottle neck for further advancement of the technique. Stimulated Raman photothermal (SRP) microscopy has been reported to overcome the sensitivity limit of SRS. By measuring the thermal effect of SRS process with a third probe beam, a 500-fold modulation depth improvement and 17-fold solution sample limit of detection improvement have been achieved with SRP.

Data analysis is a key step towards fully realize the sensitivity potential of SRP. In SRP, the complex frequency features as well as spatially heterogeneity of the signal pose challenges for any existing data analysis algorithm. To overcome these challenges, we proposed dynamic matched filtering (DMF) method.While matched filtering is designed to detect specific signals within received signals by correlating them with templates, maximizing signal-to-noise ratio (SNR) in the presence of white noise, the fixed-template approach of conventional matched filtering falls short in meeting the processing demands of SRP signals. Therefore, we propose DMF, which individually processes each pixel’s single period and dynamically generates unique decay signal models as templates for matching filtering. With this algorithm, we expect to maximize the SNR we can possibly extract from the photothermal signal curve.

We plan to validate the performance of DMF algorithm with LOD as test bed. A comparison between DMF and conventional lock-in measurement will be performed. Afterward, this algorithm will be further demonstrated in SRP imaging of cell samples.

Microphotoacoustic Emitter for High-precision Neuromodulation Fabricated through Photothermal Curing of PDMS

Mentor: Prof. Chen Yang
Member: Zhuqin Xu

High-precision neuromodulation plays a crucial role in advancing neuroscience knowledge and providing clinical treatments. In this work, we present a novel approach utilizing the photothermal effect to fabricate micron-level structures on the tip of a tapered fiber and make it into a photoacoustic emitter (TFOE) for neural stimulation with a resolution of approximately 30 μm. The TFOE fabricated through this new method could generate a stable and effective ultrasound with a pressure of about 0.4Mpa and a central frequency of 5MHz. We optimized the design of the TFOE by carefully selecting the materials for the absorber and optoacoustic transducer, as well as controlling the thickness of the layered structure.Compared to other fabrication protocols, our proposed process, involving fiber tapering, candle soot coating and PDMS coating, is more replicable and controllable.In the final PDMS coating step, the thickness of the PDMS layer can be adjusted by varying laser power and curing time, enabling us to modulate the central frequency and peak pressure of the ultrasound signal generated by TFOE. Successful neurostimulation by TFOE was validated by calcium (Ca²⁺) imaging. By optimizing and improving the PA efficiency of TFOE, our approach can use lower power to stimulate neurons, reducing the possible thermal toxicity. Moreover, we developed a mature fabrication process of a pressure-frequency-controllable ultrasound emitter via the photothermal curing process of PDMS. These findings present new opportunities for complex and programmable micrometer-scale neuron stimulation and offer valuable insights for ultrasound neurostimulation research.

3D Reconstruction in Computational Miniature Mesoscope Incorporated with a Scattering Model

Mentor: Prof. Lei Tian
Member: Zhengyi Pan

Light field microscopy (LFM) is a fast and high-resolution 3D imaging technique that records both the spatial and angular information of a specimen in one shot. However, it suffers from degradation due to scattering, which limits its application to living organisms. To address this challenge, we incorporate a recently proposed computationally efficient, incoherent multiscale multiple-scattering model into the 3D reconstruction process of the computational miniature mesoscope, which is a novel LFM system. We investigate the effects of different parameters on the reconstruction quality, such as the angular component selection for gradient descent, and the lower bound of the estimated fluorescence distribution. We also assess the reconstruction performance on a scattering phantom. We demonstrate that the integrated scattering model can enhance the 3D reconstruction of the mesoscope, but further exploration and optimization are needed.

Simulation of Different Fiber Designs on Topological Confined Modes (TCM) Count

Mentor: Prof. Siddarth Ramachandran
Member: Cheng peng

Topological confined mode (TCM) is a special kind of mode that allows OAM beams to propagate in fibers beneath the cut-off refractive index in relatively not-so-lossy conditions. However, the multi-mode transmission would suffer from cross-talk between high order L, m=1 mode OAM and high order m, low order L as their refractive index become so close. We want to figure out how fiber designs would affect the TCM regime where OAM beams could transmit stably and be ‘lossless’, while it would also not be bothered by the high-order m OAMs through simulations. Step-index fibers and ring-core fibers are considered and we test the delta refractive index between cladding and core based on the previous finding.

Year 2019

Micromachining using femtosecond laser

Mentor: Prof. Michelle Sander
Member: Danchen Jia

2D materials with unique physical properties have shown great potential in a variety of research fields ranging from optoelectronics to biological engineering. Our project aims to study the dynamics of 2D material ablation and defect generation using femtosecond laser sources to pattern its surface.

We have successfully achieved clean ablation and defect generation at the focus of the laser, which enables tight focusing of the laser beam and ablation at the diffraction limit. We characterized the optical setup and mechanical stages to allow tight focusing of the laser beam on the sample and studied how changes of the incident fluence affected the micromachining of the sample, by either tuning the laser power or the laser irradiation time. The material properties were analyzed with Raman spectroscopy. This laser micromachining method can also be attractive to pattern other 2D material.

Year 2018

Generation And Application Of Adaptive Bessel Beam

Mentor: Prof. Jixin Cheng
Member: Luzhe Huang

One specific direction which is being explored in our lab is enhancing the efficiency of white LEDs. The main challenge here is developing TiN films, which still maintaining the metallic/plasmonic properties of the films. An Atomic Layer Deposition (ALD) technique is currently under development in our lab, and the fabrication conditions need to be optimized to develop films of high optical quality. In order to confirm that the titanium nitride is optically metallic and to know the location of the plasmon resonance, we need to measure the optical constants e1 and e2.

As we know, the metallic behavior of TiN films exhibits a sensitive dependence on the substrate and deposition details. And my work in Boston University is to measure the optical constants of TiN films with spectroscopic ellipsometer (SE). In order to achieve high accuracy for SE analysis, I build a combined Drude-Lorentz model describing the optical response of the conduction and valence electrons of TiN and TiN_x. Due to high parameter uncertainty in Drude-Lorentz model, we can obtain several different models for one kind of TiN in the certain wavelength range which is decided by the experimental data we used to train this model. But the accurate model should has physical meaning in all wavelength range. So to build a model meaningful, I look into the physical meaning of the Drude and Lorentz expression and analyze the trend of permittivity of TiN. And after we analyze the optical constants we obtained from ellipsometer, we adjust our recipe to produce more metallic TiN and use this sample to verify our model. After all, I will calculate the real and imaginary parts of dielectric constant for TiN thin films.

High Precision Neuro-modulation By Focused Ultrasound Generated By A Fiber Photoacoustic Converter

Mentor: Prof. Jixin Cheng
Member: Yiqing Shen

Our project focuses on generation of focused ultrasound through photoacoustic effect at the tip of fiber with a concaved surface. This tightly focused ultrasound at a fiber tip is a platform technology, which can do precise neuron modulation and precision ablation. Since it is a fiber based device, it is much easier to multiplex compared with traditional ultrasound transducer. Previously, Jay Guo’s team designed and fabricated a concave photoacoustic lens of 6 mm in diameter, achieving a focal spot of 75 μm in lateral and 400 μm in axial widths. Our goal is to design and fabricate a fiber photoacoustic converter of around 1 mm in diameter with a small focus area.

At first, we wanted to fabricate a 1 mm concave photoacoustic lens and fix it with fiber to make a photoacoustic converter. It is very challenging to coat thin (<100μm) photoacoustic films on a 1 mm concave lens with a deviation of around 1 μm, since 1 mm concave lens is so tiny to deal with. So we came up two new ways to build a device generating focused ultrasound. The first one is that we input a small amount of mixture of epoxy and graphite, which serves as a photoacoustic film after cure, into a capillary tube. Then because of the surface energy, a thin film with concave surface is formed. However, using this method, it is difficult to control the positon and size of the ultrasonic focus, since the curvature is decided by characteristics of material. The second way is to fabricate a convex acoustic lens. We have built a stimulation which proved that after passing a convex acoustic lens surrounded by water, an ultrasound plane wave will be focused and the positon and size of the ultrasonic focus can be modulated by changing parameter of the acoustic lens.

Volumetric Stimulated Raman Imaging With A High-speed Deformable Mirror

Mentor: Prof. Jixin Cheng
Member: Hongli Ni

Our project aims to realize fast and high-quality Stimulated Raman Scattering(SRS)  volumetric imaging using a MEMS deformable mirror. Our setup is based on an SRS microscope. The SRS microscope enables sub-cellular detection of chemical distribution. The combination of SRS and volumetric imaging has potential in giving a global understanding of a complex 3D system, which is invaluable in biomedical analysis. Current volumetric imaging methods have disadvantages in imaging speed or aberrations. In our project, we adopt a MEMS deformable mirror offered by Professor Bifano to do volumetric imaging. Deformable mirror modulates the wavefront of incident light by changing its shape freely in a high speed(20 kHz). We are able to scan images in different depth by changing wavefront, then synthesis 3D image combining all 2D slices. Unlike ultrasound or tunable lens, the deformable mirror is a reflective component without chromatic aberration. Meanwhile, we can compensate for monochromatic aberrations by designing mirror shapes carefully.

I helped measure system characteristics including axial scanning range and imaging resolution. I wrote LabVIEW programs to control the deformable mirror and a piezo positioner determining the axial position of the objective. I moved the objective axially to acquire 3D images of focuses. The relative positions and shapes of focuses with different deformable mirror shapes are calculated by analyzing their 3D images. I made reasonable guesses for scanning range and resolution degradation. These guesses assisted in finding ways to improve image quality. We now resolved the issue of scanning range degradation and are in the way to enhance axial resolution. In the above experiments, I learned a lot of useful skills such as adjusting the SRS system to optimize signals, making samples and most importantly, shooting problems, reasoning causes and giving solutions. My mentor Peng also showed how to advance research by discussing and reading. We discussed our project with many people and read a plenty of paper about similar projects. I am now exploring the application of sparse sampling in our project with help of Haonan.  Sparse sampling is capable of boosting imaging speed with little decrease in image quality.Finally great thanks to Professor Cheng, Professor Bifano, my mentor Peng, Max, Haonan, Huate, and Nick. They all taught and helped me a lot.

Fiber Based on Optoacoustic Converter Simulation Based on Comsol

Mentor: Prof. Chen Yang
Member: Yiming Fu

Photoacoustic effect can be used to produce specific ultrasound frequency for neuronal stimulation which has a high sensitivity with ultrasound at 1MHz and lower. In order to make the fiber based optoacoustic converter (FOC) produce lower frequency ultrasound and intensity, it is necessary to change the thickness and material. In experiment, it would be time-consuming to find out how to design a FOC with lower frequency and higher intensity but it is very convenient in simulation. I will use COMSOL to tell experiment how to change the thickness and material in order to have a desired frequency and higher intensity.

I have built two kinds of model. The simple one is a fiber with an absorption layer which is like a rectangle. The complex one is similar to the experiment in which the absorption layer is like an ellipse. The simple one can have a very similar result from a paper named ‘NUMERICAL SIMULATION OF FIBER-OPTIC PHOTOACOUSTIC GENERATOR USING NANOCOMPOSITE MATERIAL’ in which the AuNPs is used as the light absorption material and PDMS is used as thermal expansion material. In simulation I use the PDMS as the thermal expansion material and use heat source to replace the light absorption process. The similar results validates my model and I can use my model to find out how the thickness and material property can influence the frequency and intensity of PA signal. The first conclusion of my research is that when the thickness increases, the frequency of the highest intensity will decrease and the maximum intensity will also decrease. The second conclusion is that when the Young’s modulus decreases, the proportion of the lower frequency will become larger which means there will be more signals distributing in low frequency. I still want to do research in other structure, but there is no paper which can validate my model and I am not convinced that the conclusion reached from that invalidated model can be used to instruct the experiment. So in the future, I might do more research in the structure perspective.

Analysis of The Generation and Properties of Nano Particle Photo Acoustic Signals

Mentor:  Prof. Chen Yang
Member: Yifei Geng

Ultrasound can stimulate neuronal cell excitation. Developing different methods for ultrasound stimulation of neuronal cells have recently become a research hot point.
We hope to find a way to accurately control the excitation of target neurons. By adsorbing some nanoparticles made by photoacoustic conversion materials on the surface of neuronal cells, ultrasonic signals can be generated when light is incident, achieving a localized stimulation effect.

During the summer research period, I mainly used the comsol software to model the nanoparticle photoacoustic signal generation process to explore the source of the PA signal , which can help us to understand the mechanism of photoacoustic nanoparticle. And I also explored the influence of different material parameters and particle size on the PA signal waveform, amplitude and frequency. Then we can select the frequency that we need and optimize material parameters to improve the conversion efficiency. The results (especially the temperature and acoustic pressure waveform over time) of the established model have been well matched with previous papers and experiments. The temperature distribution in the interior, surface and surrounding medium of the nanoparticle, the change of temperature and the change of PA signal with time at each reference point under the input laser irradiation are obtained, so the process of generating PA signal by the nanoparticle is basically clarified.

Spectroscopic Ellipsometer Modeling Plasmonic TiN

Mentor:  Prof. Chen Yang
Member: Yang Liu

One specific direction which is being explored in our lab is enhancing the efficiency of white LEDs. The main challenge here is developing TiN films, which still maintaining the metallic/plasmonic properties of the films. An Atomic Layer Deposition (ALD) technique is currently under development in our lab, and the fabrication conditions need to be optimized to develop films of high optical quality. In order to confirm that the titanium nitride is optically metallic and to know the location of the plasmon resonance, we need to measure the optical constants e1 and e2.

As we know, the metallic behavior of TiN films exhibits a sensitive dependence on the substrate and deposition details. And my work in Boston University is to measure the optical constants of TiN films with spectroscopic ellipsometer (SE). In order to achieve high accuracy for SE analysis, I build a combined Drude-Lorentz model describing the optical response of the conduction and valence electrons of TiN and TiNx. Due to high parameter uncertainty in Drude-Lorentz model, we can obtain several different models for one kind of TiN in the certain wavelength range which is decided by the experimental data we used to train this model. But the accurate model should has physical meaning in all wavelength range. So to build a model meaningful, I look into the physical meaning of the Drude and Lorentz expression and analyze the trend of permittivity of TiN. And after we analyze the optical constants we obtained from ellipsometer, we adjust our recipe to produce more metallic TiN and use this sample to verify our model. After all, I will calculate the real and imaginary parts of dielectric constant for TiN thin films.

3D Self-folding Silk Nanoladder for Spinal Cord Injury Recovery

Mentor:  Prof. Chen Yang
Member: Ran Cheng

When suffer from spinal cord injuries, it is hard for spinal cord neurons to reconstruct morphological and functional reconnections due to the poor regeneration properties of the central nervous system neurons. In order to prevent a glial scars formation, which further stop regeneration, a biocompatible scaffold is needed to restore the injured site and also mimic the structures of the extracellular matrix to guide the axon growth. So we decided to fabricate a biocompatible 3D self-folding bi-layered film patterned with silk nanoladder with controllable biodegrading rate, to achieve slow release of growth factor, and comparable biomechanical properties, to bridge the severed spinal cord gap, which is usually 2-3mm.

My main research here at first is to explore how to make the self-folding film self-roll into a multi-layered structure in the 37-degree water, so that we can achieve a higher performance on space ratio (53-75%, ratio between the active space, the space to allow the neurons to grow in, and the dead space) and guidance area (41-42mm2) by controlling the layer numbers, compared with porous materials have been reported (space ratio 29%, guidance area 38mm2) in the literature. By introducing different thick edges (short thick edge or long thick edge) or not and the number of the thick edges (0,1 or 2), also combined with the aspect ratio, I can control the folding direction along long-axis, short -axis and even diagonal direction to obtain the numbers of the folding layers from 1 layer to even 5 layers. However, when I observed the cross-section of the self-folding tube under the microscope, I found that there are some parts of the layers attached with each other, so to solve the problem, I introduced sacrificed layer to make the space between layers to be uniform. Gelatin was chosen to be the sacrificed material because it can be easily removed in the 37-degree water bath due to the low gel temperature. After trials, I successfully achieved space control by introducing gelatin as sacrificed layer. Finally, we used the self-folding tube to culture neurons and by 3D reconstruction, we are very delighted to find out that neurons can grow healthily on all layers. Based on the great results in vitro, it is very promising to move on to experiments in vivo to test the function on spinal cord injuries recovery. It is also very promising that we can obtain artificial spinal cord or even apply our system to culture some other kinds of cells, for example, bone cells in the future.

Specialized Physical surface for Synaptogenesis Induction in Vitro

Mentor: Prof. Chen Yang
Member: Shihan Zhu

Synapse is a structure that permits a neuron to pass an electrical or chemical signal to another neuron or to the target cell. So, to recover the function of neural system after injury, it is important to regenerate neurites and reconstruct synaptic connection together. In the former research, we have found that when we cultured neurons on the platform modified with nano-protrusions, we can find the filopodia attached to the protrusions and form the structures like part of synapse. Besides, when culturing neuron on the substrate with high density protrusions, more signals from post-synapse can be observed. Inspired by this “pseudo-synapse” structure, I want to tell the impact of the physical surface with nano-protrusions on the formation of in vitro synaptogenesis.

I have two hypotheses about the impact of protrusions to synaptogenesis. The first one is that the pattern of protrusions directs the growth of neurites and promotes different neurons to contact and form synapses. The other one is that the promotion of synaptogenesis is achieved through forming the synapse-like structure on protrusions. My research aims at comparing the distribution of beads and synapses and validating my hypotheses by observing the possible co-localization of the two elements. In the experiment, coverslips coated with 1 um polystyrene beads were chosen to mimic the substrates with nano-protrusions. Then I treated these coverslips with PDL solution and cultured the rat embryonic neurons on them. After culturing for 15 days in vitro, I labeled different synaptic components like synapsin, PSD-95 via immunofluorescence staining to see the signal from pre-synapses and post-synapses separately. The beads I used in this experiment are mixed with phycoerthrin which are a kind of protein with red fluorescence. By choosing the secondary antibody conjugated with dyes whose color is different from the beads, I can observe the signal from synapses and beads individually under distinct excitation wavelength of laser. Through comparing the overlapping of location of beads and synapses, I can validate my hypotheses. Based on the results of this experiment, we can conclude the effect of protrusions to the construction of neural network and further explore the mechanism of the connection between physical surface and neuron growth.

Deep Learning for Fourier Ptychography Microscope Reconstruction

Mentor: Prof. Lei Tian
Member: Yajie Wang, Shanshan Han, Zhiyuan Yu

The big goal for our project which is as the title involve the work with Fourier Ptchography Microscope(FPM) reconstruction and Deep Learning(DL) is to reconstruct a high resolution image form low resolution images and to predict the confidence of the result by giving a map of reliability of each pixel in the predicted high resolution image. As we all know, traditional model-based reconstruction method is time-consuming and will introduce artifacts. However, DL is a powerful tool to solve complex inverse math problems which are suitable for our project. And with GPU computing, the reconstruction procedure can be done within few seconds, much faster than traditional way. Moreover, the confidence map is quite important because we are able to measure our result accuracy in quantity. To achieve this, we use generated images by the widely accepted FPM method as ground truth and five bright field and dark field images as input.

The main object of this project is to reconstruction the low-resolution pictures of Hela cell to high-resolution with the deep learning technique. Since the deep learning technique needs ground truth, which are obtained by Fourier Ptychography Algorithm, we write codes to obtain the ground truth and collect the input data. During the project, we meet the problem of phase unwrapping. we try to fix it by simply masking the unwrapped place and add proper phase value. This method is fast and good, but not the best. Then we try more method and finally fix the problem by phase unwrapping algorithm based on least squares. So, the result of center picture patch looks good. Then we begin to reconstruct patches not in the center, which is more difficult due to the noise. Now most of the patches of the hole picture can be used in deep learning, which reduce the cost of collect input data and ground truth data. For deep learning, we combine Unet and DenseNet. we use laplace loss for the first 400 epochs and then add fourier loss for the next 200 epochs to enhance the contrast. This method works well on former data, however less effective on our newly obtained data, which have over-smoothing problem. To solve this problem we try dilated convolution layer, sub-pixel convolution, Xception layer, NPCC loss function new-designed dense block, the result get better. To increase the resolution of our prediction, we are now working on GAN.

Plasmonic Meta-Surface Object Detection

Mentor: Prof. Roberto Paiella
Member: Yunfeng Zhou

Plasmonic meta-surfaces have recently emerged in photonics as versatile light manipulation structures, which are particularly attractive due to their ability to modify light in distances much shorter than the wavelength. In recent work performed in the Paiella lab at BU, plane waves incident from a specific angle could be detected by such structures, with the target incident angle determined by the period of nanoparticles, resulting in angle-resolved light detection. As an extension to this technique, this work aims to explore design of a more complicated meta-surface structure capable of detecting a complicated wave front such as a specific pattern of light scattered off nano-scale target particles.

In this work, we first begin with the math functions that explain the relationship between phase of wave front and grating properties. Then, by analyzing phase patterns of different target particles, we select parameters that best produce an easily-detected wave front. We subsequently design a meta-surface to detect these particles. We also solve some practical problems that had an effect on our experiment result. In our simulation experiment, digitized numbers formed by diffraction pattern of gold particles of 10 microns in size are detected by our gratings, and it showed about 0.2% difference in transmission ratio between different digitized numbers. We believe that under-sampling is a critical problem that has hindered us from expanding this difference, which could be solved when fabrication technique is further advanced.