This is the folder contains all the results used in the pre-trained CNN for FLIM lifetime denoising and phasor denoising.
#For fluoresence microscopy image denoising using ImageJ plugin: https://github.com/ND-HowardGroup/Instant-Image-Denoising
We use these (DnCNN and Noise2Noise) CNN models as pre-trained CNNs for denoising FLIM lifetime and phasor denoising.
#For K-means segmentation: https://github.com/ND-HowardGroup/Kmeans-FLIM-Phasors
Noisy Input phasor: 3D zebrafish sample
Denoised Output Phasor: From Neural network:
Noisy segments: 2D Mouse Kidney shows the microtubules
Denoised segments: 2D Mouse Kidney shows the microtubules show the upstream and downstream microtubules.
Details: plant sample1 Input: 2D single channel image from plant images dataset: (from our custom made InstantFLIM FD-MPM-FLIM system at excitaiton: 800nm and sample: bean leaf plant in vivo imaging) Denoised: Image denoising using our ImageJ plugin (using pre-trained DnCNN CNN model): (time: 80 ms in GPU, image size: 512x512) Target: Target image generated by taking average of 5 noisy images in the same FOV:
Input Noisy HSV lifetime | Median HSV filtering | DnCNN HSV denoising | Target HSV lifetime |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
Input Noisy lifetime | Median filtering | DnCNN denoising Plugin | Target Image |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
Input Noisy lifetime histogram | Median filtering | DnCNN denoising | Target HSV lifetime |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
Input Noisy HSV lifetime | Median HSV filtering | DnCNN HSV denoising | Target HSV lifetime |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
Details: plant sample2 Input: 2D single channel image from plant images dataset: upper epidermis layer: (from our custom made InstantFLIM FD-MPM-FLIM system at excitaiton: 800nm and sample: bean leaf plant in vivo imaging) Denoised: Image denoising using our ImageJ plugin (using pre-trained DnCNN CNN model): (time: 80 ms in GPU, image size: 512x512) Target: Target image generated by taking average of 5 noisy images in the same FOV:
Input Noisy HSV lifetime | Median HSV filtering | DnCNN HSV denoising | Target HSV lifetime |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
Input Noisy lifetime | Median filtering | DnCNN denoising Plugin | Target Image |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
Input Noisy HSV lifetime | Median HSV filtering | DnCNN HSV denoising | Target HSV lifetime |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
Details: BPAE sample1 Input: 2D single channel image from BPAE images: (from our custom made InstantFLIM FD-MPM-FLIM system at excitaiton: 800nm and sample: BPAE ex vivo imaging) Denoised: Image denoising using our ImageJ plugin (using pre-trained DnCNN CNN model): (time: 80 ms in GPU, image size: 512x512) Target: Target image generated by taking average of 5 noisy images in the same FOV:
Input intensity | Noisy HSV lifetime | DnCNN HSV denoising | Target HSV lifetime |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
Noisy lifetime Phasor | DnCNN lifetime Phasor | Target lifetime Phasor |
---|---|---|
![]() |
![]() |
![]() |
Details: Mouse Kidney sample1 Input: 2D single channel image from fixed Mouse Kidney images: (from our custom made InstantFLIM FD-MPM-FLIM system at excitaiton: 800nm and sample: fixed mouse kideny ex vivo imaging) Denoised: Image denoising using our ImageJ plugin (using pre-trained DnCNN CNN model): (time: 80 ms in GPU, image size: 512x512) Target: Target image generated by taking average of 5 noisy images in the same FOV:
Input intensity | Noisy HSV lifetime | DnCNN HSV denoising | Target HSV lifetime |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
Noisy lifetime Phasor | DnCNN lifetime Phasor | Target lifetime Phasor |
---|---|---|
![]() |
![]() |
![]() |
Details: BPAE sample3: large FOV Input: 2D single channel image from BPAE images: (from our custom made InstantFLIM FD-MPM-FLIM system at excitaiton: 800nm and sample: BPAE ex vivo imaging) Denoised: Image denoising using our ImageJ plugin (using pre-trained DnCNN CNN model): (time: ~90 ms in GPU, image size: 560x560) Target: Target image generated by taking average of 5 noisy images in the same FOV:
Input intensity | Noisy HSV lifetime | DnCNN HSV denoising | Target HSV lifetime |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
Details: Mouse Kidney sample4: large FOV Input: 2D single channel image from fixed Mouse Kidney images: (from our custom made InstantFLIM FD-MPM-FLIM system at excitaiton: 800nm and sample: fixed mouse kideny ex vivo imaging) Denoised: Image denoising using our ImageJ plugin (using pre-trained DnCNN CNN model): (time: ~90 ms in GPU, image size: 560x560) Target: Target image generated by taking average of 5 noisy images in the same FOV:
Input intensity | Noisy HSV lifetime | DnCNN HSV denoising | Target HSV lifetime |
---|---|---|---|
![]() |
![]() |
![]() |
![]() |
Details: Input: 2D single channel FLIM image of FLIM BPAE sample captured using our Instant FLIM system: FLIM Image denoising using our ImageJ plugin (from pre-trained Noise2Noise image denoising model): (time: <80 ms in GPU, image size: 512x512) Target: Target image generated by taking average of 5 noisy images in the same FOV: The improvement in PSNR in the denoised lifetime image is 8.26 dB compared to the noisy input lifetime image.
Figure Caption: From left to right: noisy composite lifetime, denoised composite lifetime, target composite lifetime. Composite lifetime is the hue saturation value (HSV) representation of intensity and lifetime images together, where intensity and the fluorescence lifetimes are mapped to the pixels’ brightness and hue, respectively. The top row indicates a field of view (FOV) of 512
Details:
Input: 2D single channel FLIM image of FLIM BPAE sample captured using our Instant FLIM system:
FLIM Image denoising using our ImageJ plugin (from pre-trained DnCNN ML model): (time: <80 ms in GPU, image size: 512x512)
Target: Target image generated by taking average of 5 noisy images in the same FOV:
Figure Caption: From left to right: Fluorescence intensity image (a), noisy composite lifetime (b), BM3D image denoised composite lifetime (c), Pre-trained DnCNN denoised composite lifetime (d), target composite lifetime (e). Composite lifetime is the hue saturation value (HSV) representation of intensity and lifetime images together, where intensity and the fluorescence lifetimes are mapped to the pixels’ brightness and hue, respectively. The top row indicates a field of view (FOV) of 512
#Images: The testing dataset of in vivo Zebrafish and in vio mouse kidney images are given in this folder and for the BAPE sample images can be downloaded from here https://curate.nd.edu/show/mw22v40954f Additional images are added from the instant FLIM and commertial FLIM systems in the datasets folder. Fixed BPAE cells and fixed mouse kidney cells are imaged using Instant FLIM system the data is provided in the Datasets folder.
#Citation for dataset: Please cite the Fluorescence Microscopy Lifetime Denoising (FMLD) dataset using the following format: Mannam, Varun. 2020. “Fluorescence Microscopy Lifetime Denoising (FMLD) Dataset.” Notre Dame. https://doi.org/10.7274/r0-18da-9m58.
© 2019 Varun Mannam, University of Notre Dame
Licensed under the GPL