This PDF file contains the front matter associated with SPIE Proceedings Volume 11952 including the Title Page, Copyright information, and Table of Contents.

Optical coherence tomography (OCT) and light sheet fluorescence microscopy (LSFM) are well-established imaging techniques preferred in developmental biology, e.g., embryonic imaging. However, each technique has its own drawbacks, such as resolution and molecular specificity with OCT and field-of-view (FOV) and speed with LSFM. To overcome these limitations for small animal embryo imaging, we have developed a co-aligned multimodal imaging system combining OCT and LSFM. The OCT probe and LSFM excitation beams were combined and scanned with a galvanometer-mounted mirror through the same objective lens. The light sheet thickness was ~13 μm. The LSFM collection arm consisted of a 0.8 numerical aperture water immersion objective, tube lens, and CCD camera, resulting in a transverse resolution of ~2.1 μm. The OCT system was based on a 100 kHz swept-source laser with a central wavelength of 1050 nm and had a lateral resolution of ~15 µm and an axial resolution of ~7 μm. Images of fluorescent microbeads and a fluorescent-tagged mouse embryo at gestational day 9.5 showed the capabilities of the multimodal imaging system. Since the OCT system and LSFM system were co-aligned, image registration was straightforward and enabled high-throughput multimodal imaging without the need for complex registration techniques.

In recent years, significant research has been performed on developing powerful and efficient Convolutional Neural Network (CNN) architectures. To utilize these architectures in pixel-level regression tasks such as tomographic image reconstruction, a feature extraction encoder is often combined with a symmetrical decoder to generate an encoder-decoder structure, such as a U-Net. However, a more powerful decoder focusing on high-frequency features can provide higher pixel-level accuracy. In this work, we investigate the use of asymmetrical encoder-decoder architectures in medical image reconstruction tasks. The state-of-the-art EfficientNet architecture utilizes depthwise convolutions and channel attention within inverted residual bottleneck blocks to generate highly compressed features while maintaining a significant FLOPS efficiency advantage compared to regular convolutional encoders. We develop an asymmetric encoder-decoder architecture, which uses the EfficientNet as an encoder. The proposed decoder architecture combines the multi-resolution features generated by the EfficientNet encoder using an incremental feature expansion strategy, which leads to better preservation of the structural details in reconstructed images. We have tested our asymmetrical encoder-decoder approach on undersampled MRI reconstruction tasks using the Calgary Campinas multi-channel brain MR dataset. Results demonstrate that the proposed asymmetric approach vastly outperforms a symmetric Efficient U-Net, achieving a 3dB improvement in PSNR. SSIM was also improved, and the asymmetric network was found to recover small structural details more effectively. Furthermore, the proposed asymmetric Efficient U-Net provides a four-fold reduction in inference time when compared to the conventional U-Net architecture.

Breast conserving surgery re-excision rates are high due to insufficient intraoperative guidance available to the surgeon. Post-operative studies have demonstrated that micro-CT and optical structured light have potential to determine margin status, but typical systems for these modalities are not specifically designed for use in surgical guidance. This study reports on the design and analysis of such a multimodal imaging system and is the first of its kind built specifically for use during breast conserving surgery. The custom system is based on a commercial IVIS SpectrumCT micro-CT scanner (PerkinElmer, Hopkinton, MA) without standard optical and bioluminescence imaging hardware and instead houses multiline lasers (Streamline Lasers, Osela Inc., Lachine, QC, Canada) mounted to a translation stage (DDSM100, Thorlabs, Newton, NJ), and a fast readout CMOS camera (Blackfly S USB 3.0, FLIR, Wilsonville, OR) for reflectance-based optical active line scanning. The system requires minimal training to operate and performs imaging and visualization of scans rapidly. Image processing uses the Insight Toolkit (ITK 5.1.2, Kitware, Clifton Park, NY), CT reconstruction uses the Reconstruction Toolkit (RTK 2.2.0, RTK Consortium, Villeurbanne, France) with CUDA (NVIDIA, Santa Clara, CA) for speed enhancement, and data visualization leverages the Visualization Toolkit (VTK 8.2.0, Kitware, Clifton Park, NY). Resolution and image quality metrics are comparable to current pre-clinical research systems, while scans are performed faster and with streamlined software for interacting with image data in near real-time. This study represents the advancement of multimodal micro-CT and optical structured light imaging toward clinical translation for margin detection during breast conserving surgery.

Nerve graft repair surgery has been accepted for treating peripheral nerve injuries in which the transected nerve ends are incapable of primary end-to-end tensionless neurorrhaphy. In this paper, we present ex vivo proof-of-concept of a functional intra-operative guidance using voltage-sensitive dye (VSD) imaging. In particular, we here evaluate the efficacy of photoacoustic (PA) and fluorescence (FL) imaging modalities, in each of which near-infrared VSD contrast can be obtained by transmembrane VSD redistribution mechanism upon neuronal depolarization events. Realistic nerve graft surgery was mimicked by ex vivo sciatic nerve freshly excised from an anesthetized pig model. The dual-modal PA/FL imaging system was configured to monitor ex vivo sample chamber with wide field-of-view covering a significant portion of the nerve sample. The ex vivo sample chamber was equipped with an arbitrary electrical stimulation and recording system to trigger and monitor the neuronal electrophysiology, respectively. The proof-of-concept study suggested the high VSD signal sensitivity in functional PA VSD imaging with its unique depth profiling capability over the thick nerve tissue (<2-mm diameter). Otherwise, FL imaging indicated unspecific signal trends that might suffer from depth-unspecific imaging mode, which makes it challenging to extract nerve-related signals from background VSD signal clutter unbound to nerve tissue.

Diffuse optical tomography (DOT) is a promising non-invasive optical imaging technique that can provide functional information of biological tissues. Since diffuse light undergoes multiple scattering in biological tissues and boundary measurements are limited, DOT reconstruction is ill-posedness and ill-conditioned. To overcome these limitations, Tikhonov regularization is the most popular algorithm. Recently, deep learning based reconstruction methods have attracted increasing attention, and promising results have been reported. However, they lack generalization for unstructured physical model. Therefore, a model-base convolution neural network framework (Model-CNN) is developed. It composes of two layers: data consistency layer and depth layer, which increases the interpretability of the model. Its performance is evaluated with numerical simulations. Our results demonstrate that Model-CNN can get better reconstructed results than those obtained by Tikhonov Regularization in terms of ABE, MSE, and PSNR.

The design of an all-reflective laser-scanning microscope capable of imaging samples three-dimensionally with multiple modes of nonlinear imaging at high resolution and with a large field of view is described in this presentation. The all-reflective design is based on off-axis parabolic mirror sections, and was designed to require minimal alignment. An all fiber alignment-free femtosecond laser with dual output is used to enable simultaneous multiphoton and stimulated Raman imaging processes, creating a powerful but hassle-free system. This paper will describe the design, performance, and imaging outputs of the system.

We proposed and implemented a deep learning scheme using convolution neural networks (CNNs) with batch normalization (BNCNN) to construct a sensor-image DOI computation model with the aim of reconstructing tissue optical-property images as well as identifying and localizing breast tumors. A non-iterative learning reconstruction method was developed to recover optical properties, focusing on one-dimensional convolution layers followed by dense layers. Besides simulated data for model training, validation and testing, for the comparison of model performance, measurement data sets were employed to test on the same trained network which results outperform Tikhonov regularization method and other artificial neural networks as well.

Scientists today have powerful tools to alter a wide range of genes in animal models with ease to understand biological processes in development and disease conditions. However, there is a lack of imaging technology to assess the effects of these genetic modifications in live mammals. Without these in vivo imaging capabilities, scientists cannot fully understand how the individual organism’s response to overall process. In this work, we have developed a multi-modal optical coherence microscopy (OCM) and dual-channel fluorescence microscopy (DC-FM) to enable the evaluation of biological processes in transgenic mice models. The combined system achieved a simultaneous recording of reflectance and fluorescence signals of deep tissue layers at the speed of 250 kHz and a lateral resolution of ~ 2 μm over a field of view of 1.3 x 1.3 mm² . OCM and DC-FM also achieved an axial resolution of 2.8 μm and 23 μm, respectively. To evaluate the performances of the system in imaging the cornea of transgenic mice, a conditional dual-reporter mouse strain, KerartTA/tet-O-Cre/RosamTmG triple transgenic mouse strain, which express membrane-bound tomato red (mT), was harnessed. Upon exposure to doxycycline (Dox) in KerartTA/tet-O-Cre/RosamTmG mice, it would then express a membrane-bound green fluorescent protein (mG) in the corneal stromal cells. While OCM evaluated the change in thickness and structures of the cornea, DC-FM provided green and red proteins expression in the cornea. This system will enable longitudinal in vivo studies on transgenic mice models to advance the understanding of developmental and disease mechanisms.

We implemented a ring-scanning mechanism in a prone diffuse optical imaging (DOI) system for the application of breast tumor detection before. In the study, simultaneous multiple-sinusoids driving light source and flexible opto-measurement channels were considered to update the DOI system. The designed flexible channels in the scanning module of imaging system prevent optical information loss measured on a noncircular phantom/object if using fixed optical channels. Further, simultaneous multiple-sinusoids driving light sources speed up the acquisition of optical information for the frequency domain DOI. Examination phantoms were designated to justify the proposed measurement schemes.

Diffuse Optical Tomography (DOT) is a promising non-invasive and relatively low-cost biomedical image technology. The aim of DOT is to reconstruct optical properties of the tissue from boundary measurements. However, the DOT reconstruction is a severely ill-posed problem. To reduce the ill-posedness of DOT and to improve image quality, imageguided DOT has attracted more attention. In this paper, a reconstruction algorithm for DOT is proposed based on the convolutional neural network (CNN). It uses both optical measurements and magnetic resonance imaging (MRI) images as the input of the CNN, and directly reconstructs the distribution of absorption coefficient. The merits of the proposed algorithm are without segmenting MRI images and modeling light propagation. The performance of the proposed algorithm is evaluated using numerical simulation experiments. Our results reveal that the proposed method can achieve superior performance compared with conventional reconstruction algorithms and other deep learning methods. Our result shows that the average SSIM of reconstructed images is above 0.88, and the average PSNR is more than 35 dB.

In the study the multimodal optical coherence tomography (MM OCT) including microstructural cross-polarization OCT (CP OCT) imaging with the application of attenuation coefficients combined with compression OCT-elastography (OCE) with quantitative morphological segmentation based on specific stiffness ranges for delineation of breast cancer margins was applied. The research was carried out on different morphological and molecular subtypes of human breast cancer. The findings of this study suggest that OCE and CP OCT of breast cancer images may, in the future, enable real-time feedback to the surgeon about accurate resection margin location in patients with breast cancer.

Line-field confocal optical coherence tomography (LC-OCT) is an optical technology developed for in vivo skin imaging. Combining the principles of OCT and reflectance confocal microscopy, LC-OCT generates high-resolution three-dimensional (3D) images with an isotropic spatial resolution of 1.3 micron and up to 500 microns in depth. Confocal Raman microspectroscopy is a label-free optical technique which allows for point measurement of the molecular content of a sample with micrometer resolution. While LC-OCT provides morphological information, Raman spectroscopy brings chemical information but lacks image guidance for targeting specific points of interest in the sample. Combining the two modalities would therefore provide complementary information and guidance for Raman measurements. We present a method to co-localize LC-OCT and Raman acquisitions for ex vivo applications. This co-localization approach allows acquisition of Raman spectra at specific locations targeted in a 3D LC-OCT image, with an accuracy of ± 20 μm. The co-localization method was developed using a LC-OCT device designed for ex vivo imaging and a custom Raman system. The principle of co-localization relies on both the use of a specific sample holder that can be positioned under each device with high repeatability and of a coordinate-based calibration between the two devices. Co-localization was validated using pig skin samples containing tattoo ink of known composition. LC-OCT images allowed to target specific regions in the samples where the presence of tattoo ink was revealed by detection of a specific Raman signature.

A novel wearable MRg-NIRST system for breast cancer detection has been designed and developed. In this prototype, eight (8) flex circuit strips, each with six (6) photo-detectors (PDs) and six (6) source fibers, are attached to the breast to collect diffused light. A 6x48 fiber switch and 48 side-firing fibers deliver intensity modulated laser light at six (6) near-infrared wavelengths. Light intensity at each of 2304 source-detector positionsis obtained for T₂-MRI guided 3D NIRST image reconstruction. In phantom testing, reconstructed images showed the contrast between tumor/inclusion and normal/background.