Use the setup_roi_segmentation.ipynb notebook to adjust these parameters
| Parameter | Definition | Recommended Initial Value | Notes |
|---|---|---|---|
| Data Parameters: | |||
| input_dir | Directory with images to be analyzed | All z-layers and channels for a specific sample must be combined into a single file (see Step 3 of the Protocol Procedure) | |
| output_dir | Directory to save ROI segmentation results | ROI masks will be added as an extra channel to the input image and saved in this directory | |
| channel | Channel index, starting from 0, that will be used to segment ROI | Cellpose allows using nuclei channel to improve whole-cell segmentation. To use this option, provide two channel indices as a list, where the first index corresponds to the nuclei staining, and the second index corresponds to the cytoplasm staining. Examples: 0 – the first channel will be used to segment ROI (either cells or nuclei) [1, 0] – the second channel (1) will be used as an auxiliary nuclei stain, the first channel (0) will be used to segment whole cells |
|
| Frequent Parameters: | |||
| diameter | Target ROI (cell or nucleus) diameter in pixels | An example image displayed in the notebook will contain scale in pixels to help determine the target ROI diameter. Set to None to automatically detect the ROI diameter | |
| Less Frequent Parameters: | |||
| model_type | Cellpose model to use for segmentation: ‘nuclei’ for nucleus segmentation, ‘cyto’ or ‘cyto2’ for cell segmentation | cyto2 | We found that ‘cyto’ and ‘cyto2’ models work better than ‘nuclei’ for segmenting nuclei with irregular shapes; for the full list of available models please refer to cellpose documentation |
| gpu | If True, cellpose segmentation will run on GPU; if False, cellpose will use CPU | True | GPU processing is significantly faster; use gpu=True whenever possible |
| clear_border | If True, will remove cells touching image border (in xy only) | False | |
| Advanced Parameters: | |||
| do_3D | If True, cellpose segmentation is performed in 3D; if False, cellpose segments ROI in each individual z-layer, and the ROI are combined in 3D in the postprocessing | False | 3D segmentation is resource intensive, though sometimes more accurate. If do_3D=True results in “CUDA out of memory” error, either set do_3D= False, or set gpu=False |
| flow_threshold | Cellpose parameter: the maximum allowed error of the flows for each mask | 0.4 | Advanced parameter. Increase if cellpose returns too few masks; decrease if cellpose returns too many ill-shaped masks |
| cellprob_threshold | Cellpose parameter: defines which pixels are used to run dynamics and determine masks | 0 | Advanced parameter. Decrease if cellpose returns too few ROI; increase if cellpose returns too many ROI; values should be between -6 and 6. |
| remove_small_mode | '2D', or '3D'. Used to remove small ROI by volume (3D) or area (2D) | 3D | Set to ‘3D’ unless testing on cropped images. Set to ‘2D’ if the image contains only a few z-layers. If set to ‘3D’, small ROI are excluded based on volume; this will exclude a ROI if only small part of it is contained in the field of view. |
| remove_small_diam_fraction | Size threshold used to exclude small ROI, provided as a fraction of the ‘diameter’ parameter | 0.5 | Advanced parameter. Increase to remove more ROI, decrease remove fewer ROI |
Use the setup_puncta_analysis.ipynb notebook to adjust these parameters
| Parameter | Definition | Recommended Initial Value | Notes |
|---|---|---|---|
| Data Parameters: | |||
| input_dir | Directory with images to be analyzed | All z-layers and channels for a specific sample must be combined into a single file (see Step 3 of the Protocol Procedure). If ROI segmentation was done, set this to the “output_dir” of the ROI segmentation. Alternatively, ignore this parameter and specify the “parameter_file” | |
| output_dir | Output directory to save puncta analysis results | ||
| roi_segmentation | If True, the last channel of the input images will be used as ROI mask | Set to False if the ROI segmentation step was skipped. Set to True if the image from “input_dir” contain ROI masks as the last channel. Alternatively, ignore this parameter and specify the “parameter_file” | |
| puncta_channels | List of channel indices, starting form 0, that will be used to segment puncta | Examples: [1] – puncta will be segmented in the second channel [2, 3] – puncta will be segmented in the third and fourth channels |
|
| Frequent Parameters: | |||
| minsize_um | Minimum target puncta size in µm | 0.2 | Will be used as the minimum sigma for the Laplacian of Gaussian detector. Decrease to detect smaller puncta, increase to avoid detection of smaller puncta |
| maxsize_um | Maximum target puncta size in µm | 2 | Will be used as the maximum sigma for the Laplacian of Gaussian detector. Increase to detect larger puncta, decrease to avoid detection of larger puncta |
| threshold_detection | Threshold used by LoG detector to exclude low intensity blobs | 0.001 | Should be close to 0 and can be both positive and negative. Start with threshold_detection=0 and first adjust minsize_um and maxsize_um to make sure that all puncta of relevant size are detected. After that, gradually increase the value of threshold_detection to remove low-intensity detection |
| segmentation_mode | Determines the way the “threshold_segmentation” is applied. For mode 0: absolute threshold is applied in LoG space; for mode 1: a threshold relative to the background is applied in LoG space; for mode 2: a threshold relative to the background is applied in image intensity space. | 0 | Advanced parameter. Set to 0 if the background fluorescent signal in all ROI is relatively uniform. Set to 1 if there is a large range of ROI background fluorescence values |
| threshold_segmentation | Threshold for puncta segmentation. Used in combination with the “segmentation_mode” | 0.001 | For mode 0, start with values between 0.001 and 0.003; for mode 1, start with values between 20 and 100; for mode 2, start with values between 2 and 3. Decrease or increase to detect more/bigger or fewer/smaller puncta |
| Less Frequent Parameters: | |||
| threshold_background | Threshold used to remove low intensity puncta centers, provided relative to the ROI background value (see “background_percentile”) | 3 | Example: threshold_ background=3 will remove all puncta centers with fluorescent intensity lower than 3 background values. Set to 0 to keep all puncta centers. Only applied if the ROI masks are provided |
| global_background | If False, the background value is calculated individually for each ROI. If True, the background value is calculated globally as the global_background_percentile of all ROI | False | Set to False if there is a large range of cell fluorescence values. This will increase sensitivity in cells with low fluorescence and decrease sensitivity in cells with high fluorescence |
| remove_out_of_roi | If True, puncta (parts) that extend beyond ROI will be removed. If False, all puncta will be kept | False | |
| Advanced Parameters: | |||
| num_sigma | Number of sigma values for the Laplacian of Gaussian detection | 5 | Advanced parameter. Decrease to save computational resources, increase to improve the accuracy of puncta centers detection |
| overlap | Parameter used by the LoG detector to remove the smaller one of two overlapping blobs | 1 | Advanced parameter. Set to 1 to only remove completely overlapping blobs. Decrease to remove blobs that are further apart. Should be between 0 and 1 |
| background_percentile | Intensity percentile (between 0 and 100) used to calculate the background value of the ROI | 50 | Advanced parameter. 50 corresponds to the median value. |
| global_background_percentile | Percentile (between 0 and 100) of ROI background values to calculate the global background value | 95 | Advanced parameter. Only used if global_background=True |
| maxrad_um | Maximum puncta radius in in µm. Used to remove large puncta | None | Set to None to keep all puncta |