Flexible and user friendly Conditional Microscopy
CyberSco.py is a software written in Python for performing Conditional Microscopy. It was conceived to be available to none expert users with a simple intuitive GUI interface and easy syntax for predefined experiments. It uses Machine Learning segmentation for extracting information for the conditional steps.
Go to https://www.anaconda.com/products/individual and install Anaconda
Creation of a virtual environment, click here to expand
in a terminal write:
conda create -n env_name python=3.7.7 anacondaFor activating :
conda activate env_nameFor deactivating :
conda deactivatefull information for installation procedure can be found here : GPU support for Windows
installation in short, Click here to expand
Here is a summary of the different steps for the GPU installation:
1. cuda_10.1.243_426.00_win10.exe
can be found at https://www.filehorse.com/download-nvidia-cuda-toolkit/42676/
2. cudnn-10.1-windows10-x64-v7.6.5.32.zip
can be found at https://developer.nvidia.com/rdp/cudnn-archive
In the unfolded list of Download cuDNN v7.6.5 (November 5th, 2019), for CUDA 10.1
3. Copy 3 files from cudnn to Cuda
- bin: Copy <cuDNN directory>\cuda\bin\*.dll to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vxx.x\bin
- include : Copy <cuDNN directory>\cuda\include\*.h to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vxx.x\include
- lib : Copy <cuDNN directory>\cuda\lib\x64\*.lib to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vxx.x\lib\x64
4. check that you have in the environment variables those 2 paths :
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vxx.x\bin
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vxx.x\libnvvp
Click here to expand
Go to :
Xcite dlls
Extract the zip file, turn on the XCite device, then in device manager,
add new device using the extracted folder with all the dlls
Go inside CyberSco.py/drivers/pyAndorSDK3 and run :
python3 –m pip install .Open an Anaconda terminal in CyberSco.py folder, activate the virtual environment and run:
python setup.py install Go in the folder CyberSco.py and click on
run.bat
You can also create a shortcut on the Desktop for an even faster access.
It will launch two consoles, one for the camera and another for the CyberSco.py interface. Do not click inside of these cmd windows as it can perturb the script execution. The loading page takes around half a minute to finish due to machine learning models loading. Once the models loaded a Begin button appears.
Machine learning models are stored in the folder CyberSco.py/models. They can be used for segmentation or for event detection. For the moment, CyberSco.py only accepts models produced with Tensorflow.
Configuration files for the models, Click here to expand
The user needs to provide also three other configuration files in the folder CyberSco.py/modules/settings:
- The file models.yaml which contains the correspondences between the models and their shortcut.
- The file curr_models.yaml which contains the shortcut for the principal segmentation model.
- finally the file event_model.yaml which contains the shortcut for the model devoted to the event detection.
The different experiments presented in this paper rely on the utilization of two segmentation models with the UNET architecture. The trainings were performed using Nvidia GeForce GTX 1080 GPU card with Tensorflow 2.2.0 and with data augmentation spanning the transformations of rotation (6 angles), flip operation, noise, colors, and contrast.
The two segmentation models, Click here to expand
The first model is used for segmenting the yeast cells at 20x and the second one for segmenting the yeast cells at 60x.
The first model (20x) was trained on 15 epochs (in 15 minutes) from a training set of 20 pictures. The masks were obtained from RFP pictures in conjunction with OTSU algorithm.
| BF 20x | BF with segmentation 20x |
|---|
| 
The second model (60x) was trained on 5 epochs (in 5 minutes) from a training set of 20 pictures. This training set was obtained from fluorescence signal RFP images which were used for creating a model which trained multiple times permitted to enlarge the masks until having a good cell segmentation.
| BF 60x | BF with segmentation 60x |
|---|---|
 |
 |
In the current CyberSco.py version, the tracking for a given cell is simply done by selecting
the cell the nearest from one image to the following.
When changing of position, the tracking is performed using correlation between images.
Tracking image example, Click here to expand
Here are screenshots to explain how to use the Graphic User Interface(GUI):
Interface overview with functionalities, Click here to expand
The interface is divided in tree parts. A window for selecting the view and making snapshots, a middle panel for building the MDA experiment and a green vertical bar for accessing to various utilities like the Settings Channels, a panel for controlling the quality of the segmentation etc ...
Change the current view or select the view for the MDAs. Positions in X,Y,Z can be entered directly or obtained by clicking on the arrows after having setting displacement step.
Take snapshot with different illuminations with superimposing on the BF image.
Create various illumination settings used in the MDAs.
Check the quality of the segmentation in real time.
Retrieve the hand control of the microscope and make composite image of the chip in a larger field of view.
Check which devices are correctly connected and read the information about the GPU.
Making a MDA, Click here to expand
It is easy to build a MDAs experiment using the tree graphical tool. The tree is available once the user has selected, predefined MDA or free MDA In the case of the predefined MDA all the elements of the tree are yet there. The user can only duplicate, move them but cannot add new ones.
The tree encapsulates graphically the MDA through the hierarchy of the serial
instructions (sent to the microscope and the associated devices) and the loops.
To add a new elementary action just drag and drop it from the three select menus
(blue selectors) inside the tree. You can modify the hierarchy position of the
actions using ctrl+arrow. To modify the parameters of each action in
the tree, click on the corresponding wrench icon.
The tree tool for conceiving a MDA (for free MDA option) is three folds: A part for creating, changing or deleting MDAs, a part for dragging and dropping actions in the tree (three blue selectors) and finally the tree itself containing the loops and actions of the MDAs.
When clicking on a blue selector, various actions are available. The firsts selector contains... the second one.. and the last one ..
It is possible to set the parameters for each step of the MDA by clicking on the wrench button. This opens a windows in which the parameters can be modified.
Predefined MDA use the tree tool but without the possibility to add new elements in the tree. The user can only duplicate and change the elements yet in place in the tree. You can find all the conditional experiments used in the paper as python file plugins in the folder modules>predef>plugins
plugin structure, Click here to expand
In CyberSco.py, a plugin program consists of a python class with the name of the experiment which inherits from the MDA class.
This class contains the class initialisation __init__() of which comment is used to produce the name of the experiment and its description that will appear as a tooltip window when hovering on the experiment name in the interface.
The syntax must be name : name_of_the_experiment for the name and description : place_here_the_description for the description.
The plugin is then divised in three blocks.
- The main MDA block determined by the
define()method which contains the MDA instructions written serially and finishing by theself.launch_loop()function. - The MDA initialisation block determined by the two methods :
init_on_positions()andinit_conditions() - And lastly the MDA conditional block determined by the method
check_conditions()which will be executed after each iteration.
Plugin code example for Sucrose lag experiment, Click here to expand
from datetime import datetime
from modules.mda import MDA
class PROTOCOL_SUCROSE(MDA):
'''
Sucrose protocol for observing lag time according to the number of cells..
'''
def __init__(self, ldevices=None):
'''
name : Sucrose lag experiment
description : the cells (which are able to use sucrose cooperatively) of different chambers are fed with glucose and counted in real time. Their supply is then stopped and replaced with sucrose when the number of cells reaches a given limit. For each chamber this limit is different. The counting continues and it is possible at the end of the experiment to observe from the cells counting, the lag phenomenon which is characterized by a time during which the cells do not divide or very slowly until being able to exploit cooperatively the sucrose.
'''
MDA.__init__(self, ldevices)
def define(self, debug=[0]):
'''
'''
self.refocus() # add refocusing
self.take_pic() # add take BF pic
##
self.analyse_pic() # analyse the pic
self.cond = 'sucrose' # apply the conditions 2, sucrose
##
self.launch_loop() # Loop
def init_on_positions(self):
'''
Initialize the protocol
'''
for pos in self.list_pos:
pos.event = self.event
pos.first_time = self.first_time
if pos.event.name: # same event on all the positions
pos.event.exists = True
def init_conditions(self):
'''
Setup parameters and initial conditions
'''
self.init_on_positions()
##
lthresh = [ 100, 500, 2000, 500 ] # list of the thresholds
for i,pos in enumerate( self.list_pos ):
pos.thresh_cells = lthresh[i] # initial delay
self.delay = self.delay_init = 2
self.repeat = 2400 # number of repetitions
def check_conditions(self, rep):
'''
At a given threshold trigger the sucrose
'''
for pos in self.list_pos:
if pos.nb_cells > pos.thresh_cells and not pos.switched :
# pos.num_gate exists
if pos.num_gate:
pos.switched = True # position blocked
# adding num gate to self.gates_blocked
self.gates_switched += [ pos.num_gate ]
print( f'self.gates_switched '
'{ self.gates_switched } for nb cell of { pos.nb_cells } ' )
# change gates value
self.ga.set_pos_indices( self.gates_switched, 1 )
Distributed under the MIT License. See LICENSE.txt for more information.
Pascal Hersen pascal.hersen@curie.fr