def load_transf(self): cwd = os.getcwd() dpath, _filter = QtWidgets.QFileDialog.getOpenFileName(self, "Choose a transformation file", cwd, 'JSON (*.json)') if dpath != '': # Load the transformation with open(dpath, 'r') as f: workspace = json.load(f) image = workspace['image'] d_transf = workspace['manual'] l_cells = workspace['cells'] transf = Transform((d_transf['x_shift'], d_transf['y_shift']), d_transf['rotation'], d_transf['scale'], d_transf['flip_lr'], d_transf['flip_ud']) try: self.orientation_cb.setCurrentIndex(d_transf['atlas_orientation']) self.slice_sl.setValue(int(d_transf['atlas_slice'])) except KeyError: pass self.data_path = image self.transf = transf self.cells = l_cells self.cell_pos = [tuple(c['pos']) for c in l_cells] self.cell_scatter.setData(pos=self.cell_pos)
def clear_transf(self): self.transf = Transform()
def __init__(self): super().__init__() self.statusBar().showMessage('Loading...') self._logger.info('Loading...') Image.MAX_IMAGE_PIXELS = None # Necessary to open super large tiff files # Load atlas and annotations self.atlas = np.load('contourify2.npy') self.template = get_atlas('average_template_25.nrrd') self.raw_atlas = get_atlas('annotation_25.nrrd') self.colors = id_colors(self.onto) self.c_atlas = np.load('color_atlas.npy') self._logger.info('Atlas opened') self.p_max = 2**16 # Change the number value to open by default a specific atlas (0: coronal, 1: horizontal, 2: sagittal) self._c_orient = 0 self._sel_regions = {} self.orientation_cb.setCurrentIndex(2) self._data_path = None self.slice_image = SliceImage('') self.pic = np.zeros((5, 5), dtype=np.uint8) self._transf_raw = Image.new('L', (5, 5)) # Create a greyscale 8-bit image. Size of 5x5 self.transf_raw = self._transf_raw # Modifie la taille de la croix à l'ouverture, mais se réinitialise à l'ouverture de l'image. # Néanmoins, la fenêtre est plus petite, mais la sélection n'est plus la même (ne pointe plus sur la même chose) self._raw_inset = np.zeros((5, 5), dtype=np.uint8) self.raw_inset = self._raw_inset self._transf = Transform() self.transf = self._transf # Modifie le niveau de zoom de l'image de droite : plus tu augmentes, plus le zoom baisse (plus de pixel à l'écran donc moins zoomé) self._inset_size = 1000 self.cells = [] self.cell_pos = [] self.actions = [] self.rot_sl.setValue(90) self.scale_sl.setValue(int(self.transf.scale * self.scale_sl.factor)) # Capture mouse movements to scroll the inset self._proxy = pg.SignalProxy(self.vb_anat.scene().sigMouseMoved, rateLimit=0.1, delay=.1, slot=self.mouse_moved) self.setWindowTitle('Atlaser Sotfware') self.help_menu = None
class AtlasExplorer(Viewer): def __init__(self): super().__init__() self.statusBar().showMessage('Loading...') self._logger.info('Loading...') Image.MAX_IMAGE_PIXELS = None # Necessary to open super large tiff files # Load atlas and annotations self.atlas = np.load('contourify2.npy') self.template = get_atlas('average_template_25.nrrd') self.raw_atlas = get_atlas('annotation_25.nrrd') self.colors = id_colors(self.onto) self.c_atlas = np.load('color_atlas.npy') self._logger.info('Atlas opened') self.p_max = 2**16 # Change the number value to open by default a specific atlas (0: coronal, 1: horizontal, 2: sagittal) self._c_orient = 0 self._sel_regions = {} self.orientation_cb.setCurrentIndex(2) self._data_path = None self.slice_image = SliceImage('') self.pic = np.zeros((5, 5), dtype=np.uint8) self._transf_raw = Image.new('L', (5, 5)) # Create a greyscale 8-bit image. Size of 5x5 self.transf_raw = self._transf_raw # Modifie la taille de la croix à l'ouverture, mais se réinitialise à l'ouverture de l'image. # Néanmoins, la fenêtre est plus petite, mais la sélection n'est plus la même (ne pointe plus sur la même chose) self._raw_inset = np.zeros((5, 5), dtype=np.uint8) self.raw_inset = self._raw_inset self._transf = Transform() self.transf = self._transf # Modifie le niveau de zoom de l'image de droite : plus tu augmentes, plus le zoom baisse (plus de pixel à l'écran donc moins zoomé) self._inset_size = 1000 self.cells = [] self.cell_pos = [] self.actions = [] self.rot_sl.setValue(90) self.scale_sl.setValue(int(self.transf.scale * self.scale_sl.factor)) # Capture mouse movements to scroll the inset self._proxy = pg.SignalProxy(self.vb_anat.scene().sigMouseMoved, rateLimit=0.1, delay=.1, slot=self.mouse_moved) self.setWindowTitle('Atlaser Sotfware') self.help_menu = None def switch_cellmode(self, checked): if checked: self.raw_inset = np.zeros((5, 5), dtype=np.uint8) self.transf_raw = self.apply_transf(self.slice_image.raw_img, False, False) self._proxy.rateLimit = 30 self.statusBar().showMessage('Ready for cell selection', 1500) def draw_cross(self): """ Draw the cross cursor on the inset zoom picture Place it at the center of the inset, which depends on the exact size of the self._raw_inset array """ # inset size w, h = self._raw_inset.shape w2, h2 = w//2, h//2 # Cross on the inset picture self._h_line_l.setLine(0, h2, w2, h2) self._h_line_r.setLine(w2, h2, w, h2) self._v_line_u.setLine(w2, h2, w2, h) self._v_line_d.setLine(w2, h2, w2, 0) def clear_cells(self): self.cells = [] self.cell_pos = [] self.cell_scatter.setData(self.cell_pos) def clear_transf(self): self.transf = Transform() def mouse_moved(self, pos): """ Update the zoom inset from the raw image when the mouse cursor moves over the low resolution atlas aligned image Parameters ---------- pos: list of SceneEvents Contains the mouse position (in scene coordinates, will need conversion) """ # FIXME: When too close to the edges, not possible to get a precise position # Ne marche pas car : # Zone de clic = zone de l'atlas # Si on sort la souris de l'atlas, ce n'est plus précis car on ne peut plus cliquer dans tous les cas if not self.cell_select_cb.isChecked(): return pos = pos[0] v_range = self.vb_anat.viewRange() x, y = self.convert_mouse_pos(pos.x(), pos.y(), v_range[0][0], v_range[1][0]) x, y = x, y all_scales = self.slice_image.downfactor / self.transf.scale raw_x = int((x - self.transf.translation[1]) * all_scales) raw_y = int((y - self.transf.translation[0]) * all_scales) # If the mouse is out of the picture, don't bother if raw_x > self.transf_raw.height or raw_x < 0 or raw_y > self.transf_raw.width or raw_y < 0: return if self.transf_raw.width < 10: return raw_x, raw_y = raw_y, raw_x start_x = np.clip(raw_x - self._inset_size, 0, self.transf_raw.width - self._inset_size) start_y = np.clip(raw_y - self._inset_size, 0, self.transf_raw.height - self._inset_size) stop_x = np.clip(raw_x + self._inset_size, self._inset_size, self.transf_raw.width - 1) stop_y = np.clip(raw_y + self._inset_size, self._inset_size, self.transf_raw.height - 1) self.raw_inset = np.array(self.transf_raw.crop((start_x, start_y, stop_x, stop_y))) self.draw_cross() def closeEvent(self, a0: QtGui.QCloseEvent) -> None: """ closeEvent of Viewer. Perform some cleanup """ # if self.bioformat: # javabridge.kill_vm() # self._logger.info('Kill the Java VM') a0.accept() super().closeEvent(a0) def export_points(self): try : for i in range(0, len(self.cells)): if (self.cells[i]['Region'] == 'None'): d = defaultdict(str) d[1] = "" d[2] = "" d = dict(d) self.cells[i].update(d) self.cells[i]['structure'] = self.cells[i].pop('structure') del self.cells[i]['Region'] else: d = defaultdict(str) d[1] = self.cells[i]['Region'][len(self.cells[i]['Region']) - 2] d[2] = self.cells[i]['Region'][len(self.cells[i]['Region']) - 3] d = dict(d) self.cells[i].update(d) self.cells[i]['structure'] = self.cells[i].pop('structure') del self.cells[i]['Region'] im_path = Path(self.data_path) path = im_path.parent / f'cells_{im_path.stem}.csv' keys = self.cells[i].keys() with open(path, 'a', newline='') as file: dict_writer = csv.DictWriter(file, keys) dict_writer.writerows(self.cells) self.cells = [] except KeyError: pass # Displays the image in the right windows when cell mode selection is activated def apply_transf(self, img, scale=True, trans=True): new_img = img.copy() if self.transf.flipped_lr: new_img = Image.fromarray(np.flipud(np.array(new_img))) if self.transf.flipped_ud: new_img = Image.fromarray(np.fliplr(np.array(new_img))) n_width = int(new_img.width*self.transf.scale) n_height = int(new_img.height * self.transf.scale) if scale: new_img = new_img.resize((n_width, n_height)) if self.transf.rotation != 90: new_img = new_img.rotate(self.transf.rotation, expand=True) else: new_img = Image.fromarray(np.flipud(np.array(new_img).T)) w, h = new_img.size if trans: x_shift, y_shift = self.transf.translation else: x_shift, y_shift = 0, 0 buffer_img = Image.new(img.mode, (w+abs(x_shift), h+abs(y_shift))) buffer_img.paste(new_img, (x_shift, y_shift)) return buffer_img def update_img(self): if self._data_path is None: return # self.slice_image.img = self.apply_transf(self.slice_image.dw_img) self.pic = np.array(self.apply_transf(self.slice_image.dw_img)) # The old transformation is no longer valid for the inset display and cell selection self.cell_select_cb.setChecked(False) # Erase the previously transformed image to avoid confusion # Now the transformed picture needs to be recomputed by switching into cell selection mode self.raw_inset = np.zeros((5, 5), dtype=np.uint8) self.transf_raw = Image.new('L', (5, 5)) # Useless to keep track of mouse movements too often since we have nothing to display self._proxy.rateLimit = 0.1 def convert_mouse_pos(self, x, y, min_x, min_y): # Original position zoom corrected ex, ey = x, y # Correct for zoom px_width, px_height = self.anat_image.pixelSize() x /= px_width y /= px_height # Inverse transformation from display to data coords = np.intp(pg.transformCoordinates(self.anat_image.inverseDataTransform(), np.array([x, y]))) dx = min_x + coords[0] dy = min_y + coords[1] # Divide dx and dy by the number of the atlas scale (here, 10) dx = dx dy = dy return dx, dy def cell_clicked(self, x, y, mx, my): dx, dy = self.convert_mouse_pos(x, y, mx, my) dx, dy = int(dx), int(dy) # Get structure id if possible c_atlas = self.get_slice(self.raw_atlas, self.c_slice) if dx < 0 or dx >= c_atlas.shape[0] or dy < 0 or dy >= c_atlas.shape[1]: return reg_id = c_atlas[dx, dy] # Get region name l_reg = [(i, r) for i, r in enumerate(self.tree_model.regions) if r.id == reg_id] if len(l_reg) == 0: return else: l_reg = l_reg[0] # id de la structure récupérée reg = l_reg[1] # nom de la structure récupérée parent_reg = get_parent(self.onto, l_reg[0]) if parent_reg == None: parent_reg = str(None) reg_ix = self.tree_model.match(self.tree_model.index(0, 0, QtCore.QModelIndex()), QtCore.Qt.DisplayRole, QtCore.QVariant(reg.abbr), 1, QtCore.Qt.MatchExactly|QtCore.Qt.MatchRecursive) if reg_ix: reg_ix = reg_ix[0] self.tree.setCurrentIndex(reg_ix) if self.cell_select_cb.isChecked(): #setOpacity(laValeurDeLaCase) self.cells.append({'pos': (dx + 1, dy + 1), 'Region': parent_reg, 'structure': str(reg)}) self.cell_pos.append((dx + 1, dy + 1)) self.actions.append(((self.cells.pop, self.cell_pos.pop), (-1, -1))) self.cells = check_duplicate(self.cells) self.cell_scatter.setData(pos = self.cell_pos) self._logger.debug(f'Coordonnées enregistrées: {self.cells[0]}') for i in range(0, len(self.cell_pos) - 1): if (dx + 1, dy + 1) == self.cell_pos[i]: self.cell_pos.remove(self.cell_pos[i]) self.cell_scatter.setData(pos = self.cell_pos) for i in range(0, len(self.cells) - 1): if (dx + 1, dy + 1) == self.cells[i]['pos']: self.cells.remove(self.cells[i]) self._logger.debug("Supression du point dans self.cells réalisée") def zoom_change(self, value): self.slice_image.c_zoom = value self.auto_scale() def atlas_alpha(self, value): self.update_atlas(self.slice_sl.value()) def undo(self): actions, args = self.actions.pop(-1) for a, ix in zip(actions, args): a(ix) self.cell_pos = check_duplicate(self.cell_pos) self.cell_scatter.setData(pos = self.cell_pos) def show_atlas(self): self.atlas_image.setVisible(self.show_atlas_cb.isChecked()) def channel_change(self, value): self.slice_image.c_channel = value self.auto_scale() def z_change(self, value): self.slice_image.c_zslice = value self.auto_scale() def brain_change(self, value): self.slice_image.c_slice = value self._logger.debug('Changing brain slice') self.auto_scale() def luminosity_change(self, value): self.apply_brightness(value) def contrast_change(self, value): self.pic = self.apply_contrast(self.slice_image.img, value) def apply_brightness(self, value): self.anat_image.setLevels((0, self.slice_image.p_max * (2 - value / 50))) @staticmethod def apply_contrast(img, value): def contrast_lut(v): m = 2**16 # Max in 16 bits v /= m # Normalize to 1 # 50 is the middle of the slider, so no extra contrast. # Over 50: exponent from 1 to 2 # Under 50: exponent from 0 to 1 pv = np.clip(np.power(v, value / 50), 0, 1) pv *= m # Go back to original scale return pv pic = np.array(img, dtype=np.float32) return contrast_lut(pic) def next_slice(self): self.slice_sl.setValue(self.c_slice + 1) self.apply_brightness(self.lum_sl.value()) def prev_slice(self): self.slice_sl.setValue(self.c_slice - 1) self.apply_brightness(self.lum_sl.value()) def flip_lr(self): self.transf.flipped_lr = not self.transf.flipped_lr self.apply_brightness(self.lum_sl.value()) def flip_ud(self): self.transf.flipped_ud = not self.transf.flipped_ud self.apply_brightness(self.lum_sl.value()) def rotation(self, value): self.transf.add_rotation(value) self.update_transf_sliders() def translation(self, x_shift, y_shift): x_shift, y_shift = int(-x_shift), int(-y_shift) self.transf.add_translation((x_shift, y_shift)) self.apply_brightness(self.lum_sl.value()) def scaling(self, value): scale = value / self.slice_image.img.height / 5 self.transf.add_scale(scale) self.update_transf_sliders() self.apply_brightness(self.lum_sl.value()) def sl_rot_changed(self, value): self.transf.rotation = value / self.rot_sl.factor self.apply_brightness(self.lum_sl.value()) # def sl_scale_changed(self, value): # if value == 1: # return # self.transf.scale = value / self.scale_sl.factor def update_transf_sliders(self): self.rot_sl.setValue(int(self.transf.rotation))# * self.rot_sl.factor)) # self.scale_sl.setValue(int(self.transf.scale * self.scale_sl.factor)) self.fliplr_pb.setChecked(bool(self.transf.flipped_lr)) self.flipud_pb.setChecked(bool(self.transf.flipped_ud)) @property def transf_raw(self): return self._transf_raw @transf_raw.setter def transf_raw(self, value): self._transf_raw = value self.raw_inset = np.zeros((5, 5), dtype=np.uint8) @property def raw_inset(self): return self._raw_inset @raw_inset.setter def raw_inset(self, value): self._raw_inset = value self.zoom_image.setImage(value) @property def transf(self): return self._transf @transf.setter def transf(self, value): self._transf = value self.transf.rotation_changed.connect(self.update_img) self.transf.translation_changed.connect(self.update_img) self.transf.scale_changed.connect(self.update_img) self.transf.fliplr_changed.connect(self.update_img) self.transf.flipud_changed.connect(self.update_img) self.update_img() self.update_transf_sliders() @property def sel_regions(self): return self._sel_regions @sel_regions.setter def sel_regions(self, value): self._sel_regions = value self.update_atlas(self.slice_sl.value()) @property def pic(self): return self._pic @pic.setter def pic(self, value): self._pic = value self.anat_image.setImage(value) @property def data_path(self): return self._data_path @data_path.setter def data_path(self, value): """ When the data path changes, open the corresponding data file Parameters ---------- value: str Path to the data as selected from the dialog box """ self._data_path = value self.load_image() @property def c_orient(self): return self._c_orient @c_orient.setter def c_orient(self, value): self._c_orient = value self.slice_sl.setRange(0, self.template.shape[value] - 1) def get_slice(self, volume, value): if self.c_orient == 0: s = volume[value, ...] d = (1, 0, 2) if len(s.shape) == 3 else (1, 0) s = s.transpose(d) elif self.c_orient == 1: s = volume[:, value, ...] else: if len(volume.shape) == 3: s = volume[:, :, value] else: s = volume[:, :, value, :] return s def update_atlas(self, value): self.c_slice = value c_atlas = self.get_slice(self.c_atlas, value) s_atlas = self.get_slice(self.atlas, value) c_template = self.get_slice(self.template, value) atlas_alpha = int(self.alpha_sl.value()/100*255) ids = np.unique(s_atlas) alpha = np.zeros(s_atlas.shape) + atlas_alpha alpha[np.isin(s_atlas, list(self.sel_regions))] = 0 gi = np.logical_and(alpha == atlas_alpha, np.any(c_atlas > 0, 2)) c_atlas[gi, ...] = 255 c_atlas = np.dstack((c_atlas, alpha)) self.template_image.setImage(c_template) self.atlas_image.setImage(c_atlas) def change_orientation(self, index): self.c_orient = index self.update_atlas(self.slice_sl.value()) def select_data_file(self): cwd = os.getcwd() dpath, _filter = QtWidgets.QFileDialog.getOpenFileName(self, "Choose an image file", cwd, 'Image (*.tiff *.tif *.vsi *.ndpis)') if dpath != '': # Set the new data path. The data_path setter will take care of the rest self.data_path = dpath # Helping manuel to get shortcuts and tips def help(self): img = cv2.imread("helpManuel.png") cv2.imshow('Help manuel', img) def auto_scale(self): atlas_shape = self.c_atlas.shape[:2] self.actions = [] self.cells = [] self.cell_pos = [] # On fixe le scale à 1 (et on ne transforme plus cette valeur comme ci-dessous) self.transf.scale = 1 # if self.transf.scale != 1: # self.transf.scale = min(atlas_shape[0] / self.slice_image.img.width, # atlas_shape[1] / self.slice_image.img.height) # self.update_transf_sliders() self.update_img() def update_n_slices(self, n_zslices, n_channels, n_res_levels, n_brainslices): # self.z_sl.setRange(0, n_zslices - 1) self.channel_sl.setRange(0, n_channels - 1) # self.zoom_sl.setRange(0, n_res_levels - 1) self.brain_sl.setRange(0, n_brainslices - 1) self._logger.debug(f'Number of z slices: {n_zslices}') def update_max_value(self, p_max): self.p_max = p_max self.apply_brightness(self.lum_sl.value()) def img_updated_event(self): self.pic = self.apply_contrast(self.slice_image.img, self.contrast_sl.value()) self.apply_brightness(self.lum_sl.value()) self._logger.debug(f'Pic max: {self.pic.max()}') self._logger.debug(f'Display range: {self.anat_image.levels}') def load_image(self): """ Load the image defined in self.data_path Loading procedure depends on format. Handles tif through PIL, vsi through python-bioformats """ # Load the imge self.slice_image = SliceImage(self.data_path) self.slice_image.n_slices_known.connect(self.update_n_slices) self.slice_image.img_updated.connect(self.img_updated_event) self.slice_image.max_value_changed.connect(self.update_max_value) self.slice_image.load() # Reset contrast and brightness # self.contrast_sl.setValue(50) # self.lum_sl.setValue(50) if self.slice_image.is_tiff: # self.z_sl.setEnabled(False) self.channel_sl.setEnabled(False) self.scale_sl.setValue(0.2) else: # self.z_sl.setEnabled(True) # self.zoom_sl.setEnabled(True) self.channel_sl.setEnabled(True) # self.z_sl.setValue(self.slice_image.c_zslice) self.channel_sl.setValue(self.slice_image.c_channel) # self.zoom_sl.setValue(self.slice_image.c_zoom) if self.slice_image.is_ndpis: self.brain_sl.setEnabled(True) else: self.brain_sl.setEnabled(False) # Finish initialization self.auto_scale() self.lum_sl.setEnabled(True) self.contrast_sl.setEnabled(True) self.transf_raw = self.slice_image.raw_img def select_region(self, current, previous): c_item = self.tree.model().data(current, QtCore.Qt.UserRole) sel_regions = set([c_item.region.id]) children = set([ch.region.id for ch in self.get_all_children(c_item)]) sel_regions.update(children) self.sel_regions = sel_regions self._logger.debug('Region selected') def get_all_children(self, region): for ch in region.child_items: yield ch if ch.child_items: yield from self.get_all_children(ch) def save_transf(self): if self.data_path is None: return self._logger.debug(self.cells) params = dict(manual=self.transf.params, image=self.data_path, cells=self.cells) params['manual']['atlas_orientation'] = self.c_orient params['manual']['atlas_slice'] = self.slice_sl.value() params['manual']['prefactor'] = self.slice_image.downfactor im_path = Path(self.data_path) path = im_path.parent / f'params_{im_path.stem}.json' with open(path, 'w') as f: json.dump(params, f, indent=4) def load_transf(self): cwd = os.getcwd() dpath, _filter = QtWidgets.QFileDialog.getOpenFileName(self, "Choose a transformation file", cwd, 'JSON (*.json)') if dpath != '': # Load the transformation with open(dpath, 'r') as f: workspace = json.load(f) image = workspace['image'] d_transf = workspace['manual'] l_cells = workspace['cells'] transf = Transform((d_transf['x_shift'], d_transf['y_shift']), d_transf['rotation'], d_transf['scale'], d_transf['flip_lr'], d_transf['flip_ud']) try: self.orientation_cb.setCurrentIndex(d_transf['atlas_orientation']) self.slice_sl.setValue(int(d_transf['atlas_slice'])) except KeyError: pass self.data_path = image self.transf = transf self.cells = l_cells self.cell_pos = [tuple(c['pos']) for c in l_cells] self.cell_scatter.setData(pos=self.cell_pos)
def __init__(self): super().__init__() self.statusBar().showMessage('Loading...') self._logger.info('Loading...') Image.MAX_IMAGE_PIXELS = None # Necessary to open super large tiff files # Load atlas and annotations self.atlas = np.load('contourify2.npy') self.template = get_atlas('average_template_25.nrrd') self.raw_atlas = get_atlas('annotation_25.nrrd') self.colors = id_colors(self.onto) self.c_atlas = np.load('color_atlas.npy') self._logger.info('Atlas opened') self.p_max = 2**16 self._c_orient = 0 self._sel_regions = {} self.orientation_cb.setCurrentIndex(2) self._data_path = None self.slice_image = SliceImage('') self.pic = np.zeros((5, 5), dtype=np.uint8) self._transf_raw = Image.new( 'L', (5, 5)) # Create a greyscale 8-bit image. Size of 5x5 self.transf_raw = self._transf_raw self._raw_inset = np.zeros((5, 5), dtype=np.uint8) self.raw_inset = self._raw_inset self._transf = Transform() self.transf = self._transf self._inset_size = 1000 self.cells = [] self.cell_pos = [] self.actions = [] ### AJOUT PAR LE GROUPE ### self.rot_sl.setValue(90) ########################### self.scale_sl.setValue(int(self.transf.scale * self.scale_sl.factor)) # Capture mouse movements to scroll the inset ### MODIFICATION ET AJOUTS PAR LE GROUPE ### self._proxy = pg.SignalProxy(self.vb_anat.scene().sigMouseMoved, rateLimit=0.1, delay=.1, slot=self.mouse_moved) # On modifie la fenêtre qui repère les mouvements de la souris : c'est la fenêtre de gauche et non plus la fenetre en bas à gauche (qui n'existe plus) self.setWindowTitle('Atlaser Sotfware') # On met un titre à la fenetre self.help_menu = None # On initialise à None le manuel d'aide (qui n'est pas ouvert par défaut) self.myROIdata = [ ] # On initialise la liste stockant les coordonnées sauvegardées avec l'outil de sélection implémenté
class AtlasExplorer(Viewer): def __init__(self): super().__init__() self.statusBar().showMessage('Loading...') self._logger.info('Loading...') Image.MAX_IMAGE_PIXELS = None # Necessary to open super large tiff files # Load atlas and annotations self.atlas = np.load('contourify2.npy') self.template = get_atlas('average_template_25.nrrd') self.raw_atlas = get_atlas('annotation_25.nrrd') self.colors = id_colors(self.onto) self.c_atlas = np.load('color_atlas.npy') self._logger.info('Atlas opened') self.p_max = 2**16 self._c_orient = 0 self._sel_regions = {} self.orientation_cb.setCurrentIndex(2) self._data_path = None self.slice_image = SliceImage('') self.pic = np.zeros((5, 5), dtype=np.uint8) self._transf_raw = Image.new( 'L', (5, 5)) # Create a greyscale 8-bit image. Size of 5x5 self.transf_raw = self._transf_raw self._raw_inset = np.zeros((5, 5), dtype=np.uint8) self.raw_inset = self._raw_inset self._transf = Transform() self.transf = self._transf self._inset_size = 1000 self.cells = [] self.cell_pos = [] self.actions = [] ### AJOUT PAR LE GROUPE ### self.rot_sl.setValue(90) ########################### self.scale_sl.setValue(int(self.transf.scale * self.scale_sl.factor)) # Capture mouse movements to scroll the inset ### MODIFICATION ET AJOUTS PAR LE GROUPE ### self._proxy = pg.SignalProxy(self.vb_anat.scene().sigMouseMoved, rateLimit=0.1, delay=.1, slot=self.mouse_moved) # On modifie la fenêtre qui repère les mouvements de la souris : c'est la fenêtre de gauche et non plus la fenetre en bas à gauche (qui n'existe plus) self.setWindowTitle('Atlaser Sotfware') # On met un titre à la fenetre self.help_menu = None # On initialise à None le manuel d'aide (qui n'est pas ouvert par défaut) self.myROIdata = [ ] # On initialise la liste stockant les coordonnées sauvegardées avec l'outil de sélection implémenté ################################### def switch_cellmode(self, checked): if checked: self.raw_inset = np.zeros((5, 5), dtype=np.uint8) self.transf_raw = self.apply_transf(self.slice_image.raw_img, False, False) self._proxy.rateLimit = 30 self.statusBar().showMessage('Ready for cell selection', 1500) def draw_cross(self): """ Draw the cross cursor on the inset zoom picture Place it at the center of the inset, which depends on the exact size of the self._raw_inset array """ # inset size w, h = self._raw_inset.shape w2, h2 = w // 2, h // 2 # Cross on the inset picture self._h_line_l.setLine(0, h2, w2, h2) self._h_line_r.setLine(w2, h2, w, h2) self._v_line_u.setLine(w2, h2, w2, h) self._v_line_d.setLine(w2, h2, w2, 0) def clear_cells(self): self.cells = [] self.cell_pos = [] self.cell_scatter.setData(self.cell_pos) def clear_transf(self): self.transf = Transform() def mouse_moved(self, pos): """ Update the zoom inset from the raw image when the mouse cursor moves over the low resolution atlas aligned image Parameters ---------- pos: list of SceneEvents Contains the mouse position (in scene coordinates, will need conversion) """ # FIXME: When too close to the edges, not possible to get a precise position if not self.cell_select_cb.isChecked(): return pos = pos[0] v_range = self.vb_anat.viewRange() x, y = self.convert_mouse_pos(pos.x(), pos.y(), v_range[0][0], v_range[1][0]) x, y = x, y all_scales = self.slice_image.downfactor / self.transf.scale raw_x = int((x - self.transf.translation[1]) * all_scales) raw_y = int((y - self.transf.translation[0]) * all_scales) # If the mouse is out of the picture, don't bother if raw_x > self.transf_raw.height or raw_x < 0 or raw_y > self.transf_raw.width or raw_y < 0: return if self.transf_raw.width < 10: return raw_x, raw_y = raw_y, raw_x start_x = np.clip(raw_x - self._inset_size, 0, self.transf_raw.width - self._inset_size) start_y = np.clip(raw_y - self._inset_size, 0, self.transf_raw.height - self._inset_size) stop_x = np.clip(raw_x + self._inset_size, self._inset_size, self.transf_raw.width - 1) stop_y = np.clip(raw_y + self._inset_size, self._inset_size, self.transf_raw.height - 1) self.raw_inset = np.array( self.transf_raw.crop((start_x, start_y, stop_x, stop_y))) self.draw_cross() def closeEvent(self, a0: QtGui.QCloseEvent) -> None: """ closeEvent of Viewer. Perform some cleanup """ a0.accept() super().closeEvent(a0) ### AJOUT PAR LE GROUPE ### def export_points(self): """ Sauvegarde les données stockées dans le dictionnaire cells Save cells dictionnary data """ try: for i in range(0, len(self.cells)): # On parcourt le dictionnaire if (self.cells[i]['Region'] == 'None' ): # Certaines régions stockées n'ont pas de parent d = defaultdict( str ) # Initialisation du dictionnaire stockant les string (librairie collections) d[1] = "" # Le parent 1 (colonne 1) prend par défaut la valeur vide d[2] = "" # Idem pour la deuxième colonne d = dict(d) # On convertit d en dictionnaire classique self.cells[i].update( d ) # On ajoute à cells, pour chaque structure, seulement les deux derniers parents self.cells[i]['structure'] = self.cells[i].pop( 'structure' ) # On retire la clé structure qui correspond à toute l'arborescence de la structure del self.cells[i][ 'Region'] # On supprime la clé région car on n'a plus besoin de celle-ci pour la sauvegarde else: d = defaultdict( str ) # Initialisation du dictionnaire stockant les string (librairie collections) d[1] = self.cells[i]['Region'][ len(self.cells[i]['Region']) - 2] # On récupère le parent le plus proche de la structure enregistrée dans cells d[2] = self.cells[i]['Region'][ len(self.cells[i]['Region']) - 3] # On récupère le parent de la structure stockée dans d[1] d = dict(d) # idem qu'avant self.cells[i].update(d) # idem qu'avant self.cells[i]['structure'] = self.cells[i].pop( 'structure') # idem qu'avant del self.cells[i]['Region'] # idem qu'avant im_path = Path( self.data_path) # On initialise le chemin de sauvegarde path = im_path.parent / f'cells_{im_path.stem}.csv' # On définit le chemin de sauvegarde et le nom du fichier à la sortie # Le fichier aura le même nom que l'image # et est sauvegardé dans le même dossier que celle-ci (souhait de l'utilisateur) keys = self.cells[i].keys( ) # On récupère les clés du dictionnaire cells mis-à-jour au-dessus with open( path, 'a', newline='' ) as file: # Pour chaque élément de cells, on écrit dans le fichier CSV. # Le mode a permet d'ajouter dans un fichier déjà existant dict_writer = csv.DictWriter(file, keys) dict_writer.writerows(self.cells) self.cells = [] # On vide le dictionnaire après une sauvegarde. except KeyError: # On évite les erreurs de clés pouvant arriver avec des régions sans parent pass ########################## # Displays the image in the right windows when cell mode selection is activated def apply_transf(self, img, scale=True, trans=True): new_img = img.copy() if self.transf.flipped_lr: new_img = Image.fromarray(np.flipud(np.array(new_img))) if self.transf.flipped_ud: new_img = Image.fromarray(np.fliplr(np.array(new_img))) n_width = int(new_img.width * self.transf.scale) n_height = int(new_img.height * self.transf.scale) if scale: new_img = new_img.resize((n_width, n_height)) if self.transf.rotation != 90: new_img = new_img.rotate(self.transf.rotation, expand=True) else: new_img = Image.fromarray(np.flipud(np.array(new_img).T)) w, h = new_img.size if trans: x_shift, y_shift = self.transf.translation else: x_shift, y_shift = 0, 0 buffer_img = Image.new(img.mode, (w + abs(x_shift), h + abs(y_shift))) buffer_img.paste(new_img, (x_shift, y_shift)) return buffer_img def update_img(self): if self._data_path is None: return self.pic = np.array(self.apply_transf(self.slice_image.dw_img)) # The old transformation is no longer valid for the inset display and cell selection self.cell_select_cb.setChecked(False) # Erase the previously transformed image to avoid confusion # Now the transformed picture needs to be recomputed by switching into cell selection mode self.raw_inset = np.zeros((5, 5), dtype=np.uint8) self.transf_raw = Image.new('L', (5, 5)) # Useless to keep track of mouse movements too often since we have nothing to display self._proxy.rateLimit = 0.1 def convert_mouse_pos(self, x, y, min_x, min_y): # Original position zoom corrected ex, ey = x, y # Correct for zoom px_width, px_height = self.anat_image.pixelSize() x /= px_width y /= px_height # Inverse transformation from display to data coords = np.intp( pg.transformCoordinates(self.anat_image.inverseDataTransform(), np.array([x, y]))) dx = min_x + coords[0] dy = min_y + coords[1] # Divide dx and dy by the number of the atlas scale (here, 10) dx = dx dy = dy return dx, dy ### MODIFICATION PAR LE GROUPE ### def cell_clicked(self, x, y, mx, my): dx, dy = self.convert_mouse_pos(x, y, mx, my) dx, dy = int(dx), int(dy) # Get structure id if possible c_atlas = self.get_slice(self.raw_atlas, self.c_slice) if dx < 0 or dx >= c_atlas.shape[0] or dy < 0 or dy >= c_atlas.shape[1]: return reg_id = c_atlas[dx, dy] # Get region name l_reg = [(i, r) for i, r in enumerate(self.tree_model.regions) if r.id == reg_id] if len(l_reg) == 0: return else: l_reg = l_reg[0] # id de la structure récupérée reg = l_reg[1] # nom de la structure récupérée ### AJOUT PAR LE GROUPE ### parent_reg = get_parent( self.onto, l_reg[0]) # On récupère la liste de tous les parents if parent_reg == None: # S'il n'y a pas de parents, on convertit en string None pour le stockage dans le dictionnaire de string parent_reg = str(None) ########################### reg_ix = self.tree_model.match( self.tree_model.index(0, 0, QtCore.QModelIndex()), QtCore.Qt.DisplayRole, QtCore.QVariant(reg.abbr), 1, QtCore.Qt.MatchExactly | QtCore.Qt.MatchRecursive) if reg_ix: reg_ix = reg_ix[0] self.tree.setCurrentIndex(reg_ix) ### AJOUT ET MODIFICATIONS PAR LE GROUPE ### if self.cell_select_cb.isChecked( ): # Quand le mode de sélection est activé self.cells.append({ 'pos': (dx + 1, dy + 1), 'Region': parent_reg, 'structure': str(reg) }) # On stocke les coordonnées (+1 pour pouvoir tracer à l'écran sur le pixel choisi), la région et son arborescence et la structure cliquée (nom) self.cell_pos.append( (dx + 1, dy + 1) ) # On ajoute les coordonnées à la liste des points dessinés à l'écran self.actions.append( ((self.cells.pop, self.cell_pos.pop), (-1, -1))) self.cells = check_duplicate( self.cells) # On contrôle les doublons dans cells self.cell_scatter.setData( pos=self.cell_pos) # On dessine les ronds enregistrés # On contrôle qu'il n'y ait pas de doublons dessinés for i in range(0, len(self.cell_pos) - 1): if (dx + 1, dy + 1) == self.cell_pos[i]: self.cell_pos.remove(self.cell_pos[i]) self.cell_scatter.setData(pos=self.cell_pos) # On contrôle qu'il n'y ait pas de doublons stockés dans le dictionnaire cells qui va être exporté à la fin for i in range(0, len(self.cells) - 1): if (dx + 1, dy + 1) == self.cells[i]['pos']: self.cells.remove(self.cells[i]) ############################################## def zoom_change(self, value): self.slice_image.c_zoom = value self.auto_scale() def atlas_alpha(self, value): self.update_atlas(self.slice_sl.value()) ### AJOUT PAR LE GROUPE ### def undo(self): """ Suppression des doublons affichés à l'écran. Se fait automatiquement quand on dessine un autre point que celui qui est en double. Accessible avec CTRL + Z. A utiliser en fin de manipulation pour avoir un dictionnaire de données propre et sans doublons Delete duplicates drawn circles on screen. It is automatic when the user clicks on another area that the one with duplicates. Can be used manually with CTRL + Z. Should be used when finishing the cells selection to clean data that will be save """ actions, args = self.actions.pop(-1) for a, ix in zip(actions, args): a(ix) self.cell_pos = check_duplicate( self.cell_pos ) # On check les doublons dans la liste contenant les points dessinés à l'écran et on les supprime self.cell_scatter.setData( pos=self.cell_pos) # On redessine tout le dictionnaire (update) ################################### def show_atlas(self): self.atlas_image.setVisible(self.show_atlas_cb.isChecked()) def channel_change(self, value): self.slice_image.c_channel = value self.auto_scale() def z_change(self, value): self.slice_image.c_zslice = value self.auto_scale() def brain_change(self, value): self.slice_image.c_slice = value self._logger.debug('Changing brain slice') self.auto_scale() def luminosity_change(self, value): self.apply_brightness(value) def contrast_change(self, value): self.pic = self.apply_contrast(self.slice_image.img, value) self.apply_brightness(self.lum_sl.value()) def apply_brightness(self, value): self.anat_image.setLevels( (0, self.slice_image.p_max * (2 - value / 50))) # ## AJOUT PAR LE GROUPE ### # def update(self, roi): # #List that contains the position of the ROI # roidata = [] # roidata.append(roi.pos()) # #times 8 for each position to negate the downscaled tiff # roidata.append(roi.size() * 8) # self.myROIdata = roidata # print (self.myROIdata) # def select_tool(self): # # Création de l'outil de selection, sur l'image et d'une taille correcte # myROI = pg.ROI([1000,1000], [350,500]) # # Ajout de l'outil sur la fenetre qui contient l'image # self.vb_anat.addItem(myROI) # # Gère le scaling horizontal # myROI.addScaleHandle([1, 0.5], [0.5, 0.5]) # myROI.addScaleHandle([0, 0.5], [0.5, 0.5]) # # Gère le scaling vertical # myROI.addScaleHandle([0.5, 0], [0.5, 1]) # myROI.addScaleHandle([0.5, 1], [0.5, 0]) # # Gère le scaling horizontal et vertical # myROI.addScaleHandle([1, 1], [0, 0]) # myROI.addScaleHandle([0, 0], [1, 1]) # myROI.sigRegionChanged.connect(self.update) # self.update(myROI) ############################# @staticmethod def apply_contrast(img, value): def contrast_lut(v): m = 2**16 # Max in 16 bits v /= m # Normalize to 1 # 50 is the middle of the slider, so no extra contrast. # Over 50: exponent from 1 to 2 # Under 50: exponent from 0 to 1 pv = np.clip(np.power(v, value / 50), 0, 1) pv *= m # Go back to original scale return pv pic = np.array(img, dtype=np.float32) return contrast_lut(pic) ### AJOUT PAR LE GROUPE ### def next_slice(self): self.slice_sl.setValue(self.c_slice + 1) self.apply_brightness( self.lum_sl.value() ) # Evite le problème de reset de la valeur en appliquant la valeur actuelle de luminosité def prev_slice(self): self.slice_sl.setValue(self.c_slice - 1) self.apply_brightness( self.lum_sl.value() ) # Evite le problème de reset de la valeur en appliquant la valeur actuelle de luminosité def flip_lr(self): self.transf.flipped_lr = not self.transf.flipped_lr self.apply_brightness( self.lum_sl.value() ) # Evite le problème de reset de la valeur en appliquant la valeur actuelle de luminosité def flip_ud(self): self.transf.flipped_ud = not self.transf.flipped_ud self.apply_brightness( self.lum_sl.value() ) # Evite le problème de reset de la valeur en appliquant la valeur actuelle de luminosité def rotation(self, value): self.transf.add_rotation(value) self.update_transf_sliders() def translation(self, x_shift, y_shift): x_shift, y_shift = int(-x_shift), int(-y_shift) self.transf.add_translation((x_shift, y_shift)) self.apply_brightness( self.lum_sl.value() ) # Evite le problème de reset de la valeur en appliquant la valeur actuelle de luminosité def scaling(self, value): scale = value / self.slice_image.img.height / 5 self.transf.add_scale(scale) self.update_transf_sliders() self.apply_brightness( self.lum_sl.value() ) # Evite le problème de reset de la valeur en appliquant la valeur actuelle de luminosité def sl_rot_changed(self, value): self.transf.rotation = value / self.rot_sl.factor self.apply_brightness( self.lum_sl.value() ) # Evite le problème de reset de la valeur en appliquant la valeur actuelle de luminosité ################################## def update_transf_sliders(self): self.rot_sl.setValue(int(self.transf.rotation)) self.fliplr_pb.setChecked(bool(self.transf.flipped_lr)) self.flipud_pb.setChecked(bool(self.transf.flipped_ud)) @property def transf_raw(self): return self._transf_raw @transf_raw.setter def transf_raw(self, value): self._transf_raw = value self.raw_inset = np.zeros((5, 5), dtype=np.uint8) @property def raw_inset(self): return self._raw_inset @raw_inset.setter def raw_inset(self, value): self._raw_inset = value self.zoom_image.setImage(value) @property def transf(self): return self._transf @transf.setter def transf(self, value): self._transf = value self.transf.rotation_changed.connect(self.update_img) self.transf.translation_changed.connect(self.update_img) self.transf.scale_changed.connect(self.update_img) self.transf.fliplr_changed.connect(self.update_img) self.transf.flipud_changed.connect(self.update_img) self.update_img() self.update_transf_sliders() @property def sel_regions(self): return self._sel_regions @sel_regions.setter def sel_regions(self, value): self._sel_regions = value self.update_atlas(self.slice_sl.value()) @property def pic(self): return self._pic @pic.setter def pic(self, value): self._pic = value self.anat_image.setImage(value) @property def data_path(self): return self._data_path @data_path.setter def data_path(self, value): """ When the data path changes, open the corresponding data file Parameters ---------- value: str Path to the data as selected from the dialog box """ self._data_path = value self.load_image() @property def c_orient(self): return self._c_orient @c_orient.setter def c_orient(self, value): self._c_orient = value self.slice_sl.setRange(0, self.template.shape[value] - 1) def get_slice(self, volume, value): if self.c_orient == 0: s = volume[value, ...] d = (1, 0, 2) if len(s.shape) == 3 else (1, 0) s = s.transpose(d) elif self.c_orient == 1: s = volume[:, value, ...] else: if len(volume.shape) == 3: s = volume[:, :, value] else: s = volume[:, :, value, :] return s def update_atlas(self, value): self.c_slice = value c_atlas = self.get_slice(self.c_atlas, value) s_atlas = self.get_slice(self.atlas, value) c_template = self.get_slice(self.template, value) atlas_alpha = int(self.alpha_sl.value() / 100 * 255) ids = np.unique(s_atlas) alpha = np.zeros(s_atlas.shape) + atlas_alpha alpha[np.isin(s_atlas, list(self.sel_regions))] = 0 gi = np.logical_and(alpha == atlas_alpha, np.any(c_atlas > 0, 2)) c_atlas[gi, ...] = 255 c_atlas = np.dstack((c_atlas, alpha)) self.template_image.setImage(c_template) self.atlas_image.setImage(c_atlas) def change_orientation(self, index): self.c_orient = index self.update_atlas(self.slice_sl.value()) def select_data_file(self): cwd = os.getcwd() dpath, _filter = QtWidgets.QFileDialog.getOpenFileName( self, "Choose an image file", cwd, 'Image (*.tiff *.tif *.vsi *.ndpis)') if dpath != '': # Set the new data path. The data_path setter will take care of the rest self.data_path = dpath ### AJOUT PAR LE GROUPE ### # Helping manuel to get shortcuts and tips def help(self): """ Ouvre une image contenant les raccourcis du logiciel utilisables par l'utilisateur Open an image with all the shortcuts the user can use """ img = cv2.imread("helpManuel.png") # ouverture de l'image manuel cv2.imshow('Help manuel', img) # affichage de l'image à l'écran ############################ def auto_scale(self): atlas_shape = self.c_atlas.shape[:2] self.actions = [] self.cells = [] self.cell_pos = [] self.transf.scale = 1 self.update_img() def update_n_slices(self, n_zslices, n_channels, n_res_levels, n_brainslices): self.channel_sl.setRange(0, n_channels - 1) self.brain_sl.setRange(0, n_brainslices - 1) self._logger.debug(f'Number of z slices: {n_zslices}') def update_max_value(self, p_max): self.p_max = p_max self.apply_brightness(self.lum_sl.value()) def img_updated_event(self): self.pic = self.apply_contrast(self.slice_image.img, self.contrast_sl.value()) self.apply_brightness(self.lum_sl.value()) self._logger.debug(f'Pic max: {self.pic.max()}') self._logger.debug(f'Display range: {self.anat_image.levels}') def load_image(self): """ Load the image defined in self.data_path Loading procedure depends on format. Handles tif through PIL, vsi through python-bioformats """ # Load the imge self.slice_image = SliceImage(self.data_path) self.slice_image.n_slices_known.connect(self.update_n_slices) self.slice_image.img_updated.connect(self.img_updated_event) self.slice_image.max_value_changed.connect(self.update_max_value) self.slice_image.load() # Reset contrast and brightness if self.slice_image.is_tiff: self.channel_sl.setEnabled(False) self.scale_sl.setValue(0.2) else: self.channel_sl.setEnabled(True) self.channel_sl.setValue(self.slice_image.c_channel) if self.slice_image.is_ndpis: self.brain_sl.setEnabled(True) else: self.brain_sl.setEnabled(False) # Finish initialization self.auto_scale() self.lum_sl.setEnabled(True) self.contrast_sl.setEnabled(True) self.transf_raw = self.slice_image.raw_img def select_region(self, current, previous): c_item = self.tree.model().data(current, QtCore.Qt.UserRole) sel_regions = set([c_item.region.id]) children = set([ch.region.id for ch in self.get_all_children(c_item)]) sel_regions.update(children) self.sel_regions = sel_regions self._logger.debug('Region selected') def get_all_children(self, region): for ch in region.child_items: yield ch if ch.child_items: yield from self.get_all_children(ch) def save_transf(self): if self.data_path is None: return self._logger.debug(self.cells) params = dict(manual=self.transf.params, image=self.data_path, cells=self.cells) params['manual']['atlas_orientation'] = self.c_orient params['manual']['atlas_slice'] = self.slice_sl.value() params['manual']['prefactor'] = self.slice_image.downfactor im_path = Path(self.data_path) path = im_path.parent / f'params_{im_path.stem}.json' with open(path, 'w') as f: json.dump(params, f, indent=4) def load_transf(self): cwd = os.getcwd() dpath, _filter = QtWidgets.QFileDialog.getOpenFileName( self, "Choose a transformation file", cwd, 'JSON (*.json)') if dpath != '': # Load the transformation with open(dpath, 'r') as f: workspace = json.load(f) image = workspace['image'] d_transf = workspace['manual'] l_cells = workspace['cells'] transf = Transform((d_transf['x_shift'], d_transf['y_shift']), d_transf['rotation'], d_transf['scale'], d_transf['flip_lr'], d_transf['flip_ud']) try: self.orientation_cb.setCurrentIndex( d_transf['atlas_orientation']) self.slice_sl.setValue(int(d_transf['atlas_slice'])) except KeyError: pass self.data_path = image self.transf = transf self.cells = l_cells self.cell_pos = [tuple(c['pos']) for c in l_cells] self.cell_scatter.setData(pos=self.cell_pos)