fidoptBioSR wide-field to linear SIM, microtubules, 2D
One cell of BioSR's microtubules (Qiao et al., Nature Methods 2021): a 2D wide-field image at the lowest signal level, the mean of nine raw SIM images, with the linear SIM image as target.
Read before comparing (8)
- Structure. BioSR images four structures, and each is its own dataset (BioSR_CCP_Syn, BioSR_ER_Syn, BioSR_MT_Syn, BioSR_Actin_Syn), since the pretrained BioSR models such as DFCAN are trained one a structure. The four replace BioSR_Syn, the mixed set, which held other draws.
- Case. The source is the mean of the cell's nine raw SIM images at level 1, the noisiest, in camera counts with an offset of about 100. The target is the provider's SIM reconstruction from the high-signal raw images, 1004 x 1004 pixels of 31.3 nm, scaled per image to a peak of 65535 that nothing in the source predicts; divided by that peak, it lies between 0 and 1.
- Alignment. Linear interpolation puts source pixel k on SIM pixel 2k, as for BioTISR_Syn. The shift left is at most 0.32 SIM pixel on the twenty cases prepared first and 0.25 on the five test cells added later; prepare.py refuses a case over one pixel.
- Mask. The SIM image smoothed by a Gaussian of 0.3 µm, above Otsu's threshold, grown by 1 µm: 49 to 78 % of the field on the twenty cases, 48 to 72 % on the five. It is made from the target and given to every method.
- Baseline. Inside the mask at a range of 1, a constant at the target's median scores an MAE of 0.059 (0.047 to 0.074), a PSNR of 20.8 dB (19.2 to 22.0) and an SSIM of 0.49 (0.40 to 0.56) on the 15 test cases; 0.063, 20.7 dB and 0.46 on development. SSIM is ranked, as other ranked synthesis sets give a constant 0.47 to 0.62.
- Line to beat. On the ten test cases drawn first, the source mapped linearly onto the target by least squares inside the mask scored an MAE of 0.034 (0.027 to 0.042) and 26.5 dB. No method can do that fit without the target, but a method must beat it to do better than rescaling its input.
- Overlap. Each cell lies in one zip of the deposit, so no cell is in two BioSR datasets. No other dataset of the corpus holds BioSR images; BioTISR_Syn, from the same laboratory, is a separate deposit of 3D-SIM time-lapses of live COS-7 cells.
- Outside numbers. The test cells are the provider's own (its code trains on cells 6 to 35 and validates on 1 to 5), but all of BioSR is public with its targets: a model trained on all of it has seen the test cases.
| Accuracy | Cost | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| # | Method | MAE | Change | PSNR µm | SSIM | Time s | RAM GB | GPU GB | Images |
| 1 | DFCAN · microtubulesCPUappnote | 0.0 ±0.0 | −113.2 | 27.30 ±1.47 | 0.776 ±0.034 | 12 | 1.9 | ||
| Source as given | 113.2 ±1.7 | -41.09 ±0.13 | 0.000 ±0.000 | ||||||
The order averages each method's ranks on MAE, PSNR µm, SSIM, so the first one is not always the best in every column. ± is the 95% interval of the mean over the 15 cases. Two neighbours are tied when a paired Wilcoxon test on the shared cases does not find the upper one better on MAE (5%): the places a method could hold are then shown under its own. The best value of each column is in colour. Click a row to see its images, a column title to sort by it. Time and memory measured on Laptop.
DFCAN · microtubules. At signal level 1 its background is brighter than the target's, so its MAE ties a constant image's on some cells (mt_036: 0.063 against 0.062) while its structure is right (correlation 0.90, SSIM 0.69 against the constant's 0.43). The authors' model as they run it.
MAE: mean absolute error of a synthetic image against its target, inside the body, in the image's units. PSNR: the same by signal-to-noise ratio, higher is better. SSIM: the same by structural similarity, 1 is perfect.
Case by case
Click a score to look at that result: the target image, and over it the method's synthetic image. Underlined scores have their image; for the others you see the case without the result.
About the data
- Test cases
- The provider's 15 test cells for microtubules, whole (Cell_036 to Cell_050, the default test_cell_list of DataAugumentation_ForTest.m in the DL-SR repository), and 10 development cells drawn among its training and validation cells (Cell_001 to Cell_035). One cell is a case, at signal level 1.
- Truth
- target image as released the provider's linear SIM image of the cell (SIM_gt.mrc), reconstructed from the cell's high-signal raw SIM images, divided by its largest value
- Source
- figshare.com/articles/dataset/BioSR/13264793
doi.org/10.6084/m9.figshare.13264793.v9
doi.org/10.1038/s41592-020-01048-5
github.com/qc17-THU/DL-SR - Licence
- CC BY 4.0 (the figshare record's licence field)
10 cases for tuning, never tested
mt_025, mt_022, mt_003, mt_010, mt_017, mt_014, mt_035, mt_024, mt_026, mt_019