Related Experiment Video
Updated: Dec 18, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.2K
Dense-UNet: a novel multiphoton in vivo cellular image segmentation model based on a convolutional neural network
Sijing Cai1,2, Yunxian Tian3,4, Harvey Lui3,4
1Key Laboratory of Optoelectronic Science and Technology for Medicine of Ministry of Education, Fujian Provincial Key Laboratory of Photonics Technology, Fujian Normal University, Fuzhou, China.
Quantitative Imaging in Medicine and Surgery
|June 19, 2020
Summary
Deep learning models like Dense-UNet improve multiphoton microscopy (MPM) image segmentation for in vivo skin cells. This advanced CNN technology enhances accuracy and detail capture for clinical applications.
Area of Science:
- Biomedical Imaging
- Computational Biology
- Medical Technology
Background:
- Multiphoton microscopy (MPM) shows promise for clinical biopsies.
- Lack of efficient image processing, particularly automatic segmentation, hinders clinical adoption of MPM.
- Segmentation remains a significant challenge in MPM imaging.
Purpose of the Study:
- To explore the feasibility of using deep learning, specifically convolutional neural networks (CNNs), for segmenting in vivo MPM skin cell images.
- To propose and evaluate a novel deep learning model, Dense-UNet, for improved MPM image segmentation.
Main Methods:
- A novel deep learning segmentation model, Dense-UNet, was developed based on the U-Net architecture.
- Dense-UNet incorporates dense concatenation to increase network depth and enable feature reuse.
- The model was implemented in Python using TensorFlow for segmenting 128x128 resolution MPM images of in vivo skin cells.
Main Results:
- Dense-UNet achieved a segmentation accuracy of 92.54%, outperforming the standard U-Net (88.59%).
- Dense-UNet demonstrated a significantly lower loss value (0.1681) and a superior Dice coefficient (90.60%) compared to U-Net.
- The F1-Scores for Dense-UNet, U-Net, and Seg-Net were 93.35%, 90.02%, and 85.04%, respectively, highlighting Dense-UNet's effectiveness.
Conclusions:
- The Dense-UNet model's architecture effectively captures fine cellular boundary details and provides accurate localization.
- This study presents the first application of a deep CNN (Dense-UNet) for segmenting MPM in vivo images, addressing a critical gap.
- Dense-UNet offers state-of-the-art performance for segmenting low-resolution, in vivo MPM images, providing a high-precision, automated solution.
