Research
Introducing NICA lab
This video briefly introduces our lab's research direction (4 min, Language: Korean).
Video Seminar
A few research topics in our lab are introduced in this seminar (12 min, Language: English).
Acquiring Big Data from Brain
Functional Imaging
With genetic engineering, neurons can be modified to change their brightness as a function of the their activity (i.e., neurons "blink" as they fire) which literally makes the brain activity visible. The main challenge is to record the optical signals at a high spatiotemporal resolution and we develop optical imaging techniques to tackle this.
Three-dimensional fluorescence microscopy through virtual refocusing using a recursive light propagation network, Medical Image Analysis, 2022.
3DM: Deep decomposition and deconvolution microscopy for rapid neural activity imaging, Optics Express, 2021.
RLP-Net: A Recursive Light Propagation Network for 3-D Virtual Refocusing, International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), 2021.
Sparse decomposition light-field microscopy for high speed imaging of neuronal activity, Optica, 2020.
Precision Calcium Imaging of Dense Neural Populations via a Cell-Body-Targeted Calcium Indicator, Neuron, 2020.
Robotic multidimensional directed evolution of proteins: development and application to fluorescent voltage reporters, Nature Chemical Biology, 2018.
Simultaneous whole-animal 3D imaging of neuronal activity using light-field microscopy, Nature Methods, 2014.
Computational Imaging
The performance of imaging systems is impacted by a range of factors, including physics, biology, information theory, and the sampling theorem. To mitigate these limitations, we're utilizing computational imaging methods that leverage machine learning to predict more information from limited data.
IMPASTO: Multiplexed cyclic imaging without signal removal via self-supervised neural unmixing, bioRxiv, 2022.
PICASSO: Ultra-multiplexed fluorescence imaging of biomolecules through single-round imaging and blind source unmixing, Nature Communications, 2022.
Three-dimensional fluorescence microscopy through virtual refocusing using a recursive light propagation network, Medical Image Analysis, 2022.
3DM: Deep decomposition and deconvolution microscopy for rapid neural activity imaging, Optics Express, 2021.
RLP-Net: A Recursive Light Propagation Network for 3-D Virtual Refocusing, International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), 2021.
Sparse decomposition light-field microscopy for high speed imaging of neuronal activity, Optica, 2020.
Simultaneous whole-animal 3D imaging of neuronal activity using light-field microscopy, Nature Methods, 2014.
Multiplexed & Super-resolution Imaging
Fluorescence microscopy is limited to imaging only four proteins simultaneously due to the broad emission spectra of fluorescent molecules. To surpass this limitation and visualize a larger number of proteins, we are developing multiplexed imaging technologies that use machine learning algorithms for blind signal separation.
IMPASTO: Multiplexed cyclic imaging without signal removal via self-supervised neural unmixing, bioRxiv, 2022.
Nanoscale resolution imaging of the whole mouse embryos and larval zebrafish using expansion microscopy, bioRxiv, 2022.
PICASSO: Ultra-multiplexed fluorescence imaging of biomolecules through single-round imaging and blind source unmixing, Nature Communications, 2022.
Iterative expansion microscopy, Nature Methods, 2017.
Analyzing Big Data from Brain
Neuro-image Processing
State-of-the-art functional imaging methods generate more than a gigabyte of data per second, necessitating the development of automated analysis algorithms. We develop fast and scalable machine learning algorithms capable of processing such brain images without the need for labeled data.
Design Principles of Multi-Scale J-invariant Networks for Self-Supervised Image Denoising, IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2025.
From Pixels to Information: Artificial Intelligence in Fluorescence Microscopy, Advanced Photonics Research, 2023.
Statistically unbiased prediction enables accurate denoising of voltage imaging data, Nature Methods, 2023.
Robust and Efficient Alignment of Calcium Imaging Data through Simultaneous Low Rank and Sparse Decomposition, IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2023.
Nanoscale resolution imaging of the whole mouse embryos and larval zebrafish using expansion microscopy, bioRxiv, 2022.
Efficient Neural Network Approximation of Robust PCA for Automated Analysis of Calcium Imaging Data, International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), 2021.
Feasibility of 3D Reconstruction of Neural Circuits using Expansion Microscopy and Barcode-Guided Agglomeration, Front. Comput. Neurosci., Oct 2017.
Neuro-data Mining
Neural activity underlies many functions in our brain, but our understanding of the fundamental principles of neural signal processing remains limited. To gain greater insight, we apply computational methods to analyze brain activity data and quantify information flow, uncovering the functional connections between neurons. Our aim is to identify repeating patterns, discover local circuits that operate together, and extract synaptic strength information from brain activity, leading to a deeper understanding of the brain.
RASP: Robust Mining of Frequent Temporal Sequential Patterns under Temporal Variations, EDBT, 2025.
Artificial Intelligence & Biomedical Engineering
Bio-inspired AI
A key objective in the field of artificial intelligence is to create a machine intelligence that truly mirrors biological intelligence. To pursue this goal, we aim to incorporate characteristics of biological intelligence in terms of both architecture and learning processes into artificial intelligence.
Inducing Functions through Reinforcement Learning without Task Specification, Neural Information Processing Systems Workshop (NeurIPS Workshop) on Deep Reinforcement Learning, 2022.
Medical AI
Medical diagnosis often relies on patient medical images that can be formulated as image classification problems which AI excels in. However, the variation in the imaging condition and equipment poses a a significant challenge for bringing AI into real-world diagnosis. Furthermore, limited labeled data is often available in most cases. To overcome these limitations, we are combining AI techniques and an understanding of the data's characteristics.