Skip to content
Snippets Groups Projects
GUI_main.py 76.3 KiB
Newer Older
lpbsscientist's avatar
lpbsscientist committed
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
This script is the main script used to produce a GUI which should help for
cell segmentation. This script can only read .nd2 files containing the 
lpbsscientist's avatar
lpbsscientist committed
images of cells, especially it displays for each recorded positions (field
of view) the pictures in the time axis.

The script opens first a window which allows you to load an nd2 file and
to load or create an hdf file. The hdf file contains all the masks, so if
it is the first the user segments an nd2 file, a new one should be created.
And it can be then loaded for later use.  Along with new hdf file, created
by the name entered by the user (say filename), it creates three other hdf 
files (filename_predicted.h5, filename_thresholded.h5 and 
filename_segmented.h5) these contain all the steps of the NN to get to the 
segmented picture. 
After the first window is finished a second one opens, where at each time 
index, three pictures 
lpbsscientist's avatar
lpbsscientist committed
are displayed the t-1 picture, the t picture (current frame which can be
edited) and the t+1 picture. Using the arrows one can navigate through time.
On top of the picture, there is always a mask which is displayed, if no cells
are present in the mask then the mask is blank and the user does not see it. 
lpbsscientist's avatar
lpbsscientist committed
If one wants to hand anmotate the pictures, one can just start to draw on the
picture using the different functions (New Cell, Add Region, Brush, Eraser,
mattminder's avatar
mattminder committed
Save Mask, ...) and the informations will be saved in the mask overlayed on 
top of the pictures. 
lpbsscientist's avatar
lpbsscientist committed

If one wants to segment using a neural network, one can press the
corresponding button (Launch CNN) and select the time range and 
lpbsscientist's avatar
lpbsscientist committed
the field of views on which the neural network is applied.

Once the neural network has finished predicting, there are still no visible
masks, but on the field of views and time indices where the NN has been
applied, the threshold and segment buttons are enabled. By checking these
two buttons one can either display the thresholded image of the prediction or
display the segmentation of the thresholded prediction.

At this stage, one can correct the segmentation of the prediction using
the functions (New Cell, Add Region, etc..) by selecting the Segment
checkbox and then save them using the Save Seg button.
If the user is happy with the segmentation, the Cell Correspondance button 
lpbsscientist's avatar
lpbsscientist committed
can be clicked. Until then, the cells get random numbers attributed by
the segmentation algorithm. In order to keep track of the cell through time,
the same cells should have the same number between two different time pictures.
This can be (with some errors) achieved by the Cell Correspondance button,
which tries to attribute the same number to corresponding cells in time.
After that, the final mask is saved and it is always visible when you go on
the corresponding picture. This mask can also be corrected using the 
usual buttons (because the Cell Correspondance makes also mistakes). 
lpbsscientist's avatar
lpbsscientist committed

"""
import sys
import numpy as np

# For writing excel files
from openpyxl import load_workbook
from openpyxl import Workbook
lpbsscientist's avatar
lpbsscientist committed

# Import everything for the Graphical User Interface from the PyQt5 library.
mattminder's avatar
mattminder committed
from PyQt5.QtWidgets import (QApplication, QMainWindow, QDialog, 
    QMessageBox, QPushButton, QCheckBox, QAction, QStatusBar, QLabel)
lpbsscientist's avatar
lpbsscientist committed
from PyQt5 import QtGui

#Import from matplotlib to use it to display the pictures and masks.
mattminder's avatar
mattminder committed
from matplotlib.backends.qt_compat import QtWidgets
mattminder's avatar
mattminder committed
from matplotlib.backends.backend_qt5agg import NavigationToolbar2QT as NavigationToolbar

#append all the paths where the modules are stored. Such that this script
#looks into all of these folders when importing modules.
sys.path.append("./unet")
sys.path.append("./disk")
sys.path.append("./icons")
sys.path.append("./init")
sys.path.append("./misc")
lpbsscientist's avatar
lpbsscientist committed

#Import all the other python files
#this file handles the interaction with the disk, so loading/saving images
#and masks and it also runs the neural network.
import InteractionDisk_temp as nd
lpbsscientist's avatar
lpbsscientist committed
#this file contains a dialog window that takes two integers as entry to swap
#two cell values
import ExchangeCellValues as ecv
lpbsscientist's avatar
lpbsscientist committed
#this file contains a dialog window which is opened before the main program
#and allows to load the nd2 and hdf files by browsing through the computer.
import DialogFileBrowser as dfb
#this file contains a window that opens to change the value of one cell. It 
lpbsscientist's avatar
lpbsscientist committed
#is opened as soon as the user presses with the left click on a specific cell.
import ChangeOneCellValue as cocv
lpbsscientist's avatar
lpbsscientist committed
#this file contains a dialog window to browse for the excel file where
#all the extracted information on the fluoerscence is written. Or to create a 
#new excel file by typing a name in the text box. It is thought to have one 
lpbsscientist's avatar
lpbsscientist committed
#excel file per field of view.
import DialogDataBrowser as ddb
lpbsscientist's avatar
lpbsscientist committed
#this file contains a dialog window where a time range and the field of views
#can be selected to then launch a prediction of the neural network on
#a specific range of pictures.
import LaunchBatchPrediction as lbp

#this file initializes all the buttons present in the gui, sets the shortcuts
#to these buttons and also connect the buttons to the function that are 
lpbsscientist's avatar
lpbsscientist committed
#triggered when the buttons are pressed.
import InitButtons

#this file contains the layout of the main window so it justs puts the buttons
#and the pictures at the desired position in the main window.
import InitLayout
mattminder's avatar
mattminder committed

# PlotCanvas for fast plotting
from PlotCanvas import PlotCanvas
lpbsscientist's avatar
lpbsscientist committed

class NavigationToolbar(NavigationToolbar):
    """This is the standard matplotlib toolbar but only the buttons
    that are of interest for this gui are loaded. These buttons allow 
    to zoom into the pictures/masks and to navigate in the zoomed picture. 
lpbsscientist's avatar
lpbsscientist committed
    A Home button can be used to set the view back to the original view.
    """
    toolitems = [t for t in NavigationToolbar.toolitems if
                 t[0] in ('Home', 'Pan', 'Zoom','Back', 'Forward')]
lpbsscientist's avatar
lpbsscientist committed
class App(QMainWindow):
    """This class creates the main window.
    """

    def __init__(self, nd2pathstr, hdfpathstr, newhdfstr):
        super().__init__()
lpbsscientist's avatar
lpbsscientist committed
        self.title = 'YeaZ 1.0'
        # all these ids are integers which are used to set a connection between
        # the button and the function that this button calls.
        # There are three of them because it happens that one can trigger three 
        # different functions with one button.        
lpbsscientist's avatar
lpbsscientist committed
        self.id = 0
        self.id2 = 0
        self.id3 = 0 
lpbsscientist's avatar
lpbsscientist committed

        self.reader = nd.Reader(hdfpathstr, newhdfstr, nd2pathstr)
        # these variables are used to create/read/load the excel file used
        # to write the fluorescence values extracted. For each field of view,
        # the user will be asked each time to create a new xls file for the 
        # field of view or to load an existing field of view (this is the role
        # of the boolean variable)
lpbsscientist's avatar
lpbsscientist committed
        self.xlsfilename = ''
        self.nd2path = nd2pathstr
        self.FlagFluoExtraction = False
        # Set the indices for the time axis and the field of view index. These
        # indices represent everywhere the current picture (the one that can be
        # edited, i.e. the time t frame)
lpbsscientist's avatar
lpbsscientist committed
        self.Tindex = 0
        self.FOVindex = 0
        # loading the first images of the cells from the nd2 file
lpbsscientist's avatar
lpbsscientist committed
        self.currentframe = self.reader.LoadOneImage(self.Tindex,self.FOVindex)
lpbsscientist's avatar
lpbsscientist committed
        
        # check if the t+1 time frame exists, avoid failure if there is only
        # one picture in the folder/nd2 file
lpbsscientist's avatar
lpbsscientist committed
        if self.Tindex+1 < self.reader.sizet:
            self.nextframe = self.reader.LoadOneImage(self.Tindex+1, self.FOVindex)
        else:
            self.nextframe = np.zeros([self.reader.sizey, self.reader.sizex])
        
lpbsscientist's avatar
lpbsscientist committed
        self.previousframe = np.zeros([self.reader.sizey, self.reader.sizex])

        # loading the first masks from the hdf5 file
lpbsscientist's avatar
lpbsscientist committed
        self.mask_curr = self.reader.LoadMask(self.Tindex, self.FOVindex)
        self.mask_previous = np.zeros([self.reader.sizey, self.reader.sizex])
lpbsscientist's avatar
lpbsscientist committed
        
        # check if the t+1 mask exists, avoid failure if there is only
        # one mask in the hdf file
lpbsscientist's avatar
lpbsscientist committed
        if self.Tindex+1 < self.reader.sizet:
            self.mask_next = self.reader.LoadMask(self.Tindex+1, self.FOVindex)
        else:
            self.mask_next = np.zeros([self.reader.sizey, self.reader.sizex])
        
        # creates a list of all the buttons, which will then be used in order
        # to disable all the other buttons at once when one button/function
        # is pressed/used in the gui.
lpbsscientist's avatar
lpbsscientist committed
        self.buttonlist = []
        # setting buttons as attributes
        # the shortcuts for the buttons, the functions to which they are
        # connected to,... are all set up in the ButtonInit file which is called
        # in the self.initUI() method below.
lpbsscientist's avatar
lpbsscientist committed
        self.button_newcell = QPushButton("New cell")
        self.buttonlist.append(self.button_newcell)
lpbsscientist's avatar
lpbsscientist committed
        self.button_add_region = QPushButton("Add region")
        self.buttonlist.append(self.button_add_region)
lpbsscientist's avatar
lpbsscientist committed
        self.button_savemask = QPushButton("Save Mask")
        self.buttonlist.append(self.button_savemask)
lpbsscientist's avatar
lpbsscientist committed
        self.button_drawmouse = QPushButton('Brush')
        self.buttonlist.append(self.button_drawmouse)
lpbsscientist's avatar
lpbsscientist committed
        self.button_eraser = QPushButton('Eraser')
        self.buttonlist.append(self.button_eraser)
lpbsscientist's avatar
lpbsscientist committed
        self.button_exval = QPushButton('Exchange Cell Values')
        self.buttonlist.append(self.button_exval)
lpbsscientist's avatar
lpbsscientist committed
        self.button_showval = QCheckBox('Show Cell Values')
        self.buttonlist.append(self.button_showval)
lpbsscientist's avatar
lpbsscientist committed
        self.button_hidemask = QCheckBox('Hide Mask')
        self.buttonlist.append(self.button_hidemask)
lpbsscientist's avatar
lpbsscientist committed
        self.button_nextframe = QPushButton("Next Time Frame")
        self.buttonlist.append(self.button_nextframe)
lpbsscientist's avatar
lpbsscientist committed
        self.button_previousframe = QPushButton("Previous Time Frame")
        self.buttonlist.append(self.button_previousframe)
lpbsscientist's avatar
lpbsscientist committed
        self.button_cnn = QPushButton('Launch CNN')
        self.buttonlist.append(self.button_cnn)
lpbsscientist's avatar
lpbsscientist committed
        self.button_threshold = QCheckBox('Threshold prediction')
        self.buttonlist.append(self.button_threshold)
lpbsscientist's avatar
lpbsscientist committed
        self.button_segment = QCheckBox('Segment')
        self.buttonlist.append(self.button_segment)
lpbsscientist's avatar
lpbsscientist committed
        self.button_cellcorespondance = QPushButton('Tracking')
lpbsscientist's avatar
lpbsscientist committed
        self.buttonlist.append(self.button_cellcorespondance)
lpbsscientist's avatar
lpbsscientist committed
        self.button_changecellvalue = QPushButton('Change cell value')
        self.buttonlist.append(self.button_changecellvalue)        
        
lpbsscientist's avatar
lpbsscientist committed
        self.button_extractfluorescence = QPushButton('Extract Fluorescence')
        self.buttonlist.append(self.button_extractfluorescence)
lpbsscientist's avatar
lpbsscientist committed
        self.button_hide_show = QPushButton('CNN')
        self.buttonlist.append(self.button_hide_show)
        
lpbsscientist's avatar
lpbsscientist committed
        self.initUI()


    def initUI(self):
        """Initializing the widgets contained in the window. 
        Especially, it creates the widget to plot the 
        pictures/masks by creating an object of the PlotCanvas class self.m. 
lpbsscientist's avatar
lpbsscientist committed
        Every interaction with the masks or the pictures (loading new
        frames/editing the frames/masks) occurs through this class.
lpbsscientist's avatar
lpbsscientist committed
        This method initializes all the buttons with the InitButtons file.
        It connects the buttons to the functions that they should trigger,
        it sets the shortcuts to the buttons, a tool tip, 
        eventually a message on the status bar when the user hovers 
lpbsscientist's avatar
lpbsscientist committed
        over the button, etc..
        
        This function also sets all the layout in the InitLayout file. It 
lpbsscientist's avatar
lpbsscientist committed
        takes and places the widgets (buttons, canvas, toolbar).

        The function initializes a Menu Bar to have a menu which can be 
lpbsscientist's avatar
lpbsscientist committed
        improved later on.
        It sets a toolbar of the matplotlib library and hides it. But it allows
        to connect to the functions of this toolbar through "homemade" 
lpbsscientist's avatar
lpbsscientist committed
        QPushButtons instead of the ones provided by matplotlib.
        Finally, it sets a StatusBar which displays some text to describe
        the use of some buttons, or to show that the program is working on 
lpbsscientist's avatar
lpbsscientist committed
        something (running the neural network, loading frames, etc...)
lpbsscientist's avatar
lpbsscientist committed
        After all this has been initialized, the program is ready to be used.
        """
        self._main = QtWidgets.QWidget()
        self.setCentralWidget(self._main)

        # Here our canvas is created where using matplotlib, 
        # one can plot data to display the pictures and masks.
lpbsscientist's avatar
lpbsscientist committed
        self.m = PlotCanvas(self)
        # Initialize all the buttons that are needed and the functions that are 
        # connected when the buttons are triggered.
lpbsscientist's avatar
lpbsscientist committed
        InitButtons.Init(self)
        InitLayout.Init(self)
        # MENU, TOOLBAR AND STATUS BAR
        # creates a menu just in case, some other functions can be added later
        # in this menu.
lpbsscientist's avatar
lpbsscientist committed
        menubar = self.menuBar()
        self.fileMenu = menubar.addMenu('File')   
lpbsscientist's avatar
lpbsscientist committed
        self.saveactionmenu = QAction('Save')
        self.fileMenu.addAction(self.saveactionmenu)
        self.saveactionmenu.triggered.connect(self.ButtonSaveMask)
        # hide the toolbar and instead of the original buttons of matplotlib,
        # QPushbuttons are used and are connected to the functions of the toolbar
        # it is than easier to interact with these buttons (for example to 
        # to disable them and so on..)
lpbsscientist's avatar
lpbsscientist committed
        self.Nvgtlbar = NavigationToolbar(self.m, self)
        self.addToolBar(self.Nvgtlbar)
        self.Nvgtlbar.hide()
        # creates a status bar with user instructions
lpbsscientist's avatar
lpbsscientist committed
        self.statusBar = QStatusBar()
        self.setStatusBar(self.statusBar)
        self.statusBarText = QLabel()
        self.statusBar.addWidget(self.statusBarText)
lpbsscientist's avatar
lpbsscientist committed
        self.show()
lpbsscientist's avatar
lpbsscientist committed
    def mousePressEvent(self, QMouseEvent):
        """this function is implemented just to have the QLineButtons of the 
lpbsscientist's avatar
lpbsscientist committed
        change time index button, setthreshold button and the setsegmentation
        button out of focus when the user clicks somewhere
        on the gui. (to unfocus the buttons)
        """
        self.button_timeindex.clearFocus()
        if self.button_SetThreshold.isEnabled():
            self.button_SetThreshold.clearFocus()
lpbsscientist's avatar
lpbsscientist committed
        if self.button_SetSegmentation.isEnabled():
            self.button_SetSegmentation.clearFocus()
# -----------------------------------------------------------------------------
# FUNCTIONS LINKED TO NAVIGATION
# connect the functions of the toolbar to our custom QPushbuttons.
lpbsscientist's avatar
lpbsscientist committed
    def ZoomTlbar(self):
        """The button_zoom is connected to the zoom function of the toolbar 
lpbsscientist's avatar
lpbsscientist committed
        already present in the matplotlib library.
        
        Depending on the buttons that are active or checked, when the zoom 
lpbsscientist's avatar
lpbsscientist committed
        function is used, it does not disable all the buttons.
lpbsscientist's avatar
lpbsscientist committed
        If the segment and threshold button are not checked or used
        when the zoom button is clicked, it disables all the button
        using self.Disable which disables everything except the button passed
        in argument (in this case button_zoom).
lpbsscientist's avatar
lpbsscientist committed
        If the zoom button is used while the segment button is checked,
        it disables all the buttons (1st elif) except the segment button
        but once it is finished (so the zoom button becomes unchecked) 
lpbsscientist's avatar
lpbsscientist committed
        then it enables only the editing buttons (as long as the segment
        button is still checked) such as New Cell, Add Region, Eraser, 
lpbsscientist's avatar
lpbsscientist committed
        Brush,etc.. and the other toolbar buttons (3rd elif)
lpbsscientist's avatar
lpbsscientist committed
        If the zoom button is clicked while the threshold button is checked,
        it disables all the button except the threshold button (2nd elif).
        Once the zoom button is unchecked, it enables the toolbar buttons
        (4th elif)
        In any other case, it just enables all the buttons again.
        """
        self.Nvgtlbar.zoom()
        if (self.button_zoom.isChecked() and not(self.button_segment.isChecked() 
            or self.button_threshold.isChecked())):
lpbsscientist's avatar
lpbsscientist committed
            self.Disable(self.button_zoom)
lpbsscientist's avatar
lpbsscientist committed
        elif self.button_zoom.isChecked() and self.button_segment.isChecked():
            self.Disable(self.button_zoom)
            self.button_segment.setEnabled(True)
lpbsscientist's avatar
lpbsscientist committed
        elif self.button_zoom.isChecked() and self.button_threshold.isChecked():
            self.Disable(self.button_zoom)
            self.button_threshold.setEnabled(True)
lpbsscientist's avatar
lpbsscientist committed
        elif self.button_zoom.isChecked() == False and self.button_segment.isChecked():
            self.button_pan.setEnabled(True)
            self.button_home.setEnabled(True)
            self.button_back.setEnabled(True)
            self.button_forward.setEnabled(True)
            self.EnableCorrectionsButtons()
lpbsscientist's avatar
lpbsscientist committed
        elif self.button_zoom.isChecked() == False and self.button_threshold.isChecked():
            self.button_pan.setEnabled(True)
            self.button_home.setEnabled(True)
            self.button_back.setEnabled(True)
            self.button_forward.setEnabled(True)
lpbsscientist's avatar
lpbsscientist committed
        else:
            self.Enable(self.button_zoom)
lpbsscientist's avatar
lpbsscientist committed
    def HomeTlbar(self):
        """
        connects the home button to the home function of the matplotlib
        toolbar. It sets the view to the original view (no zoom)
        """
lpbsscientist's avatar
lpbsscientist committed
        self.Nvgtlbar.home()

lpbsscientist's avatar
lpbsscientist committed
    def BackTlbar(self):
        """
        It calls the back function of the matplotlib toolbar which sets the 
        view to the previous one (if the user does several zooms/pans, 
        this button allows to go back in the "history of views")
        """
lpbsscientist's avatar
lpbsscientist committed
        self.Nvgtlbar.back()

lpbsscientist's avatar
lpbsscientist committed
    def ForwardTlbar(self):
        """
        It calls the forward function of the matplotlib toolbar which sets the 
        view to the next one (if the user does several zooms/pans, 
        this button allows to go forward in the "history of views"
        """
lpbsscientist's avatar
lpbsscientist committed
        self.Nvgtlbar.forward()

lpbsscientist's avatar
lpbsscientist committed
    def PanTlbar(self):
        """The button_pan is connected to the pan function of the toolbar 
lpbsscientist's avatar
lpbsscientist committed
        already present in the matplotlib library.
        
        Depending on the buttons that are active or checked, when the pan 
lpbsscientist's avatar
lpbsscientist committed
        function is used, it does not disable all the buttons.
lpbsscientist's avatar
lpbsscientist committed
        If the segment and threshold button are not checked or used
        when the pan button is clicked, it disables all the button
        using self.Disable which disables everything except the button passed
        in argument (in this case button_pan).
lpbsscientist's avatar
lpbsscientist committed
        If the pan button is used while the segment button is checked,
        it disables all the buttons (1st elif) except the segment button
        but once it is finished (so the zoom button becomes unchecked) 
lpbsscientist's avatar
lpbsscientist committed
        then it enables only the editing buttons (as long as the segment
        button is still checked) such as New Cell, Add Region, Eraser, 
lpbsscientist's avatar
lpbsscientist committed
        Brush,etc.. and the other toolbar buttons (3rd elif)
lpbsscientist's avatar
lpbsscientist committed
        If the pan button is clicked while the threshold button is checked,
        it disables all the button except the threshold button (2nd elif).
        Once the pan button is unchecked, it enables the toolbar buttons
        (4th elif)
        In any other case, it just enables all the buttons again.
        """

        self.Nvgtlbar.pan()

        if (self.button_pan.isChecked() and not(self.button_segment.isChecked()
            or self.button_threshold.isChecked())):
lpbsscientist's avatar
lpbsscientist committed
            self.Disable(self.button_pan)
lpbsscientist's avatar
lpbsscientist committed
        elif self.button_pan.isChecked() and self.button_segment.isChecked():
            self.Disable(self.button_pan)
            self.button_segment.setEnabled(True)
lpbsscientist's avatar
lpbsscientist committed
        elif self.button_pan.isChecked() and self.button_threshold.isChecked():
            self.Disable(self.button_pan)
            self.button_threshold.setEnabled(True)
lpbsscientist's avatar
lpbsscientist committed
        elif not(self.button_pan.isChecked()) and self.button_segment.isChecked():
lpbsscientist's avatar
lpbsscientist committed
            self.button_zoom.setEnabled(True)
            self.button_home.setEnabled(True)
            self.button_back.setEnabled(True)
            self.button_forward.setEnabled(True)
lpbsscientist's avatar
lpbsscientist committed
            self.EnableCorrectionsButtons()
lpbsscientist's avatar
lpbsscientist committed
        elif not(self.button_pan.isChecked()) and self.button_threshold.isChecked():
            self.button_zoom.setEnabled(True)
            self.button_home.setEnabled(True)
            self.button_back.setEnabled(True)
            self.button_forward.setEnabled(True)
lpbsscientist's avatar
lpbsscientist committed
        else:
            self.Enable(self.button_pan)


# -----------------------------------------------------------------------------
# EXTRACTING FLUORESCENCE
            
lpbsscientist's avatar
lpbsscientist committed
    def ButtonFluo(self):
        """This function is called everytime the Extract Fluorescence button is 
        clicked (self.button_extractfluorescence). 
        
        self.FlagFluoExtraction is boolean which is True when the path to the 
lpbsscientist's avatar
lpbsscientist committed
        excel file has already been loaded into self.xlsfilename.
        This pathname changes for each field of view as it is thought to have
        one xls file per field of view. 
lpbsscientist's avatar
lpbsscientist committed
        So at the beginning and each time the user changes field of view,
        self.FlagFluoExtraction is set to False.
lpbsscientist's avatar
lpbsscientist committed
        When it is set to False, this function calls a dialog window where
        the user is asked to load an already existing xls file for the current
        field of view or to give a name to create a new xls file for
        the current field of view. (self.Dialogxls)
lpbsscientist's avatar
lpbsscientist committed
        If it set to true, it means that self.xlsfilename contains the path
        to the xls file for the current field of view and it is directly given
        to the function that writes the fluorescence into the xls file.
        (self.ExtractFluo)
lpbsscientist's avatar
lpbsscientist committed
        if self.FlagFluoExtraction:
            self.ExtractFluo(self.xlsfilename)
        else:
            self.DialogXls()
lpbsscientist's avatar
lpbsscientist committed
    def DialogXls(self):
        """This function creates a dialog window which gives two options to the
        user either to load an existing xls file or to give a new name in order
        to create a new xls file. 
lpbsscientist's avatar
lpbsscientist committed
        """
        # creates the window
lpbsscientist's avatar
lpbsscientist committed
        dwind = ddb.FileBrowser()

        # this test is True if the user presses ok in the dialog window, if the 
        # user presses cancels it returns and nothing happens.
lpbsscientist's avatar
lpbsscientist committed
        if dwind.exec_():
            xlsname = dwind.xlsname
            newxlsname = dwind.newxlsentry.text()

            # if the string containing the filepath to an existing xls file 
            # is not empty then it calls directly the function to write the 
            # data into this existing xls file and sets self.xlsfilename
lpbsscientist's avatar
lpbsscientist committed
            if xlsname:
                self.xlsfilename = xlsname
                self.ExtractFluo(xlsname)
            # if xlsname is empty then it creates a new pathfilename and puts
            # the new created xls file into the folder where nd2 is located.
            # the string containing the nd2 namepath is split
lpbsscientist's avatar
lpbsscientist committed
            else:
                xlsname = ''
                templist = self.nd2path.split('/')
lpbsscientist's avatar
lpbsscientist committed
                for k in range(0, len(templist)-1):
lpbsscientist's avatar
lpbsscientist committed
                    xlsname = xlsname + templist[k] + '/'
                # this is the new path/filename
lpbsscientist's avatar
lpbsscientist committed
                xlsname = xlsname + newxlsname + '.xlsx'
                self.xlsfilename = xlsname
lpbsscientist's avatar
lpbsscientist committed
                self.CreateXls(xlsname)
                self.ExtractFluo(xlsname)
                
            # this flag is set to true, for the current field of view each
            # time extract fluorescence is clicked it writes in the file located
            # at self.xlsfilename.
lpbsscientist's avatar
lpbsscientist committed
            self.FlagFluoExtraction = True
lpbsscientist's avatar
lpbsscientist committed
            return


    def CreateXls(self, xlsfilename):
        """In case there is no xls file existing, here a new one is created 
lpbsscientist's avatar
lpbsscientist committed
        and prepared. For each channel a new sheet is created.
        In the first row for each sheet, the time indices are written t = 0,
        t = 1, etc... but only every third column. Because in the row below,
        three values are extracted 'Total intensity', 'Area' and 'Variance'
        at each time index. So three columns for each time index are needed,
        for the three data points. 
lpbsscientist's avatar
lpbsscientist committed
        The first column is left empty (starting from third row) because
        the cell numbers will be written in there.
        """
lpbsscientist's avatar
lpbsscientist committed
        book = Workbook()
        nbrchannels = self.reader.sizec
lpbsscientist's avatar
lpbsscientist committed
        for i in range(0,nbrchannels):
            sheetname = self.reader.channel_names[i]
            # creates a sheet with the name of the corresponding channel.
lpbsscientist's avatar
lpbsscientist committed
            if i == 0:
                sheet = book.active
                sheet.title = sheetname
            else:
                sheet = book.create_sheet(sheetname)
            sheet.cell(1,1, 'Cell Number / Time axis')
            sheet.cell(2,1, 'labels')
            timeaxissize = self.reader.sizet
            # start writing the time index at column 1, column 0 is reserved for
            # cell numbers.
lpbsscientist's avatar
lpbsscientist committed
            timecolindex = 2
lpbsscientist's avatar
lpbsscientist committed
            for t in range(1,timeaxissize+1):
                sheet.cell(1,timecolindex).value = 't = {}'.format(t-1)
                sheet.cell(2,timecolindex).value = 'Total Intensity'
                sheet.cell(2,timecolindex+1).value = 'Total Area'
                sheet.cell(2,timecolindex+2).value = 'Mean Intensity'
                sheet.cell(2,timecolindex+3).value =  'Variance'
                # updates the index, where the next time index should be written
lpbsscientist's avatar
lpbsscientist committed
                timecolindex = timecolindex + 4
        try:
            book.save(xlsfilename)
        except TypeError:
            QMessageBox.critical(self, "Error", "TypeError encountered. \
                                 Make sure you have openpyxl version 3.0.1 \
                                 installed. If the problem persists contact \
                                 the developers.")
lpbsscientist's avatar
lpbsscientist committed
    def ExtractFluo(self, xlsfilename):
        """This is the function that takes as argument the filepath to the xls
        file and writes in the file.
        It iterates over the different channels (or the sheets of the file,
        each channel has one sheet.), and reads the image corresponding
        to the time, field of view and channel index. It reads the already
        existing file and makes a copy in which the data will be written in it.
lpbsscientist's avatar
lpbsscientist committed
        The first step of calculating the data is to iterate through each
        cell/segment of the mask (so each cell is a submatrix of one value
        in the matrix of the mask).
        For each of these value /cell, the area is extracted as being
        the number of pixels corresponding to this cell/value. 
lpbsscientist's avatar
lpbsscientist committed
        (it is known from the microscope settings how to convert
        the pixel in area).
        The total intensity is just the value of the pixel and it is added over
        all the pixels corresonding to the cell/value.
        The mean is then calculated as being the total intensity divided by
        the number of pixels (which here is equal to the area also).
        With the mean it is then possible to calculate the variance of the 
lpbsscientist's avatar
lpbsscientist committed
        signal for one cell/value.
lpbsscientist's avatar
lpbsscientist committed
        Then, it is checked if the value of the cell (cell number) already
        exists in the first column, if it already exists it continues to
        find the column corresponding to the time index where the values
        should be written. It sets the flag to True such that it does not
        write the cell as new one and adds it at the end of the column
lpbsscientist's avatar
lpbsscientist committed
        If the value is not found in the cell number column (new cell or
        first time writing in the file), the flag is False, thus it adds the 
lpbsscientist's avatar
lpbsscientist committed
        cell number at the end of the column.
        It then saves the xls file.
        """
        self.Disable(self.button_extractfluorescence)
        self.WriteStatusBar('Extracting the fluorescence...')
        # opens the file to read it.
lpbsscientist's avatar
lpbsscientist committed
        book = load_workbook(self.xlsfilename)


        # iterate over all the channels, so over all the sheets in the file
lpbsscientist's avatar
lpbsscientist committed
        for channel in range(0, self.reader.sizec):
            # loads the picture corresponding to the channel, time index and fov
lpbsscientist's avatar
lpbsscientist committed
            image = self.reader.LoadImageChannel(self.Tindex, self.FOVindex, channel)
            # loads the sheet to read out corresponding to the current channel
lpbsscientist's avatar
lpbsscientist committed
            sheet = book.worksheets[channel]
            # this index contains the value of the maximum number of rows in the
            # file, it is used to append at the end the cell number column a new
            # cell/value, and it is updated each time a new cell is added.
lpbsscientist's avatar
lpbsscientist committed
            tempidx = sheet.max_row
            # np.unique(array) returns an array which contains all the value
            # that appear in self.m.plotmask, so it returns every cell value
            # including the background (value 0) present in self.m.plotmask
lpbsscientist's avatar
lpbsscientist committed
            for val in np.unique(self.m.plotmask):
                # Skip background
                if val == 0:
                    continue
                
                # Calculate stats
                area = (self.m.plotmask == val).sum()
                tot_intensity = image[self.m.plotmask == val].sum()
                mean = tot_intensity/area
                var = np.var(image[self.m.plotmask==val])

                # if flag is false it means that the cell number
                # corresponding to val is not present in the xls file, first
                # column.
                flag = False
                
                # iterate over all the rows
                for row in range(sheet.max_row+1):
                     # test if in the first column 0, the number of the cell
                     # is already present
                     # if sheet.cell_value(row,0) == str(val):
                     if sheet.cell(row = row+1, column = 1).value == str(val):
                         
                         # if is present, the column corresponding to the
                         # current time index is by iterating over the cols.
                         for col in range(sheet.max_column+1):
                             # test if it is the right column
                             # if sheet.cell_value(0, col) == 't = {}'.format(self.Tindex):
                             if sheet.cell(row = 1, column = col+1).value == 't = {}'.format(self.Tindex):
                                 # write in the xls file at the row, col coord
                                 sheet.cell(row+1, col+1, str(tot_intensity))
                                 sheet.cell(row+1, col+2, str(area))
                                 sheet.cell(row+1,col+3, str(mean))
                                 sheet.cell(row+1, col+4, str(var))
                                 book.save(xlsfilename)

                                 # the flag is set to True so that it does
                                 # not execute the code where the cell is
                                 # added in the xls file in a new row.
                                 flag = True
                                 
                if not flag:
                # this lines are executed if a new cell is detected or if
                # if it is the first time to write in the file.
                    for col in range(sheet.max_column+1):
                        if sheet.cell(row = 1, column =  col+1).value == 't = {}'.format(self.Tindex):
                            # it write the cell value/cell number in the
                            # column
                            sheet.cell(tempidx+1,1, str(val))
                            
                            # writes the data extracted before
                            sheet.cell(tempidx+1,col+1,str(tot_intensity))
                            sheet.cell(tempidx+1, col+2, str(area))
                            sheet.cell(tempidx+1, col+3, str(mean))
                            sheet.cell(tempidx+1, col+4, str(var))
                            # it updates the number of rows as a new cell
                            # has been added, so there is one more row.
                            tempidx = tempidx + 1
                            book.save(xlsfilename)
lpbsscientist's avatar
lpbsscientist committed

        self.Enable(self.button_extractfluorescence)
        self.ClearStatusBar()

# -----------------------------------------------------------------------------
# NEURAL NETWORK
lpbsscientist's avatar
lpbsscientist committed
    def ShowHideCNNbuttons(self):
        
        """hide and show the buttons corresponding to the neural network.
            this function is called by the button CNN which is hidden. But
            if activated in the InitLayout.py then you can have a button
            which hides the CNN buttons (which are now on the normal also
            hidden...).
        """
        if self.button_hide_show.isChecked():
            self.button_cnn.setVisible(True)
            self.button_segment.setVisible(True)
            self.button_savesegmask.setVisible(True)
            self.button_threshold.setVisible(True)
            self.button_SetThreshold.setVisible(True)
            self.button_savethresholdmask.setVisible(True)
            self.button_SetSegmentation.setVisible(True)

        else:
            self.button_cnn.setVisible(False)
            self.button_segment.setVisible(False)
            self.button_savesegmask.setVisible(False)
            self.button_threshold.setVisible(False)
            self.button_SetThreshold.setVisible(False)
            self.button_savethresholdmask.setVisible(False)
            self.button_SetSegmentation.setVisible(False)
            

lpbsscientist's avatar
lpbsscientist committed
    def LaunchBatchPrediction(self):
        """This function is called whenever the button Launch CNN is pressed.
        It allows to run the neural network over a time range and selected
        field of views.
lpbsscientist's avatar
lpbsscientist committed
        It creates a dialog window with two entries, that define the time range
        and a list where the user can select the desired fields of view.
lpbsscientist's avatar
lpbsscientist committed
        Once it reads all the value, it calls the neural network function
        inside of self.PredThreshSeg and it does the prediction of the neural
        network, thresholds this prediction and then segments it.
        """
        
        self.WriteStatusBar('Running the neural network...')
        self.Disable(self.button_cnn)

        # creates a dialog window from the LaunchBatchPrediction.py file
lpbsscientist's avatar
lpbsscientist committed
        dlg = lbp.CustomDialog(self)
        # this if tests if the user pressed 'ok' in the dialog window
        if dlg.exec_() == QDialog.Accepted:
            # it tests if the user has entered some values
            # if not it ignores and returns.
            if not (dlg.entry1.text()!= '' and dlg.entry2.text() != ''):
                QMessageBox.critical(self, "Error", "No Time Specified")
                return 
            
            # reads out the entry given by the user and converts the index
            # to integers
            time_value1 = int(dlg.entry1.text())
            time_value2 = int(dlg.entry2.text())
    
            # it tests if the first value is smaller or equal such that
            # time_value1 is the lower range of the time range
            # and time_value2 the upper boundary of the range.
            if time_value1 > time_value2 :
                QMessageBox.critical(self, "Error", 'Invalid Time Constraints')
                return
            
            # displays that the neural network is running
            self.WriteStatusBar('Running the neural network...')
    
            #it iterates in the list of the user-selected fields 
            #of view, to return the corresponding index, the function
            #dlg.listfov.row(item) is used which gives an integer
            if len(dlg.listfov.selectedItems())==0:
                QMessageBox.critical(self, "Error", "No FOV Selected")
            
            for item in dlg.listfov.selectedItems():
                #iterates over the time indices in the range
                for t in range(time_value1, time_value2+1):                    
                    #calls the neural network for time t and selected
                    #fov
                    if dlg.entry_threshold.text() !=  '':
                        thr_val = float(dlg.entry_threshold.text())
                    else:
                        thr_val = None
                    if dlg.entry_segmentation.text() != '':
                        seg_val = int(dlg.entry_segmentation.text())
                    else:
                        seg_val = 10
                    self.PredThreshSeg(t, dlg.listfov.row(item), thr_val, seg_val)
lpbsscientist's avatar
lpbsscientist committed
                    
                    # if tracker has been checked then apply it
                    if dlg.tracking_checkbox.isChecked():
mattminder's avatar
mattminder committed
                        if t != time_value1:
                            temp_mask = self.reader.CellCorrespondance(t, dlg.listfov.row(item))
lpbsscientist's avatar
lpbsscientist committed
                            self.reader.SaveMask(t,dlg.listfov.row(item), temp_mask)
                        
                        else:
                            temp_mask = self.reader.LoadSeg(t, dlg.listfov.row(item))
                            self.reader.SaveMask(t,dlg.listfov.row(item), temp_mask)
            
mattminder's avatar
mattminder committed
            self.ReloadThreeMasks()
            
            
        self.m.UpdatePlots()
        self.ClearStatusBar()
        self.EnableCNNButtons()
        self.Enable(self.button_cnn)
lpbsscientist's avatar
lpbsscientist committed
    def PredThreshSeg(self, timeindex, fovindex, thr_val, seg_val):
          """
          This function is called in the LaunchBatchPrediction function.
          This function calls the neural network function in the
          InteractionDisk.py file and then thresholds the result
          of the prediction, saves this thresholded prediction.
          Then it segments the thresholded prediction and saves the
          segmentation. 
lpbsscientist's avatar
lpbsscientist committed
          """
          self.reader.LaunchPrediction(timeindex, fovindex)
          self.m.ThresholdMask = self.reader.ThresholdPred(thr_val, timeindex,fovindex)
          self.reader.SaveThresholdMask(timeindex, fovindex, self.m.ThresholdMask)
          self.m.SegmentedMask = self.reader.Segment(seg_val, timeindex,fovindex)
          self.reader.SaveSegMask(timeindex, fovindex, self.m.SegmentedMask)
mattminder's avatar
mattminder committed
          self.reader.SaveMask(timeindex, fovindex, self.m.SegmentedMask)
#    def LaunchPrediction(self):
#        """This function is not used in the gui, but it can be used to launch
#        the prediction of one picture, with no thresholding and no segmentation
#        """
#        if not(self.reader.TestPredExisting(self.Tindex, self.FOVindex)):
#            self.WriteStatusBar('Running the neural network...')
#            self.Disable(self.button_cnn)
#            self.reader.LaunchPrediction(self.Tindex, self.FOVindex)
#            
#            self.Enable(self.button_cnn)
#            
#            self.button_cnn.setEnabled(False)
#            self.button_threshold.setEnabled(True)
#            self.button_segment.setEnabled(True)
#            self.button_cellcorespondance.setEnabled(True)
#            self.ClearStatusBar()
lpbsscientist's avatar
lpbsscientist committed
    def SelectChannel(self, index):
        """This function is called when the button to select different channels
        is used. From the displayed list in the button, the chosen index
        corresponnds to the same index in the list of channels from the reader.
        So, it sets the default channel with the new index (called index below)
        """
        self.reader.default_channel = index
        # update the pictures using the same function as the one used to 
        # change the fields of view.
lpbsscientist's avatar
lpbsscientist committed
        self.ChangeFOV()
lpbsscientist's avatar
lpbsscientist committed

    def SelectFov(self, index):
        """This function is called when the button containing the list of 
lpbsscientist's avatar
lpbsscientist committed
        fields od view is used.
        The index correspondds to the field of view selected in the list.
        """
        # mask is automatically saved.
lpbsscientist's avatar
lpbsscientist committed
        self.reader.SaveMask(self.Tindex, self.FOVindex, self.m.plotmask)
        self.FOVindex = index    
        
        # it updates the fov in the plot with the new index.
lpbsscientist's avatar
lpbsscientist committed
        self.ChangeFOV()
        # the flag of the fluorescence extraction is set to False (such that
        # if the user extracts fluorescence data in the new field of  view,
        # there is a dialog box asking to select the corresponding xls file
        # for this field of view. IF there is no data sheet for this fov, the
        # user can enter a new name to make a new file.)
lpbsscientist's avatar
lpbsscientist committed
        self.FlagFluoExtraction = False
    def ChangeFOV(self):
        """
        it changes the fov or channel according to the choice of the user
        and it updates the plot shown and it initializes the new fov/channel
        at t=0 by default.
        """
lpbsscientist's avatar
lpbsscientist committed
        self.Tindex = 0
        # load the image and mask for the current plot
lpbsscientist's avatar
lpbsscientist committed
        self.m.currpicture = self.reader.LoadOneImage(self.Tindex,self.FOVindex)
        self.m.plotmask = self.reader.LoadMask(self.Tindex,self.FOVindex)
        # sets the image and the mask to 0 for the previous plot
lpbsscientist's avatar
lpbsscientist committed
        self.m.prevpicture = np.zeros([self.reader.sizey, self.reader.sizex], dtype = np.uint16)
        self.m.prevplotmask = np.zeros([self.reader.sizey, self.reader.sizex], dtype = np.uint16)
        # load the image and the mask for the next plot, check if it exists
lpbsscientist's avatar
lpbsscientist committed
        if self.Tindex+1 < self.reader.sizet:
            self.m.nextpicture = self.reader.LoadOneImage(self.Tindex+1, self.FOVindex)
            self.m.nextplotmask = self.reader.LoadMask(self.Tindex+1, self.FOVindex)
            
            # enables the next frame button in case it was disabled when the 
            # fov/channel was changed
lpbsscientist's avatar
lpbsscientist committed
            self.button_nextframe.setEnabled(True)
        else:
            self.m.nextpicture = np.zeros([self.reader.sizey, self.reader.sizex], dtype = np.uint16)
            self.m.nextplotmask =  np.zeros([self.reader.sizey, self.reader.sizex], dtype = np.uint16)
            
            # disables the next frame button if the mask or the picture
lpbsscientist's avatar
lpbsscientist committed
            # does not exist.
            self.button_nextframe.setEnabled(False)
            
        # once the images and masks are loaded into the variables, they are 
        # displaye in the gui.
        self.m.UpdatePlots()
        # disables the previous frame button in case it was active before 
        # changing fov/channel.
lpbsscientist's avatar
lpbsscientist committed
        self.button_previousframe.setEnabled(False)
        # updates the title of the plots to display the right time indices
        # aboves the plots.
lpbsscientist's avatar
lpbsscientist committed
        self.UpdateTitleSubplots()
        # if the button to hide the mask was checked before changing fov/channel,
        # it hides the mask again.
lpbsscientist's avatar
lpbsscientist committed
        if self.button_hidemask.isChecked():
            self.m.HideMask()
        # the button to set the time index is also set to 0/default again.
lpbsscientist's avatar
lpbsscientist committed
        self.button_timeindex.setText('')
        # enables the neural network buttons if there is already an 
        # existing prediction for the current image.
lpbsscientist's avatar
lpbsscientist committed
        self.EnableCNNButtons()
lpbsscientist's avatar
lpbsscientist committed
        
        
    def ReloadThreeMasks(self):
        """
        A function which replots all the masks at the current time and fov 
        indices. Needed after the batch prediction is completed to display
        the result of the NN.
        """
        
        if self.Tindex >= 0 and self.Tindex <= self.reader.sizet-1:
            if self.Tindex == 0:
                self.button_nextframe.setEnabled(True)
                
lpbsscientist's avatar
lpbsscientist committed
                if self.Tindex < self.reader.sizet-1:
                    self.m.nextplotmask = self.reader.LoadMask(self.Tindex+1, self.FOVindex)
                else:
                    np.zeros([self.reader.sizey, self.reader.sizex], dtype = np.uint16)
lpbsscientist's avatar
lpbsscientist committed
                
                self.m.plotmask = self.reader.LoadMask(self.Tindex, self.FOVindex)
                self.m.prevplotmask = np.zeros([self.reader.sizey, self.reader.sizex], dtype = np.uint16)
                self.m.UpdatePlots()
lpbsscientist's avatar
lpbsscientist committed
                self.button_previousframe.setEnabled(False)
                
                
            elif self.Tindex == self.reader.sizet-1:
                self.button_previousframe.setEnabled(True)
                self.m.prevplotmask = self.reader.LoadMask(self.Tindex-1, self.FOVindex)
                self.m.plotmask = self.reader.LoadMask(self.Tindex, self.FOVindex)
                self.m.nextplotmask =  np.zeros([self.reader.sizey, self.reader.sizex], dtype = np.uint16)
                self.m.UpdatePlots()
lpbsscientist's avatar
lpbsscientist committed
                self.button_nextframe.setEnabled(False)
                
            else:
                self.button_nextframe.setEnabled(True)
                self.button_previousframe.setEnabled(True)
                self.m.prevplotmask = self.reader.LoadMask(self.Tindex-1, self.FOVindex)
                self.m.plotmask = self.reader.LoadMask(self.Tindex, self.FOVindex)              
                self.m.nextplotmask = self.reader.LoadMask(self.Tindex+1, self.FOVindex)
                self.m.UpdatePlots()
lpbsscientist's avatar
lpbsscientist committed
            
            self.UpdateTitleSubplots()
lpbsscientist's avatar
lpbsscientist committed
            if self.button_hidemask.isChecked():
                self.m.HideMask()
            self.EnableCNNButtons()
        
        else:
            return