1. 32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society
Denoising of LSFCM images with compensation for the Photoblinking/Photobleaching effects Isabel Rodrigues1,2, J. Miguel Sanches1,3
1Institute for Systems and Robotics
2Instituto Superior de Engenharia de Lisboa
3Instituto Superior Técnico, Lisbon, Portugal
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

2. The Problem
Fluorescence confocal microscopy images are affected by fading effects, called photoblinking/ photobleaching (PBPB), that lead to image intensity decreasing along the time, preventing long time experiments.
Images present low signal to noise ratio (SNR).
Images are corrupted by non-additive noise mainly due to the strong amplification performed by the photomultiplier needed to detect the small fluorescent radiation emitted by the fluorophore.
The SNR increases with the amount of incident radiation as well as the PBPB effect and the Toxicity
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

3. Real data sequences
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

4. Goal
Design a reconstruction algorithm for this type of 2D + t image sequences in a Bayesian framework where
the non- additive noise is modeled by a Poisson distribution and
the spatio-temporal correlation among pixels is taken into account and
a theoretical model to describe the intensity decay due to the PBPB effect is also included
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

5. PBPB MODEL Fluorescent tagging molecules assumed in three main states:
ON-state - able to fluoresce and be observed (nON),
OFF-state - not able to fluoresce (not visible ) (nOFF),
BLEACHED-state - permanently OFF.
mean intensity of the image at the instant t
The time intervals at each state are governed by Levy Statistics - statistical aging - leads to a constant increasing on the time intervals at each state along the time.
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

6. PBPB MODEL
-- parameters associated with the concentrations and state transition dynamics of the molecules;
I – amount of incident radiation.
and are constants computed using the initial conditions.
Given the physical constraints of the problem it can be proved that are always positive real constants.
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

7. Reconstruction Optimization criterion: MAP – maximum a posteriori
Data fidelity term
Prior term
: pixel morphology intensity
: pixel noiseless intensity,
: pixel noisy intensity
: PBPB model estimated parameters
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

8. Prior
The morphology is assumed to be a MRF  P(F) is a Gibbs distribution
: log-Euclidean potential functions  edge preserving
neighbors of f in a 2nd order clique
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

9. Regularization parameters
Optimization
Non convex Energy function to minimize
By changing the variable , , a convex function is obtained.
The function is computed directly from the data.
Regularization parameters
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

10. EXPERIMENTAL RESULTS Synthetic Data
Original F :
Intensity layers: 10, 30, 60.
Two-exponentials decay :
Regularization parameters :
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

11. EXPERIMENTAL RESULTS Synthetic Data
Reconstructed sequence and estimated morphology. Images 2, 21, 63
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

12. EXPERIMENTAL RESULTS Synthetic Data
Monte Carlo test: 500 runs
Estimated morphology relative error per iteration
SNR morphology
MSE morphology
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

13. EXPERIMENTAL RESULTS Synthetic Data
Model Validation
(147.9s)
(2.2s)
(1319.2s)
(17.2s)
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

14. EXPERIMENTAL RESULTS Real Data
Noisy data, estimated morphology and respective profile plots
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society

15. THANKS
32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society