Related Experiment Video
Updated: Sep 13, 2025
11:41
Rapid Analysis and Exploration of Fluorescence Microscopy Images
Published on: March 19, 2014
12.4K
HIST-DIP: histogram thresholding and deep image priors assisted smartphone-based fluorescence microscopy imaging
Harshitha Govindaraju1, Muhammad Nabeel Tahir1, Umer Hassan1
1Rutgers, The State University of New Jersey, Dept of Electrical and Computer Engineering, 94 Bret Rd, Piscataway, 08854 USA. umer.hassan@rutgers.edu.
The Analyst
|July 28, 2025
Summary
HIST-DIP enhances smartphone fluorescence microscopy images by removing noise and improving clarity. This unsupervised method boosts image quality for accessible, real-time point-of-care diagnostics without needing training data.
Area of Science:
- Biomedical Imaging
- Computational Microscopy
- Image Processing
Background:
- Smartphone-based fluorescence microscopy offers low-cost, portable point-of-care diagnostics.
- Image quality issues like noise and blur limit current applications.
- Restoration methods often require large labeled datasets, hindering real-time use.
Purpose of the Study:
- To develop an unsupervised image restoration framework for smartphone fluorescence microscopy.
- To improve image quality (contrast, resolution, signal-to-noise ratio) without external training data.
- To enable enhanced real-time, low-cost diagnostic imaging.
Main Methods:
- Introduced HIST-DIP (HIStogram Thresholding and Deep Image Prior) framework.
- Utilized histogram thresholding to isolate fluorescence signals and remove background noise.
- Employed Deep Image Prior (DIP) for structural refinement and resolution enhancement.
Main Results:
- Significant improvements in image quality metrics: PSNR increased from 15.59 to 27.10 dB.
- SSIM improved from 0.035 to 0.82, indicating better structural similarity.
- Enhanced contrast-to-noise ratio (CNR) and signal-difference-to-noise ratio (SDNR) for sharper details.
Conclusions:
- HIST-DIP effectively restores fluorescence microscopy images from smartphones.
- The unsupervised approach eliminates the need for training datasets, ideal for point-of-care applications.
- This method holds potential for real-time, on-device diagnostic imaging enhancement.
