Related Experiment Video
Updated: Sep 23, 2025
Dual Raster-Scanning Photoacoustic Small-Animal Imager for Vascular Visualization
Published on: July 15, 2020
Deep learning acceleration of multiscale superresolution localization photoacoustic imaging
Jongbeom Kim1, Gyuwon Kim1, Lei Li2
1Departments of Electrical Engineering, Mechanical Engineering, Convergence IT Engineering, and Interdisciplinary Bioscience and Bioengineering, Graduate School of Artificial Intelligence, Medical Device Innovation Center, Pohang University of Science and Technology (POSTECH), 77 Cheongam-ro, Nam-gu, Pohang, Gyeongbuk, 37673, Republic of Korea.
Deep neural networks (DNNs) enhance superresolution imaging by reconstructing high-density images from fewer frames. This computational strategy improves both temporal and spatial resolutions for photoacoustic microscopy and tomography.
Area of Science:
- Biomedical Imaging
- Optical Physics
- Machine Learning
Background:
- Superresolution imaging enhances spatial resolution by superimposing multiple raw frames, but this sacrifices temporal resolution.
- Existing methods require numerous frames, limiting applications where speed is critical.
Purpose of the Study:
- To develop a computational strategy using deep neural networks (DNNs) to reconstruct high-density superresolution images from significantly fewer raw frames.
- To improve both spatial and temporal resolutions in photoacoustic imaging modalities.
Main Methods:
- A deep neural network (DNN)-based computational strategy was employed to reconstruct superresolution images.
- The method was applied to both 3D label-free localization optical-resolution photoacoustic microscopy (OR-PAM) and 2D labeled localization photoacoustic computed tomography (PACT).
Main Results:
- The number of raw volumetric frames for OR-PAM was reduced from tens to fewer than ten.
- The number of raw 2D frames for PACT was reduced by 12-fold.
- Simultaneous improvement in temporal and spatial resolutions was achieved for both OR-PAM and PACT.
Conclusions:
- The DNN-based approach significantly reduces the number of required raw frames for superresolution photoacoustic imaging.
- This method offers a practical tool for preclinical and clinical studies demanding fast temporal and fine spatial resolutions.
- Deep-learning powered localization photoacoustic imaging presents a significant advancement in biomedical imaging capabilities.
