def getMaskFromCenterline(image, pointset, spacing, radius=10):
data = npy.zeros(image.shape, dtype=npy.uint8)
image_size = image[0, :, :].transpose().shape
n_pt = 20
for cnt in range(3):
temp_point = pointset[npy.where(npy.round(pointset[:, -1]) == cnt)]
if not temp_point.shape[0]:
continue
zmin = int(npy.min(temp_point[:, 2]) + 0.5)
zmax = int(npy.max(temp_point[:, 2]) + 0.5)
for z in range(zmin, zmax + 1):
center_point = temp_point[npy.where(
npy.round(temp_point[:, 2]) == z)]
if center_point is not None:
if center_point.shape[0] == 0:
continue
data_point = npy.zeros([n_pt, 2], dtype=npy.float32)
for i in range(0, n_pt):
data_point[i, 0] = center_point[0, 0] + radius * npy.cos(
2 * npy.pi * i / n_pt) / spacing[0]
data_point[i, 1] = center_point[0, 1] + radius * npy.sin(
2 * npy.pi * i / n_pt) / spacing[1]
fixPoints(data_point, image_size)
data[z, :, :] = data[z, :, :] | ac_mask(
data_point.transpose(), image_size).transpose()
result = npy.cast['float32'](data)
del data
return result
def getBinaryImageFromSegmentation(image, pointset):
data = npy.zeros(image.shape, dtype=npy.uint8)
image_size = image[0, :, :].transpose().shape
for cnt in range(3):
temp_point = pointset[npy.where(npy.round(pointset[:, -1]) == cnt)]
if not temp_point.shape[0]:
continue
zmin = int(npy.min(temp_point[:, 2]) + 0.5)
zmax = int(npy.max(temp_point[:, 2]) + 0.5)
for z in range(zmin, zmax + 1):
data_point = temp_point[npy.where(
npy.round(temp_point[:, 2]) == z)]
if data_point is not None:
if data_point.shape[0] == 0:
continue
ind = sortContourPoints(data_point)
data_point = data_point[ind]
data_point = data_point[:, :2].transpose()
data[z, :, :] = data[z, :, :] | ac_mask(
data_point, image_size).transpose()
result = npy.cast['float32'](data)
del data
return result
def getBinaryImageFromSegmentation(image, pointset):
data = npy.zeros(image.shape, dtype = npy.uint8)
image_size = image[0, :, :].transpose().shape
for cnt in range(3):
temp_point = pointset[npy.where(npy.round(pointset[:, -1]) == cnt)]
if not temp_point.shape[0]:
continue
zmin = int(npy.min(temp_point[:, 2]) + 0.5)
zmax = int(npy.max(temp_point[:, 2]) + 0.5)
for z in range(zmin, zmax + 1):
data_point = temp_point[npy.where(npy.round(temp_point[:, 2]) == z)]
if data_point is not None:
if data_point.shape[0] == 0:
continue
ind = sortContourPoints(data_point)
data_point = data_point[ind]
data_point = data_point[:, :2].transpose()
data[z, :, :] = data[z, :, :] | ac_mask(data_point, image_size).transpose()
result = npy.cast['float32'](data)
del data
return result
def getMaskFromCenterline(image, pointset, spacing, radius = 10):
data = npy.zeros(image.shape, dtype = npy.uint8)
image_size = image[0, :, :].transpose().shape
n_pt = 20
for cnt in range(3):
temp_point = pointset[npy.where(npy.round(pointset[:, -1]) == cnt)]
if not temp_point.shape[0]:
continue
zmin = int(npy.min(temp_point[:, 2]) + 0.5)
zmax = int(npy.max(temp_point[:, 2]) + 0.5)
for z in range(zmin, zmax + 1):
center_point = temp_point[npy.where(npy.round(temp_point[:, 2]) == z)]
if center_point is not None:
if center_point.shape[0] == 0:
continue
data_point = npy.zeros([n_pt, 2], dtype = npy.float32)
for i in range(0, n_pt):
data_point[i, 0] = center_point[0, 0] + radius * npy.cos(2 * npy.pi * i / n_pt) / spacing[0]
data_point[i, 1] = center_point[0, 1] + radius * npy.sin(2 * npy.pi * i / n_pt) / spacing[1]
fixPoints(data_point, image_size)
data[z, :, :] = data[z, :, :] | ac_mask(data_point.transpose(), image_size).transpose()
result = npy.cast['float32'](data)
del data
return result
def calCenterFromContour(pointset, image):
mask = ac_mask(pointset[:, :2].transpose(), image.transpose().shape).transpose()
dist = cv2.distanceTransform(mask, cv.CV_DIST_L2, 3)
value = npy.max(dist)
indx, indy = npy.where(dist == npy.max(dist))
indx = npy.cast['float32'](indx)
indy = npy.cast['float32'](indy)
center = npy.array([[npy.mean(indy), npy.mean(indx)]])
return center
def calIntensityCentroidFromContour(pointset, image):
mask = ac_mask(pointset[:, :2].transpose(), image.transpose().shape).transpose()
result = image * mask
cnt = npy.sum(mask)
m00 = npy.sum(result) / cnt
xx, yy = npy.mgrid[:image.shape[0], :image.shape[1]]
m10 = npy.sum(xx * result) / cnt
m01 = npy.sum(yy * result) / cnt
center = npy.array([[m01 / m00, m10 / m00]])
return center
def calCenterFromContour(pointset, image):
mask = ac_mask(pointset[:, :2].transpose(),
image.transpose().shape).transpose()
dist = cv2.distanceTransform(mask, cv.CV_DIST_L2, 3)
value = npy.max(dist)
indx, indy = npy.where(dist == npy.max(dist))
indx = npy.cast['float32'](indx)
indy = npy.cast['float32'](indy)
center = npy.array([[npy.mean(indy), npy.mean(indx)]])
return center
def calIntensityCentroidFromContour(pointset, image):
mask = ac_mask(pointset[:, :2].transpose(),
image.transpose().shape).transpose()
result = image * mask
cnt = npy.sum(mask)
m00 = npy.sum(result) / cnt
xx, yy = npy.mgrid[:image.shape[0], :image.shape[1]]
m10 = npy.sum(xx * result) / cnt
m01 = npy.sum(yy * result) / cnt
center = npy.array([[m01 / m00, m10 / m00]])
return center
def analysis(self, data, point_data_fix=None, area=False):
if point_data_fix is None:
point_data_fix = self.gui.dataModel[
data.getFixedIndex()].getPointSet('Contour').copy()
point_data_result = data.getPointSet('Contour').copy()
image_size = image = data.getData()[0, :, :].transpose().shape
self.spacing = data.getResolution().tolist()
self.spacing[
2] = 1.0 # The resolution of z axis is nothing to do with the analysis
point_data_fix[:, :3] *= self.spacing[:3]
point_data_result[:, :3] *= self.spacing[:3]
cnt_num = npy.array([0, 0, 0])
union_area = npy.array([0.0, 0.0, 0.0])
total_area = npy.array([0.0, 0.0, 0.0])
if area:
mr_area = [0.0, 0.0, 0.0, 0.0]
us_area = [0.0, 0.0, 0.0, 0.0]
for cnt in range(3):
temp_result = point_data_result[npy.where(
npy.round(point_data_result[:, -1]) == cnt)]
temp_fix = point_data_fix[npy.where(
npy.round(point_data_fix[:, -1]) == cnt)]
if not temp_result.shape[0] or not temp_fix.shape[0]:
continue
zmin = int(
npy.min([npy.min(temp_result[:, 2]),
npy.min(temp_fix[:, 2])]) + 0.5)
zmax = int(
npy.max([npy.max(temp_result[:, 2]),
npy.max(temp_fix[:, 2])]) + 0.5)
for z in range(zmin, zmax + 1):
data_fix = temp_fix[npy.where(npy.round(temp_fix[:, 2]) == z)]
data_result = temp_result[npy.where(
npy.round(temp_result[:, 2]) == z)]
if data_fix is not None and data_result is not None:
if data_fix.shape[0] == 0 or data_result.shape[0] == 0:
continue
cnt_num[cnt] += 1
#center_fix = npy.mean(data_fix[:, :2], axis = 0)
center_fix = calCentroidFromContour(data_fix[:, :2])[0]
#center_result = npy.mean(data_result[:, :2], axis = 0)
center_result = calCentroidFromContour(
data_result[:, :2])[0]
points_fix = getPointsOntheSpline(
data_fix, center_fix, data_fix.shape[0] * 10)[:, :2]
points_result = getPointsOntheSpline(
data_result, center_result,
data_result.shape[0] * 10)[:, :2]
fix_mask = ac_mask(points_fix.transpose(), image_size)
result_mask = ac_mask(points_result.transpose(),
image_size)
temp_fix_area = npy.sum(fix_mask)
temp_result_area = npy.sum(result_mask)
total_area[cnt] += temp_fix_area + temp_result_area
union_area[cnt] += npy.sum(fix_mask | result_mask)
if area:
mr_area[cnt] += temp_fix_area
mr_area[3] += temp_fix_area
us_area[cnt] += temp_result_area
us_area[3] += temp_result_area
intersect_area = total_area - union_area
jaccard_index = intersect_area / union_area
dice_index = 2 * intersect_area / total_area
jaccard_index_all = npy.sum(intersect_area) / npy.sum(union_area)
dice_index_all = 2 * npy.sum(intersect_area) / npy.sum(total_area)
# Replace the NAN
jaccard_index[jaccard_index != jaccard_index] = 0
dice_index[dice_index != dice_index] = 0
if self.gui is not None:
message = "Jaccard Index on Vessel 0: %0.3f\nJaccard Index on Vessel 1: %0.3f\nJaccard Index on Vessel 2: %0.3f\nTotal Jaccard Index: %0.3f\n" \
% (jaccard_index[0], jaccard_index[1], jaccard_index[2], jaccard_index_all) + \
"-------------------------------------------------------\n" + \
"Dice Index on Vessel 0: %0.3f\nDice Index on Vessel 1: %0.3f\nDice Index on Vessel 2: %0.3f\nTotal Dice Index: %0.3f" \
% (dice_index[0], dice_index[1], dice_index[2], dice_index_all)
self.gui.showErrorMessage("Registration Area Index", message)
if not area:
return dice_index, dice_index_all
else:
for i in range(3):
mr_area[i] /= cnt_num[i]
us_area[i] /= cnt_num[i]
mr_area[3] /= npy.sum(cnt_num)
us_area[3] /= npy.sum(cnt_num)
return dice_index, dice_index_all, mr_area, us_area