melage 0.0.65__py3-none-any.whl

melage/widgets/brain_extraction_dl_bu.py

@@ -0,0 +1,869 @@
+__AUTHOR__ = 'Bahram Jafrasteh'
+
+"""
+    Main BrainExtractor class
+"""
+import os
+import numpy as np
+import nibabel as nib
+from PyQt5 import QtCore, QtGui, QtWidgets
+from functools import partial
+
+
+from PyQt5.QtCore import pyqtSignal
+import sys
+from PyQt5.QtWidgets import QDialog, QApplication, QPushButton, QVBoxLayout
+from PyQt5.QtCore import Qt
+from melage.utils.utils import convert_to_ras, LargestCC
+def to_str(val):
+    return '{:.2f}'.format(val)
+
+def remove_zero(f_data, value=0):
+    """
+    Remove non segmented areas from image
+    :param f_data:
+    :param value:
+    :return:
+    """
+
+    xs, ys, zs = np.where(f_data > value) #find zero values
+    tol = 4
+
+    min_max = []
+    for x in [xs, ys, zs]:
+        minx = min(x)-tol if min(x)-tol>1 else min(x)
+        maxx = max(x) + tol if max(x) + tol < f_data.shape[0]-1 else max(x)
+        min_max.append([minx, maxx])
+    f_data = f_data[min_max[0][0]:min_max[0][1] + 1, min_max[1][0]:min_max[1][1] + 1, min_max[2][0]:min_max[2][1] + 1]
+
+    return f_data, min_max
+
+def centralize_image(img, maxas=128, border=None):
+    """
+    Put image in the center
+    :param img:
+    :param maxas:
+    :param border:
+    :return:
+    """
+
+    n = img.shape
+    if type(maxas) != list:
+        maxas= [maxas, maxas, maxas]
+    pads = np.array([maxas[i] - a for i, a in enumerate(n)])
+    pads_r = pads // 2
+    pads_l = pads - pads_r
+    npads_l = pads_l * -1
+    npads_r = pads_r * -1
+    if border is None:
+        border = img[0,0,0]
+    new_img = np.ones((maxas[0], maxas[1], maxas[2]))*border
+
+    pads_r[pads_r < 0] = 0
+    pads_l[pads_l < 0] = 0
+    npads_l[npads_l < 0] = 0
+    npads_r[npads_r < 0] = 0
+    # print(pads_l, pads_r)
+    new_img[pads_r[0]:maxas[0] - pads_l[0], pads_r[1]:maxas[1] - pads_l[1], pads_r[2]:maxas[2] - pads_l[2]] = img[
+                                                                                                  npads_r[0]:n[0] -npads_l[0],
+                                                                                                  npads_r[1]:n[1] -npads_l[1],
+                                                                                                  npads_r[2]:n[2] -npads_l[2]]
+    return new_img, [pads, pads_l, pads_r, npads_l, npads_r, n]
+
+class BE_DL(QDialog):
+    """
+    This class has been implemented to use deep learning algorithms for brain extraction
+    """
+    closeSig = pyqtSignal()
+    betcomp = pyqtSignal(int)
+    datachange = pyqtSignal()
+    back_orig = pyqtSignal(int)
+
+    backbutton = pyqtSignal()
+    """
+
+    """
+    def __init__(self, parent=None
+                 ):
+        super(BE_DL, self).__init__(parent)
+        #self.load_filepath = 'widgets/Hybrid_latest.pth'
+        self._curent_weight_dir = os.path.dirname(os.path.join(os.getcwd()))
+        """
+        Initialization of Brain Extractor
+
+        Computes image range/thresholds and
+        estimates the brain radius
+        """
+    def set_pars(self, threshold=-0.5, remove_extra_bone=True):
+        #print("Initializing...")
+
+        # get image resolution
+
+        self.threshold = threshold
+        self.borderp = 0.0
+
+        # store brain extraction parameters
+        self.setupUi()
+    def setData(self, img, res):
+        # store the image
+        self.img = img
+        self.initial_mask = None
+        self.shape = img.shape  # 3D shape
+
+
+
+
+    def activate_advanced(self, value):
+        self.widget.setEnabled(value)
+
+    def setupUi(self):
+        Dialog = self.window()
+        Dialog.setObjectName("N4")
+        Dialog.resize(500, 220)
+        self.grid_main = QtWidgets.QGridLayout(self)
+        self.grid_main.setContentsMargins(0, 0, 0, 0)
+        self.grid_main.setObjectName("gridLayout")
+
+        self.hbox = QtWidgets.QHBoxLayout()
+        self.hbox.setContentsMargins(0, 0, 0, 0)
+        self.hbox.setObjectName("gridLayout")
+        self.grid_main.addLayout(self.hbox, 0, 0)
+
+        self.checkBox = QtWidgets.QCheckBox()
+        self.hbox.addWidget(self.checkBox, 0)
+        self.checkBox.setObjectName("checkBox")
+        self.checkBox.stateChanged.connect(self.activate_advanced)
+        self.comboBox_image = QtWidgets.QComboBox()
+        self.comboBox_image.setGeometry(QtCore.QRect(20, 10, 321, 25))
+        self.comboBox_image.setObjectName("comboBox")
+        for i in range(2):
+            self.comboBox_image.addItem("")
+        self.comboBox_image.currentIndexChanged.connect(self.datachange)
+
+        self.comboBox_models = QtWidgets.QComboBox()
+        self.comboBox_models.setGeometry(QtCore.QRect(20, 10, 321, 25))
+        self.comboBox_models.setObjectName("comboBox")
+        for i in range(2):
+            self.comboBox_models.addItem("")
+        self.comboBox_models.currentIndexChanged.connect(partial(self.parChanged, True))
+
+
+        self.hbox.addWidget(self.comboBox_image, 1)
+        self.hbox.addWidget(self.comboBox_models, 2)
+
+
+        self.progressBar = QtWidgets.QProgressBar()
+        self.progressBar.setEnabled(False)
+        self.progressBar.setProperty("value", 0)
+        self.progressBar.setAlignment(QtCore.Qt.AlignCenter)
+        self.progressBar.setObjectName("progressBar")
+        self.progressBar.setMaximum(100)
+
+
+
+        self.hbox2 = QtWidgets.QHBoxLayout()
+        self.hbox2.setContentsMargins(0, 0, 0, 0)
+        self.hbox2.setObjectName("gridLayout")
+        self.hbox2.addWidget(self.progressBar, 1)
+
+
+
+        self.widget = QtWidgets.QWidget()
+        self.widget.setObjectName("widget")
+        self.gridLayout = QtWidgets.QGridLayout(self.widget)
+        self.gridLayout.setContentsMargins(0, 0, 0, 0)
+        self.gridLayout.setObjectName("gridLayout")
+        self.splitter_3 = QtWidgets.QSplitter(self.widget)
+        self.splitter_3.setOrientation(QtCore.Qt.Horizontal)
+        self.splitter_3.setObjectName("splitter_3")
+
+
+        self.gridLayout.addWidget(self.splitter_3, 0, 0, 1, 1)
+        self.splitter = QtWidgets.QSplitter(self.widget)
+        self.splitter.setOrientation(QtCore.Qt.Horizontal)
+        self.splitter.setObjectName("splitter")
+        self.checkbox_thresholding = QtWidgets.QCheckBox(self.splitter)
+        #self.checkbox_thresholding.setAlignment(QtCore.Qt.AlignCenter)
+        self.checkbox_thresholding.setObjectName("checkbox_thresholding")
+        self.checkbox_thresholding.setChecked(False)
+
+        self.checkbox_otsu = QtWidgets.QCheckBox(self.splitter)
+        #self.checkbox_otsu.setAlignment(QtCore.Qt.AlignCenter)
+        self.checkbox_otsu.setObjectName("checkbox_thresholding")
+        self.checkbox_otsu.setChecked(False)
+        self.checkbox_otsu.stateChanged.connect(self.resetParams)
+
+
+        #self.checkbox_bone = QtWidgets.QCheckBox(self.splitter)
+        #self.checkbox_bone.setObjectName("histogram_bone")
+        #self.checkbox_bone.setChecked(False)
+        self.label_type = QtWidgets.QLabel(self.splitter)
+        self.label_type.setAlignment(QtCore.Qt.AlignCenter)
+        self.label_type.setObjectName("label")
+        self.comboBox_image_type = QtWidgets.QComboBox(self.splitter)
+        self.comboBox_image_type.setGeometry(QtCore.QRect(20, 10, 321, 25))
+        self.comboBox_image_type.setObjectName("comboBox")
+        self.comboBox_image_type.setCurrentIndex(0)
+        self.comboBox_image_type.currentIndexChanged.connect(partial(self.parChanged, False))
+        for i in range(2):
+            self.comboBox_image_type.addItem("")
+
+
+        #self.histogram_threshold_min = QtWidgets.QDoubleSpinBox(self.splitter)
+        #self.histogram_threshold_min.setObjectName("histogram_threshold_min")
+        #self.histogram_threshold_min.setValue(6)#(self.ht_min)*100)
+        #self.histogram_threshold_min.setMaximum(10)
+        #self.histogram_threshold_min.setMinimum(0)
+
+        #self.histogram_threshold_max = QtWidgets.QDoubleSpinBox(self.splitter)
+        #self.histogram_threshold_max.setObjectName("histogram_threshold_max")
+        #self.histogram_threshold_max.setValue((self.ht_max) * 100)
+        #self.histogram_threshold_max.setMaximum(100)
+        #self.histogram_threshold_max.setMinimum(0)
+
+        #self.histogram_threshold_min.setEnabled(True)
+        #self.histogram_threshold_max.setEnabled(False)
+
+        self.gridLayout.addWidget(self.splitter, 1, 0, 1, 1)
+
+
+
+        self.splitter_2 = QtWidgets.QSplitter(self.widget)
+        self.splitter_2.setOrientation(QtCore.Qt.Horizontal)
+        self.splitter_2.setObjectName("splitter_2")
+
+
+        self.label = QtWidgets.QLabel(self.splitter_2)
+        self.label.setAlignment(QtCore.Qt.AlignCenter)
+        self.label.setObjectName("label")
+        self.fractional_threshold = QtWidgets.QDoubleSpinBox(self.splitter_2)
+        self.fractional_threshold.setObjectName("fractional_threshold")
+
+        self.fractional_threshold.setMaximum(10)
+        self.fractional_threshold.setMinimum(-10)
+        self.fractional_threshold.setSingleStep(0.1)
+        self.fractional_threshold.setValue(self.threshold)
+
+        self.pushButton_original = QtWidgets.QPushButton(self.widget)
+        self.pushButton_original.setObjectName("pushButton_2")
+        self.pushButton_original.clicked.connect(self.back_original)
+        self.pushButton_original.setVisible(False)
+
+        #self.label_pma = QtWidgets.QLabel(self.splitter_2)
+        #self.label_pma.setAlignment(QtCore.Qt.AlignCenter)
+        #self.label_pma.setObjectName("label")
+        #self.border_pix = QtWidgets.QDoubleSpinBox(self.splitter_2)
+        #self.border_pix.setObjectName("PMA")
+        #self.border_pix.setValue((self.borderp))
+        #self.border_pix.setMaximum(150)
+        #self.border_pix.setMinimum(0)
+        #self.border_pix.setSingleStep(5)
+        #self.border_pix.valueChanged.connect(partial(self.parChanged, False))
+        self.gridLayout.addWidget(self.splitter_2, 2, 0, 1, 1)
+
+
+        self.splitter_3 = QtWidgets.QSplitter(self.widget)
+        self.splitter_3.setOrientation(QtCore.Qt.Horizontal)
+        self.splitter_3.setObjectName("splitter_3")
+
+        self.label_fl = QtWidgets.QLabel(self.splitter_3)
+        self.label_fl.setAlignment(QtCore.Qt.AlignCenter)
+        self.label_fl.setObjectName("label_5")
+        self.load_filepath = os.path.join(self._curent_weight_dir)
+        self.bt_load_weight = QtWidgets.QPushButton(self.splitter_3)
+        self.bt_load_weight.setObjectName("pushButton")
+        self.bt_load_weight.pressed.connect(self.load_weight_dialog)
+        self.bt_load_weight.setDefault(False)
+        self.pushButton = QtWidgets.QPushButton()
+
+        self.pushButton.setObjectName("pushButton")
+        self.pushButton.pressed.connect(self.accepted_emit)
+        self.hbox2.addWidget(self.pushButton, 0)
+        self.pushButton.setDefault(True)
+        self.gridLayout.addWidget(self.splitter_3, 3, 0, 1, 1)
+
+        self.widget.setEnabled(False)
+
+
+        self.grid_main.addWidget(self.widget)
+        self.grid_main.addLayout(self.hbox2, 20, 0)
+
+        self.label_pr = QtWidgets.QLabel()
+        self.label_pr.setAlignment(QtCore.Qt.AlignCenter)
+        self.label_pr.setObjectName("label_2")
+        self.label_pr.setText('fdfdf')
+        self.label_pr.setVisible(False)
+
+        self.grid_main.addWidget(self.label_pr)
+        self.retranslateUi(Dialog)
+        QtCore.QMetaObject.connectSlotsByName(Dialog)
+
+
+    def resetParams(self, state):
+
+        self.init_im_rec = None
+        self.initial_mask = None
+
+    def back_original(self):
+        self.back_orig.emit(0)
+        self.pushButton_original.setVisible(False)
+
+    def parChanged(self, model_changed=True):
+        """
+        if model parameter changed the user needs to run the algorithm again
+        :param model_changed:
+        :return:
+        """
+        self.initial_mask = None
+        if model_changed:
+            ci = self.comboBox_models.currentIndex()
+            #if ci== 0:
+            #    self.fractional_threshold.setValue(-0.5)
+            #    self.load_filepath = os.path.join(self._curent_weight_dir,'unet3d_latest.pth')
+            if ci==0:
+                self.load_filepath = os.path.join(self._curent_weight_dir,'NMPNET.pth')
+                self.fractional_threshold.setValue(-0.5)
+            elif ci == 1:
+                self.load_filepath = os.path.join(self._curent_weight_dir,'synthstrip.1.pt')
+                self.fractional_threshold.setValue(1)
+            #elif ci==3:
+            #    self.load_filepath = os.path.join(self._curent_weight_dir,'Hybrid_latest_custom.pth')
+            #    self.fractional_threshold.setValue(-0.5)
+            else:
+                return
+            self.label_fl.setText(self.load_filepath)
+    def load_weight_dialog(self):
+        opts = QtWidgets.QFileDialog.DontUseNativeDialog
+        pwd = os.path.abspath(__file__)
+        source_dir = os.path.dirname(os.path.dirname(pwd))
+        fileObj = QtWidgets.QFileDialog.getOpenFileName(self, "Open File", source_dir, "pth (*.pth *.pt)", options=opts)
+        if fileObj[0] == '':
+            return
+        self.load_filepath = fileObj[0]
+        self.label_fl.setText(self.load_filepath)
+        self.initial_mask = None
+        self.init_im_rec = None
+    def clear(self):
+        # store the image
+        self.img = None
+        self.shape = None  # 3D shape
+        self.mask = None
+        self.im_rec = None
+        self.initial_mask = None
+        self.init_im_rec = None
+
+    def retranslateUi(self, Dialog):
+        _translate = QtCore.QCoreApplication.translate
+        Dialog.setWindowTitle(_translate("Dialog", "DEEPL BET"))
+        self.checkBox.setText(_translate("Dialog", "Advanced"))
+        self.comboBox_image.setItemText(0, _translate("Dialog", "Top Image"))
+        self.comboBox_image.setItemText(1, _translate("Dialog", "Bottom Image"))
+        self.comboBox_image_type.setItemText(0, _translate("Dialog", "Ultrasound"))
+        self.comboBox_image_type.setItemText(1, _translate("Dialog", "MRI"))
+
+        self.comboBox_models.setItemText(0, _translate("Dialog", "NPM NET"))
+        self.comboBox_models.setItemText(1, _translate("Dialog", "SynthStrip"))
+
+
+        self.pushButton.setText(_translate("Dialog", "Apply"))
+        self.pushButton_original.setText(_translate("Dialog", "Back"))
+        self.checkbox_thresholding.setText(_translate("Dialog", "CUDA"))
+        self.checkbox_otsu.setText(_translate("Dialog", "Otsu"))
+
+        self.bt_load_weight.setText(_translate("Dialog", "Load Network Weights"))
+        self.label.setText(_translate("Dialog", "Threshold"))
+        self.label_type.setText(_translate("Dialog", "Image type"))
+
+
+
+        self.label_fl.setText(_translate("Dialog", self.load_filepath))
+
+
+    def closeEvent(self, a0: QtGui.QCloseEvent) -> None:
+        self.closeSig.emit()
+        super(BE_DL, self).closeEvent(a0)
+
+    def accepted_emit(self):
+        if not hasattr(self,'initial_mask'):
+            return
+        try:
+            self.label_pr.setVisible(True)
+            self.label_pr.setText('Initialization...')
+            self.pushButton_original.setVisible(False)
+
+            self.progressBar.setValue(5)
+            self._progress = 5
+
+            if self.initial_mask is None:
+                self.initial_mask, self.init_im_rec = self.initialization()
+
+            self.label_pr.setText('prediction...')
+            self.mask = self.compute_mask(self.initial_mask)
+
+            """
+            a = self.initial_mask.get_fdata().copy()
+            a *= self.mask
+            from melage.utils.utils import standardize
+            a = standardize(-a, 20)
+            a[self.mask==0] = -1
+            a = nib.Nifti1Image(a, self.initial_mask.affine, self.initial_mask.header)
+            a.to_filename('/home/binibica/eco/172_X_20210325_seg0.nii.gz')
+            """
+            if self.init_im_rec is not None:
+                from melage.utils.utils import standardize
+                a1 = standardize(self.img.get_fdata() * self.mask)
+                rec = standardize(self.init_im_rec.get_fdata() * self.mask)
+
+                mse = np.mean((a1 - rec) ** 2)
+                max_pixel_value = np.max(a1)
+                psnr = 10 * np.log10((max_pixel_value ** 2) / mse)
+                print("PSNR: {:.2f}".format(psnr))
+
+                from skimage.metrics import structural_similarity
+                print("Structural similarity : {:.2f}".format(structural_similarity(a1, rec)))
+
+
+
+                imB = self.init_im_rec.get_fdata().copy()
+                #bmin, bmax = imB[self.mask>0].min(), imB[self.mask>0].max()
+                #imB = (imB - bmin) / (bmax-bmin)  # range
+                imB[~(self.mask > 0)] = 0
+                #from scipy.ndimage import gaussian_filter
+                #alpha, sigma = 1.5, 5
+                #blurred = gaussian_filter(imB, 5)
+                #sharpened_mask =  (imB - blurred)
+
+                #sharpened_mask = (
+                #            (sharpened_mask - sharpened_mask.min()) / (sharpened_mask.max() - sharpened_mask.min()))
+                #sharpened_mask[~(self.mask > 0)] = 0
+                self.im_rec = nib.Nifti1Image(imB, self.init_im_rec.affine, self.init_im_rec.header)
+            else:
+                imB = self.img.get_fdata().copy()
+                imB[~(self.mask > 0)] = 0
+                self.im_rec = nib.Nifti1Image(imB, self.img.affine, self.img.header)
+            self.progressBar.setValue(98)
+            self.label_pr.setVisible(False)
+            self._progress = 100
+            self.progressBar.setValue(self._progress)
+            self._progress =0
+            self.betcomp.emit(True)
+            self.pushButton_original.setVisible(True)
+        except Exception as e:
+            print(e)
+            self.screen_error_msgbox(e.args[0])
+    def screen_error_msgbox(self, text= None):
+        if text is None:
+            text = 'There is an error. Screen is not captured. Please check the content.'
+        MessageBox = QtWidgets.QMessageBox(self)
+        MessageBox.setText(str(text))
+        MessageBox.setWindowTitle('Warning')
+        MessageBox.show()
+        self.progressBar.setValue(0)
+
+    def initialization(self):
+        def rescaleint8(x):
+            """
+            y = a+(b-a)*(x-min(x))/(max(x)-min(x))
+            Parameters
+            ----------
+            x
+
+            Returns
+            -------
+
+            """
+            oldMin, oldMax = int(x.min()), x.max()
+            NewMin, NewMax = 0, 1000
+            OldRange = (oldMax - oldMin)
+            NewRange = (NewMax - NewMin)
+            y = NewMin + (NewRange) * ((x - oldMin) / (OldRange))
+            return y
+        def signdf(df):
+            if df<=0:
+                return -1
+            else:
+                return 1
+        pv = self._progress
+        self.progressBar.setValue(pv+1)
+        #from melage.widgets.be_dl import MRI_bet
+        from melage.widgets.be_dl_unet3d import Unet3D
+
+        from melage.widgets.DeepLModels.Unet3DAtt import Unet3DAtt
+        from melage.widgets.Synthstrip import StripModel
+        # find the center of mass of image
+        from nibabel.processing import resample_to_output, resample_from_to
+        import torch
+        #imA = self.img
+        imA = self.img.__class__(self.img.dataobj[:], self.img.affine, self.img.header)
+
+        if self.checkbox_thresholding.isChecked():
+            if torch.cuda.is_available() :
+                # device = torch.device("cuda")
+                device = torch.device("cuda:0")
+                torch.cuda.set_device(0)
+                torch.backends.cudnn.benchmark = True
+            else:
+                device = torch.device("cpu")
+        else:
+            device = torch.device("cpu")
+        ci = self.comboBox_models.currentIndex()
+        if ci== 0:
+            #model = Unet3D(time_embed=True, spacing_embed=True)
+            model = Unet3DAtt(time_embed=True, spacing_embed=True)
+            #self.fractional_threshold.setValue(-0.5)
+        elif ci==1:
+            model = StripModel()
+            #self.fractional_threshold.setValue(-0.5)
+        #elif ci == 2:
+
+            #self.fractional_threshold.setValue(1)
+        #elif ci==2:
+        #    from melage.widgets.DeepLModels.NPP.models.model import UNet
+        #    self.fractional_threshold.setValue(0)
+        #    model = UNet()
+        #    self.fractional_threshold.setValue(-0.5)
+        else:
+            return
+        model.eval()
+
+        #transform, source = convert_to_ras(imA.affine, target='RAS')
+        #imA = imA.as_reoriented(transform)
+        #self.borderp = self.border_pix.value()
+        self.borderp = 0
+        if ci==1:
+            transform, source = convert_to_ras(imA.affine, target='LIA')
+            imAa = imA.as_reoriented(transform)
+            shape_initial = imAa.shape
+            image_used, pad_zero = remove_zero(imAa.get_fdata(), 0)
+            imAzero = nib.Nifti1Image(image_used, imAa.affine, imAa.header)
+            affine_used, header_used = imAa.affine, imAa.header.copy()
+            NewSpacing = 1
+            imAa = resample_to_output(imAzero, [NewSpacing, NewSpacing, NewSpacing])
+
+            transform, source = convert_to_ras(imAa.affine, target='LIA')
+            imAa = imAa.as_reoriented(transform)
+            target_shape = np.clip(np.ceil(np.array(imAa.shape[:3]) / 64).astype(int) * 64, 192, 320)
+        else:
+            shape_initial = imA.shape
+            border_value = imA.get_fdata()[0,0,0]
+            if self.checkbox_otsu.isChecked():
+                try:
+                    from melage.utils.utils import Threshold_MultiOtsu
+                    border_value = Threshold_MultiOtsu(imA.get_fdata(), 4)[0]
+                except:
+                    pass
+            image_used, pad_zero = remove_zero(imA.get_fdata(), border_value)
+            imAzero = nib.Nifti1Image(image_used, imA.affine, imA.header)
+            affine_used, header_used = imA.affine, imA.header.copy()
+            Spacing = imAzero.header['pixdim'][1:4]
+
+            NewSpacing = Spacing.max()  # [0]
+            if Spacing.max() != Spacing.min():
+                NewSpacing = Spacing.min()
+                Spacing = [Spacing.min()]
+            else:
+                Spacing = [Spacing[0]]
+
+            minus_sign = False
+            maxdim = np.array([el * sp / NewSpacing for el in imAzero.shape for sp in Spacing]).max()
+            df = (maxdim - 192)
+
+            if df>0:
+                resize_need = True
+                #minus_sign = True
+                prev_sign = signdf(df)
+                while True:
+                    maxdim = np.array([el * sp / NewSpacing for el in imAzero.shape for sp in Spacing]).max()
+                    df = (maxdim - 192)
+                    if signdf(df) != prev_sign:
+                        break
+                    if abs(abs(df) - self.borderp)<10: #and df <= 0:
+                        break
+                    if minus_sign:
+                        NewSpacing -= 0.1
+                    else:
+                        NewSpacing += 0.1
+                    prev_sign = signdf(df)
+                while True:
+                    imAa = resample_to_output(imAzero, [NewSpacing, NewSpacing, NewSpacing])
+                    df = (np.max(imAa.shape) - 192)
+                    if signdf(df) != prev_sign and df < 0:
+                        break
+                    prev_sign = signdf(df)
+                    if abs(df) <= self.borderp and df <= 0:
+                        break
+                    else:
+                        if signdf(df) == -1:
+                            minus_sign = True
+                            minus_signed = True
+                        else:
+                            minus_sign = False
+                    if minus_sign:
+                        NewSpacing -= 0.1
+                    else:
+                        NewSpacing += 0.1
+                transform, source = convert_to_ras(imAa.affine, target='RAS')
+                if source!='RAS':
+                    imAa = imAa.as_reoriented(transform)
+            else:
+                resize_need = False
+                transform, source = convert_to_ras(imAzero.affine, target='RAS')
+                imAa = imAzero.as_reoriented(transform)
+                #imAa = nib.Nifti1Image(imAzero.get_fdata(), imAzero.affine, imAzero.header)
+        #transform, source = convert_to_ras(imAa.affine, target='RAS')
+        #imAa = imAa.as_reoriented(transform)
+        pixdim = imAa.header['pixdim'][1:4]
+        affine = imAa.affine
+
+        if ci!=1:
+            imA = imAa.get_fdata()
+            #imA -= imA.min()
+            #imA = rescaleint8(imAa.get_fdata())/1000
+            #imA = (imA / imA.max()).clip(0, 1)
+            if ci==0:
+                imA = ((imA - imA.min()) / (imA.max() - imA.min()))
+            else:
+                imA = 2 * ((imA - imA.min()) / (imA.max() - imA.min())) - 1
+            #imA = (imA - imA.mean())/np.std(imA)
+            #imA = ((imA - imA.min()) / (imA.max() - imA.min()))
+        else:
+            imA = imAa.get_fdata()
+            imA -= imA.min()
+            if np.percentile(imA, 99)!=0:
+                imA = (imA / np.percentile(imA, 99)).clip(0, 1)
+            else:
+                imA = (imA / imA.max()).clip(0, 1)
+        if ci==1:
+            imA, info_back = centralize_image(imA, list(target_shape))
+        else:
+            imA, info_back = centralize_image(imA, 192)
+        del imAa
+        [pads, pads_l, pads_r, npads_l, npads_r, shape_img] = info_back
+
+
+
+        #rescaleint8(imA)
+        #imA -= imA.min()
+        #if np.percentile(imA, 99)!=0:
+        #    imA = (imA / np.percentile(imA, 99)).clip(0, 1)
+        #else:
+        #    imA = (imA / imA.max()).clip(0, 1)
+
+        imA = torch.from_numpy(imA).to(torch.float).unsqueeze(0).unsqueeze(0)
+
+
+        print('loading model weight...')
+        #pma = self.border_pix.value()
+        if self.comboBox_image_type.currentIndex()==0:
+            eco_mri = -1
+        else:
+            eco_mri = 1
+        #time = torch.from_numpy(np.array([pma, eco_mri])).to(device).unsqueeze(0).to(torch.float)
+        if ci!=0:
+            time = torch.from_numpy(np.array(eco_mri)).unsqueeze(0).to(torch.float)
+            time = torch.concat([time.reshape(-1,1), torch.from_numpy(pixdim).reshape(-1,1)]).squeeze().unsqueeze(0).to(device)
+        else:
+            time = torch.from_numpy(np.array(eco_mri)).unsqueeze(0).to(torch.float).to(device)
+        imA = imA.to(device)
+        model.to(device)
+        self.progressBar.setValue(40)
+        self.label_pr.setText('loading model...')
+        if ci==1:
+            state_dict = torch.load(self.load_filepath, map_location=device)
+            model.load_state_dict(state_dict['model_state_dict'])
+        else:
+            try:
+                state_dict = torch.load(self.load_filepath, map_location=device)
+                #d = OrderedDict((k.replace('norm', 'norm2') if 'norm' in k else k, v) for k, v in state_dict['model'].items())
+                if hasattr(state_dict, 'forward'):
+                    model = state_dict
+                else:
+                    try:
+                        model.load_state_dict(state_dict['state_dict'], strict=False)
+                    except:
+                        model.load_state_dict(state_dict['model'], strict=False)
+            except:
+                model = torch.load(self.load_filepath, map_location=device)
+
+        self.progressBar.setValue(50)
+        self.label_pr.setText('computing mask...')
+        im_mask = model.forward(imA, time)
+
+        if ci==0:
+            im_mask, im_rec=im_mask
+            im_mask = im_mask.detach().cpu().squeeze().numpy()
+            im_rec = im_rec.detach().cpu().squeeze().numpy()
+            maxas = [i for i in im_mask.shape]
+            im_rec = im_rec[pads_r[0]:maxas[0] - pads_l[0], pads_r[1]:maxas[1] - pads_l[1],
+                      pads_r[2]:maxas[2] - pads_l[2]]
+            im_rec = ((im_rec - im_rec.min()) * 255 / np.ptp(im_rec))
+            im_mask = im_mask[pads_r[0]:maxas[0] - pads_l[0], pads_r[1]:maxas[1] - pads_l[1],
+                      pads_r[2]:maxas[2] - pads_l[2]]
+        else:
+            im_mask = im_mask.detach().cpu().squeeze().numpy()
+            maxas = [i for i in im_mask.shape]
+            im_mask = im_mask[pads_r[0]:maxas[0] - pads_l[0], pads_r[1]:maxas[1] - pads_l[1], pads_r[2]:maxas[2] - pads_l[2]]
+            im_rec = None
+        if ci==1:
+            mask = nib.Nifti1Image(im_mask, affine, header_used)
+            mask = resample_from_to(mask, imAzero)
+            im_mask = np.zeros(shape_initial)
+            im_mask[pad_zero[0][0]:pad_zero[0][1] + 1, pad_zero[1][0]:pad_zero[1][1] + 1,
+            pad_zero[2][0]:pad_zero[2][1] + 1] = mask.get_fdata()
+        self.progressBar.setValue(80)
+        header = self.img.header.copy()
+
+        header['pixdim'][1:4] = pixdim
+        mask = nib.Nifti1Image(im_mask, affine, header)
+        if ci==1:
+            _, source = convert_to_ras(self.img.affine, target='LIA')
+            transform, _ = convert_to_ras(mask.affine, target=source)
+            mask = mask.as_reoriented(transform)
+        else:
+            mask = nib.Nifti1Image(im_mask, affine, header_used)
+            _, source = convert_to_ras(self.img.affine)
+            transform, _ = convert_to_ras(mask.affine, target=source)
+            mask = mask.as_reoriented(transform)
+            if im_rec is not None:
+                _rec = nib.Nifti1Image(im_rec, affine, header_used)
+                _rec = _rec.as_reoriented(transform)
+            else:
+                _rec = None
+            if resize_need:
+                mask = resample_from_to(mask, imAzero)
+                if _rec is not None:
+                    _rec = resample_from_to(_rec, imAzero)
+            im_mask = np.ones(shape_initial)*mask.get_fdata()[0,0,0]
+
+            im_mask[pad_zero[0][0]:pad_zero[0][1] + 1, pad_zero[1][0]:pad_zero[1][1] + 1,
+            pad_zero[2][0]:pad_zero[2][1] + 1] = mask.get_fdata()
+            mask = nib.Nifti1Image(im_mask, self.img.affine, self.img.header)
+            if ci==0:
+                im_rec = np.ones(shape_initial) * _rec.get_fdata()[0, 0, 0]
+                im_rec[pad_zero[0][0]:pad_zero[0][1] + 1, pad_zero[1][0]:pad_zero[1][1] + 1,
+                pad_zero[2][0]:pad_zero[2][1] + 1] = _rec.get_fdata()
+                im_rec = nib.Nifti1Image(im_rec, self.img.affine, self.img.header)
+
+        return mask, im_rec
+
+
+    def compute_mask(self, mask):
+        """
+        Convert surface mesh to volume
+        """
+        threshold = self.fractional_threshold.value()
+        try:
+            import surfa as sf
+            if type(mask)==sf.image.framed.Volume:
+                label = mask.copy()
+                label = (label < threshold).connected_component_mask(k=1, fill=True)
+                return label.data.astype('int')
+        except:
+            pass
+
+        #from nibabel.processing import resample_from_to
+        from skimage.measure import label as label_connector
+        from scipy.ndimage import binary_fill_holes
+
+
+        im_mask = mask.get_fdata().copy()
+        #from utils.utils import Threshold_MultiOtsu
+
+
+        ind = im_mask >= threshold
+        im_mask[ind] = 0
+        im_mask[~ind] = 1
+        """
+        
+        if self.checkbox_bone.isChecked():
+            A = self.img.get_fdata() * im_mask
+            thresholds = Threshold_MultiOtsu(A, 5)
+            im_mask[A > thresholds[-1]] = 0
+        """
+        #label correction
+        label_correction = False
+
+        labels = im_mask
+        labels = binary_fill_holes(labels)
+        labels, labels_freq = LargestCC(labels, connectivity=1)
+        argmax = np.argmax(
+            [self.img.get_fdata()[labels == el].sum() for el in range(len(labels_freq)) if el != 0]) + 1
+        #connectivity = 1
+        #labels = label_connector(labels, connectivity=connectivity)
+        # labels[labels>1]=1
+        #frequency = np.bincount(labels.flat)
+        # argmax = [el for el in np.argsort(-frequency) if el != 0][0]
+        #argmax = np.argmax([self.img.get_fdata()[labels == el].sum() for el in range(len(frequency)) if el != 0]) + 1
+        ind = labels != argmax
+        labels[ind] = 0
+        labels[~ind]=1
+
+        if label_correction:
+            connectivity = 1
+            labels = label_connector(im_mask, connectivity=connectivity)
+            # labels[labels>1]=1
+            frequency = np.bincount(labels.flat)
+            #argmax = [el for el in np.argsort(-frequency) if el != 0][0]
+            argmax = np.argmax([self.img.get_fdata()[labels==el].sum() for el in range(len(frequency)) if el != 0])+1
+            ind = labels != argmax
+            labels[ind] = 0
+            # labels[~ind]=0
+            labels = binary_fill_holes(labels)
+            for slc in range(labels.shape[0]):
+                im = labels[slc, :, :]
+                if im.max() == 0:
+                    continue
+                labels[slc, :, :] = binary_fill_holes(im)
+            for slc in range(labels.shape[1]):
+                im = labels[:, slc, :]
+                if im.max() == 0:
+                    continue
+                labels[:, slc, :] = binary_fill_holes(im)
+
+            for slc in range(labels.shape[2]):
+                im = labels[:, :, slc]
+                if im.max() == 0:
+                    continue
+
+                lbl = label_connector(im, connectivity=connectivity)
+                frequency = np.bincount(lbl.flat)
+                argmax = [el for el in np.argsort(-frequency) if el != 0][0]
+                ind_remove = frequency < 0.1 * frequency[argmax]
+                if ind_remove.any():
+                    index_remove = np.argwhere(ind_remove)
+                    for el in index_remove:
+                        ind = lbl == el
+                        lbl[ind] = 0
+                    lbl[lbl > 0] = 1
+                else:
+                    continue
+                labels[:, :, slc] = lbl.copy()
+
+            for slc in range(labels.shape[2]):
+                im = labels[:, :, slc]
+                if im.max() == 0:
+                    continue
+                labels[:, :, slc] = binary_fill_holes(im)
+
+        return labels.astype('int')
+
+
+def run():
+    import sys
+    app = QtWidgets.QApplication(sys.argv)
+    file = 't1_withoutmask.nii.gz'
+    import nibabel as nib
+    #m = nib.load(file)
+    #res = m.header["pixdim"][1]
+    #nibf = m.get_fdata()
+    from melage.utils.utils import standardize
+    #nibf = standardize(nibf)
+    window = BE_DL()
+    #window.setData(m, res)
+    window.set_pars()
+    window.show()
+    sys.exit(app.exec_())
+
+if __name__ == '__main__':
+    run()
+