def run(self):
# Network for Segmentation
self.send_msg.emit('FCNN_V2 Initializing...')
time_1 = datetime.now()
if not self.is_run:
self.fcnn = CRF_RNN(input_shape=[512, 512, 1],
batch_size=1,
input_dim=262144,
pre_train=True)
time_2 = datetime.now() - time_1
self.send_msg.emit('FCNN_V2 Initialize Finished.\nTime Using:' +
str(time_2))
reader = dcm_reader(path=self.path)
arr = self.process(reader.getPixelArray())
slices, rows, cols = arr.shape
data_slices = list()
for i in range(slices):
flatten = arr[i, :, :].reshape((1, rows * cols)).astype(np.float32)
pred = self.fcnn.predict(flatten, as_list=True)[0]
pred = np.array(list(map(list, zip(*pred[::-1]))))
pred = pred.transpose() * 255.0
Image.fromarray(pred).convert('RGB').save(image_path + str(i) +
'.png')
data_slices.append(pred[None, :, :])
self.signal.emit(int(i * 100 / slices))
self.data = np.concatenate(data_slices, axis=0)
time_3 = datetime.now() - time_1
self.send_msg.emit('Thread Finished.\nTime Cost: ' + str(time_3))
def run(self):
self.send_msg.emit('Decision Tree Initializing...')
time_1 = datetime.now()
self.dt = joblib.load(model_path + 'model.m')
self.haar = haarReader()
self.haar.setRandomWindowsFromFile(model_path + 'rands.npy')
time_2 = datetime.now() - time_1
self.send_msg.emit('Decision Tree Initialize Finished.\nTime Using:' +
str(time_2))
self.reader = dcm_reader(path=self.path)
arr = self.reader.getPixelArray()
slices, rows, cols = arr.shape
data_slices = list()
for i in range(slices):
flatten = self.haar.getFeature(arr[i, :, :])
pred = self.dt.predict(flatten).reshape((512, 512))
data_slices.append(pred[None, :, :])
self.signal.emit(int(i * 100 / slices))
self.data = np.concatenate(data_slices, axis=0)
time_3 = datetime.now() - time_1
self.send_msg.emit('Thread Finished.\nTime Cost: ' + str(time_3))
def _initialize(self, path):
self.reader = dcm_reader(path)
pixLabels = [self.ui.png1, self.ui.png2, self.ui.png3]
chart = dict()
for key, val in self.reader.data.items():
to_path = '\\'.join(key.split('\\')[:-1]) + '\\' + str(
val.SeriesNumber) + '\\'
if str(val.SeriesNumber) not in chart:
# 如果已经有上次使用过的文件夹还存在，没有被删除的话，就先删掉
if str(val.SeriesNumber) in os.listdir(
'\\'.join(key.split('\\')[:-1]) + '\\'):
shutil.rmtree(to_path)
# 新建文件夹
os.makedirs(to_path)
self.to_del.append(to_path)
# 保存图片到本地
try:
img_to_save = self._dicom_normalize(val.pixel_array)
Image.fromarray(img_to_save).convert('RGB').save(
str(val.SeriesNumber) + '.png')
chart[str(
val.SeriesNumber)] = str(val.SeriesNumber) + '.png'
except:
print('Something wrong')
continue
shutil.copy(key, to_path + key.split('\\')[-1])
# 把上面保存在本地的图片文件渲染到界面上
for index, pic in enumerate(chart.values()):
if index >= 3:
break
png = QtGui.QPixmap(pic)
png = png.scaled(175, 175, QtCore.Qt.IgnoreAspectRatio,
QtCore.Qt.SmoothTransformation)
pixLabels[index].setPixmap(png)
self.ui.statusBar.showMessage('Initialize Finished')
def run(self):
# Network for Segmentation
self.send_msg.emit('FCNN Initializing...')
time_1 = datetime.now()
if self.is_run:
self.fcnn = FCNet(learning_rate=1e-4)
self.fcnn.load(
path=
'E:\\Python\\surgeryGuidingProject\\deep_learning\\network\\')
time_2 = datetime.now() - time_1
self.send_msg.emit('FCNN Initialize Finished.\nTime Using:' +
str(time_2))
reader = dcm_reader(path=self.path)
# arr = reader.getPixelArray()
arr = self.process(reader.getPixelArray())
slices, rows, cols = arr.shape
data_slices = list()
for i in range(slices):
flatten = arr[i, :, :].reshape((1, rows * cols)).astype(np.float32)
pred = self.fcnn.predict(flatten)
pred = np.array(list(map(list, zip(*pred[::-1]))))
pred = pred.transpose()
# pred = np.array(list(map(list, zip(*pred[::-1]))))
# levelset = LevelSet()
# levelset.init_phi(pred)
# levelset.evolution(iter_num=10, showed=False)
# mask = levelset.get_contour(dtype=np.uint8)
# data_slices.append(mask[None, :, :])
Image.fromarray(pred).convert('RGB').save(image_path + str(i) +
'.png')
data_slices.append(pred[None, :, :])
self.signal.emit(int(i * 100 / slices))
self.data = np.concatenate(data_slices, axis=0)
time_3 = datetime.now() - time_1
self.send_msg.emit('Thread Finished.\nTime Cost: ' + str(time_3))
def evolution_once(self):
self._phi = self.sess.run(self.evolution_op,
feed_dict={self.phi_op: self._phi})
def show_contour(self):
rgb_src = np.array(Image.fromarray(self.src[50, :, :]).convert('RGB'))
mask = np.zeros_like(self._phi[50, :, :]).astype(np.int32)
mask[self._phi[50, :, :] > 0] = 255
contours = measure.find_contours(mask, level=0.5)
plt.figure()
plt.imshow(rgb_src)
for ct in contours:
plt.plot(ct[:, 1], ct[:, 0], linewidth=2, color='c')
plt.show()
# for ct in contours:
# rgb_src[ct[:, 0], ct[:, 1]] = (0, 255, 255)
# Image.fromarray(rgb_src).save('6.png')
print('Evolution result saved!')
if __name__ == '__main__':
reader = dcm_reader(
path=
'E:\\C++\\Projects\\surgeryGuidingProject_copy\\8848\\new\DICOM\\20170510\\08290000\\6\\'
)
arr = reader.getPixelArray()
arr = reader.dicom_normalize(arr)
levelset3 = LevelSet()
levelset3.init_phi(arr)
levelset3.evolution()