Published: Benchmark Study of Deep Super-Resolution Models for Digital Holography

🔬 Exciting Publication Alert! 🔬 We are delighted to share that our FS-13 hyperspectral imaging system (FS-1X Series) was used in a recent study published in the IEEE Sensors Journal (2025) Researchers from Chongqing University of Posts and Telecommunications and Army Medical University developed CBIA-Net, a novel deep learning framework, to achieve rapid and non-destructive detection of wound bacteria. Using hyperspectral images captured by the FS-13, they built a large dataset covering seven typical wound pathogens at multiple concentrations and achieved 97% accuracy, precision, and recall in bacterial classification. This research demonstrates the powerful potential of CHNSpec’s FS-1X Series in biomedical imaging and clinical diagnostics, extending the applications of hyperspectral technology far beyond traditional fields. 👏 Congratulations to the research team for this innovative achievement, and thank you for trusting CHNSpec technology! #HyperspectralImaging #BiomedicalResearch #ClinicalDiagnostics #CHNSpec

Building the Future of Flow Cytometry: At SpikoCyte we’re rethinking what a cytometer can be. Our mission is to design a compact, intelligent system that combines cutting-edge technologies to unlock faster, more sensitive, and label-free cell analysis. Here’s what we’re integrating into our platform: - Microfluidics → to precisely handle and control cells in flow - Advanced Optics → to capture rich scattering signals - Neuromorphic Event-Based Cameras (DVS) → for real-time, high-speed vision without redundant data - Spiking Neural Networks → trained to classify cells efficiently, mimicking how the brain processes information By bringing these components together, we’re creating a new generation of cytometers that are smarter, faster, and more efficient—pushing the boundaries of biomedical research and diagnostics. #SpikoCyte #FlowCytometry #Microfluidics #Optics #Neuromorphic #EventBasedVision #SpikingNeuralNetworks #BiotechInnovation

✨ Exciting news! ✨ Our paper "Pixel Perfect MegaMed: A Megapixel-Scale Vision-Language Foundation Model for Generating High-Resolution Medical Images" has been accepted to the DGM4MICCAI Workshop at #MICCAI2025 🎉 The work has been selected for a short oral presentation, and I will have the honor of presenting it this September in South Korea 🇰🇷 . In this work, we use Stable Diffusion XL to create a vision-language foundation model tailored for medical imaging, scaling generation to megapixel resolution (1024×1024). This enables more detailed and clinically meaningful synthetic image generation. We further use a progressive upscaling module that extends generation up to 2048×2048 resolution, enhancing both global anatomical structure and local detail. We show that these high-resolution synthetic images not only preserve fine-grained anatomical details but also significantly improve classification performance in low-data regimes, supporting robust diagnostic modeling. Looking forward to enlightening discussions at #MICCAI2025. A huge thank you to my co-authors and collaborators Hujun Ni, Amar Kumar, and Professor Tal Arbel for their support throughout this project. This research was carried out in the Probabilistic Vision Group (PVG) at McGill University and Mila - Quebec Artificial Intelligence Institute. 🔗 Project Website: https://lnkd.in/eV_aZSmb 📄 ArXiv: https://lnkd.in/ehJ7VSuX #miccai2025 #medicalimaging #foundationmodels #stablediffusion #deeplearning #generativemodels

Taking image generation with foundation models in medical imaging to the next level ! One of the first papers to achieve ultra high-resolution medical image generation with Stable Diffusion XL. The work was brilliantly led by Zahra TehraniNasab, co-authored by Hujun Ni, and supervised by Prof. Tal Arbel. Beyond pushing the image resolution, we also show that these high-quality synthetic images significantly improve classification performance in low-data regimes, underscoring their practical value for medical AI. I’m proud of our team’s creativity and hard work in driving this forward. Excited to see it presented this September at DGM4MICCAI in South Korea 🇰🇷. #SyntheticImageGeneration #AIInHealthcare #FoundationModels #MedicalImaging #MICCAI2025

𝗙𝗿𝗼𝗺 𝗽𝗶𝘅𝗲𝗹𝘀 𝘁𝗼 𝗺𝗲𝗴𝗮𝗽𝗶𝘅𝗲𝗹𝘀: 𝗔𝗜 𝗳𝗼𝗿 𝘂𝗹𝘁𝗿𝗮 𝗵𝗶𝗴𝗵-𝗿𝗲𝘀𝗼𝗹𝘂𝘁𝗶𝗼𝗻 𝗺𝗲𝗱𝗶𝗰𝗮𝗹 𝗶𝗺𝗮𝗴𝗲𝘀 Interested in knowing how 𝘃𝗶𝘀𝗶𝗼𝗻-𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀 can generate 𝘂𝗹𝘁𝗿𝗮 𝗵𝗶𝗴𝗵-𝗿𝗲𝘀𝗼𝗹𝘂𝘁𝗶𝗼𝗻 𝗺𝗲𝗱𝗶𝗰𝗮𝗹 𝗶𝗺𝗮𝗴𝗲𝘀 (up to 𝟮𝟬𝟰𝟴×𝟮𝟬𝟰𝟴)? Our latest work, led by Zahra TehraniNasab and accepted for 𝗼𝗿𝗮𝗹 𝗽𝗿𝗲𝘀𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 at the 𝗗𝗚𝗠𝟰𝗠𝗜𝗖𝗖𝗔𝗜 𝗪𝗼𝗿𝗸𝘀𝗵𝗼𝗽 at #𝗠𝗜𝗖𝗖𝗔𝗜𝟮𝟬𝟮𝟱, demonstrates how 𝗦𝘁𝗮𝗯𝗹𝗲 𝗗𝗶𝗳𝗳𝘂𝘀𝗶𝗼𝗻 𝗫𝗟 can be adapted to generate synthetic medical images at 𝗺𝗲𝗴𝗮𝗽𝗶𝘅𝗲𝗹 𝗿𝗲𝘀𝗼𝗹𝘂𝘁𝗶𝗼𝗻𝘀. These images not only preserve fine anatomical detail but also improve classification in low-data regimes — a big step forward for #𝗠𝗲𝗱𝗶𝗰𝗮𝗹𝗔𝗜. 👏 Congratulations to Zahra, co-author Amar Kumar (Probabilistic Vision Group (PVG), McGill University & Mila - Quebec Artificial Intelligence Institute), and undergraduate researcher Hujun Ni! #MedicalAI #StableDiffusion #FoundationModels #MedicalImaging #AIHealthcare #DeepLearning #GenerativeAI #MICCAI2025 #PVG #MilaQuebec #McGillUniversity

Researchers Unlock High-precision Position Measurements Using Two-photon Interferometry and SPAD Arrays Researchers achieve highly precise position measurements by exploiting a refined form of two-photon interference that overcomes limitations caused by differences between individual photons, opening possibilities for advanced imaging and potentially new approaches to photonic processing #quantum #quantumcomputing #technology https://lnkd.in/e2m6uEJt

Over the past year, I’ve pursued two research projects that took a lot of effort to develop and establish. I’m thrilled to share that both were accepted as Lecture presentations at the IEEE International Ultrasonics Symposium (IUS 2025) being held in Utrecht, Netherlands! 🔹 GIAnI (Growing Indenter Anisotropy Imaging): A new ultrasound elastography technique that, to our knowledge, is the first capable of generating 2D images of tissue mechanical anisotropy—opening new doors for understanding tissue microstructure and pathology. 🔹 Deep-DoPIo: A deep learning framework that uses transformer architectures to predict tissue shear elasticity from ultrasound data—highlighting how AI can advance ultrasound elastography beyond traditional empirical models. While I couldn’t be there in person this year, I’m very grateful to Caterina Gallippi, Ph.D. for presenting both talks. Manuscripts from these projects are in progress, and I hope to be able to share them soon. I’d love to connect with anyone interested in discussing these ideas or future collaborations! #ultrasound #IUS #IUS2025 #elastography #deeplearning #medicalimaging

Excited to share that our paper “Chalcogenide-Based Brain-Inspired Photo-Synapses for Neuromorphic Vision Sensor: An Experimental and Theoretical Study” has been accepted for publication in Small. In this work, we explore 2D chalcogenide memristors, focusing on SnS₂ as a multifunctional vdW material to replicate biological visual systems. Our combined experimental and simulation work reveals that SnS₂ memristors enable wavelength-sensitive pattern recognition and memory, achieving 98.51% accuracy on MNIST datasets. https://lnkd.in/g9UmwutW #NeuromorphicComputing #2DMaterials #MachineVision #ArtificialIntelligence #BrainInspired

🚀 Glad to share the latest work, fresh from the press, is now published in APL Photonics: "Computational terahertz phase imaging using a sequence of multi-plane intensity measurements" For years, terahertz imaging has promised incredible possibilities, from non-destructive material testing to biomedical diagnostics and security screening. However, one big hurdle has been recovering the spatial phase information of imaging objects, which carries crucial structural information, without relying on fragile interferometric experimental approaches. In this work, we explored a different path: ✨ recording only intensity images at multiple planes, ✨ and then using a powerful and innovative computational algorithm to retrieve the phase. Specifically, our approach introduces two non-interferometric imaging schemes, a lens-less configuration and a 4f imaging system, paired with a computational second-order optimization framework. By leveraging a smart optimisation strategy, we achieved faster convergence and more accurate phase reconstructions than conventional methods. 🌱 Beyond the algorithm, we validated the technique experimentally using insect wings, engineered test objects, and even biological samples such as leaves, demonstrating sensitivity to water distribution and internal structures. I'm deeply grateful to my contributing authors, Pitambar Mukherjee, Frédéric Fauquet, Kedar Khare, Sylvain Gigan, and Patrick Mounaix, and to our labs at Laboratoire Kastler Brossel - LKB, IMS Laboratory, for the collaborative spirit that made this possible. We would also like to thank the funding agencies ANR, the French National Research Agency, and CNRS for their support. To read more, the paper is open access here: https://lnkd.in/dAuHNF7j

This is a recent paper from our ongoing collaboration between the Radiology Department at UNMC and Electrical & Computer Engineering at UNL! In this project, we apply AI and deep learning tools to advance in vivo molecular imaging, aiming for greater robustness and efficiency. Our first study demonstrates that CEST MRI can be accelerated by up to 10× while maintaining high-quality data in a normal mouse model. Next, we are evaluating the robustness of the method to handle disease mouse models. This is a nice project. Grateful to be part of this team! https://lnkd.in/dmD-CQNb https://lnkd.in/d7D4qNpR