def dicom(path, save_path):
print("dicom start!")
#mritopng.convert_folder(r'C:\Users\user\PycharmProjects\dicom', path)
png_list = []
dcm_list = []
for file in [doc for doc in os.listdir(path) if doc.endswith(".png")]:
png_list.append(file)
for file2 in [doc for doc in os.listdir(path) if doc.endswith(".dcm")]:
dcm_list.append(file2)
new_dir = path + '\\' + 'new'
if not (os.path.isdir(new_dir)):
os.makedirs(os.path.join(new_dir))
for k in range(len(dcm_list)):
dicom_path = path + '\\' + dcm_list[k]
shutil.move(dicom_path, new_dir)
mritopng.convert_folder(new_dir, save_path)
for i in range(len(png_list)):
png_path = path + '\\' + png_list[i]
shutil.move(png_path, save_path)
print(png_path + ' success')
def test_folder_convert(self):
curr_path = os.path.dirname(os.path.realpath(__file__))
mri_dir_path = os.path.join(curr_path, 'data', 'samples')
png_dir_path = os.path.join(curr_path, 'data', 'png_dir')
mritopng.convert_folder(mri_dir_path, png_dir_path)
valid_files = {'dicom1.png', '000012.dcm.png', '000017.dcm.png'}
converted_files = set(os.listdir(png_dir_path))
# cleanup before assertion.
shutil.rmtree(png_dir_path)
self.assertEqual(valid_files, converted_files, 'Invalid dicom left a stray png file.')
folder_path = r"D:\FYP\imageConvert\dicom_image\C3N-04611\02-11-2011-K Neck ContrastAdult-21611\4-Neck Native 200 Br40 S3-09851"
destinate_path = 'D:/FYP/imageConvert/dicom_image/C3N-04611/02-11-2011-K Neck ContrastAdult-21611/4-Neck Native 200 Br40 S3-09851/PNG/'
def convertImg(dicomImage):
print("Image: " + dicomImage)
outputImg = dicomImage.split('.')
mritopng.convert_file(os.path.join(folder_path, dicomImage),
os.path.join(folder_path, outputImg[0] + '.jpg'),
auto_contrast=True)
convertedImg = os.path.join(folder_path, outputImg[0] + '.jpg')
return convertedImg
mritopng.convert_folder(folder_path, destinate_path)
def rename():
i = 0
for filename in os.listdir(folder_path):
dst = str(i) + ".png"
src = destinate_path + filename + ".png"
dst = destinate_path + dst
# rename() function will
# rename all the files
os.rename(src, dst)
i += 1
#Convert dicoms to png (without mode equalization)
import mritopng
# Convert a since file
mritopng.convert_folder('/home/anvit/Desktop/Data/RSNA/stage1/',
'/home/anvit/Desktop/Data/RSNA/images/')
# Convert a whole folder recursively
#mritopng.convert_folder('/home/user/DICOM/', '/home/user/PNG/')
def acrelicfunc(ls=alst):
fname = ls[0]
dcmfile = []
dcmlist = []
fiducoords = []
voxeli = []
mritopng.convert_folder(fname, fname + '/png')
os.chdir(fname)
for file in os.listdir():
if 'png' not in file.lower():
dcmlist.append(dicom.read_file(file))
else:
pass
pnglist = []
os.chdir(fname + '\png')
for file in os.listdir():
pnglist.append(file)
# pnglist.pop()
for file in pnglist:
img = cv2.imread(file, 0)
orig = img.copy()
shape = img.shape
_, img = cv2.threshold(img, 195, 255, cv2.THRESH_BINARY)
_, contours, _ = cv2.findContours(img, cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
objs = list(filter(lambda x: cv2.contourArea(x) > 800, contours))
if img is not None:
img.fill(0)
else:
img = np.zeros(shape, np.uint8)
if len(objs) == 1 and cv2.contourArea(objs[0]) > 3000: # side view
# print("Block")
obj = objs[0]
approx = cv2.approxPolyDP(obj, 0.02 * cv2.arcLength(obj, True),
True)
x, y, w, h = cv2.boundingRect(approx)
img = cv2.drawContours(img, [obj], 0, 1, -1)
img[y - 2:y + h + 2, x - 2:x + w + 2] = 0
_, contours, _ = cv2.findContours(img, cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
boxes = [cv2.boundingRect(cnt) for cnt in contours]
boxes = list(
filter(lambda x: x[2] * x[3] < 200 and x[2] * x[3] > 20,
boxes))
vert_groups = []
horz_groups = []
for i in range(0, len(boxes) - 1):
for j in range(i + 1, len(boxes)):
x1, y1, _, _ = boxes[i]
x2, y2, _, _ = boxes[j]
dist_x = abs(x1 - x2)
dist_y = abs(y1 - y2)
if dist_x > 20 and dist_x < 40 and dist_y < 5:
horz_groups.append((boxes[i], boxes[j]))
elif dist_y > 20 and dist_y < 40 and dist_x < 5:
vert_groups.append((boxes[i], boxes[j]))
fids = []
for group in horz_groups:
x1, y1, w1, _ = group[0]
x2, y2, w2, _ = group[1]
gap_dist = min(abs(x2 - x1 - w1), abs(x1 - x2 - w2))
if gap_dist >= 15:
fids.append(((x1 + x2 + gap_dist) // 2, (y1 + y2) // 2))
for group in vert_groups:
x1, y1, _, h1 = group[0]
x2, y2, _, h2 = group[1]
gap_dist = min(abs(y2 - y1 - h1), abs(y1 - y2 - h2))
if gap_dist >= 15:
fids.append(((x1 + x2) // 2, (y1 + y2 + gap_dist) // 2))
else: # either top view or noise
# print("Circles")
fids = []
for obj in objs:
(x, y), r = cv2.minEnclosingCircle(obj)
actual_area = cv2.contourArea(obj)
ideal_area = 3.14 * r * r
perc_filled = actual_area / ideal_area
if r > 10 and r < 35 and perc_filled > 0.9:
fids.append((int(x), int(y)))
print(str(len(fids)) + " fiducials detected.")
print(fids)
print(file)
if len(fids) != 0:
print(file)
ds = dcmlist[pnglist.index(file)]
vector = ds.ImageOrientationPatient
pixelsize = ds.PixelSpacing
voxel = ds.ImagePositionPatient
for k in fids:
if abs(vector[0]) == 1:
if abs(vector[4]) == 1:
x = k[0]
y = k[1]
plane = 'xy'
voxelx = voxel[0] + vector[0] * pixelsize[0] * x
voxely = voxel[1] + vector[4] * pixelsize[1] * y
voxelz = voxel[2]
else:
x = k[0]
z = k[1]
plane = 'xz'
voxelx = voxel[0] + vector[0] * pixelsize[0] * x
voxely = voxel[1]
voxelz = voxel[2] + vector[5] * pixelsize[1] * z
else:
y = k[0]
z = k[1]
plane = 'yz'
voxelx = voxel[0]
voxely = voxel[1] + vector[1] * pixelsize[1] * y
voxelz = voxel[2] + vector[5] * pixelsize[0] * z
print(plane, (voxelx, voxely, voxelz))
fiducoords.append((voxelx, voxely, voxelz))
voxeli.append(voxelx)
print(fiducoords)
print(len(fiducoords))
correct = []
for i in fiducoords:
check = True
for j in fiducoords[fiducoords.index(i):]:
dist1 = abs((i[0] - j[0]))
dist2 = abs(i[1] - j[1])
dist3 = abs((i[2] - j[2]))
if dist1 < 1 and dist2 < 1 and dist3 < 1:
check = False
if check:
correct.append(i)
print(len(fiducoords), fiducoords)
print(len(correct), correct)
for (a, b, c) in correct:
print(sqrt(a * a + b * b + c * c))
def getKey(item):
return fiducoords[0]
sorted(fiducoords, key=getKey)
print(fiducoords)
def mainFunc(ls=lst):
for fname in ls:
print(fname)
mritopng.convert_folder(fname, fname + '/png')
dcmfile = []
dcmlist = []
fiducoords = []
os.chdir(fname)
for file in os.listdir():
if 'png' not in file.lower():
dcmlist.append(dicom.read_file(file))
else:
pass
pnglist = []
os.chdir(fname + '/png')
for file in os.listdir():
pnglist.append(file)
pnglist.pop()
for file in pnglist:
img = cv2.imread(file, 0)
# The algorithm is good for images of size 256x256. So we have to scale
# the input image down, and then scale the result back up.
dimensions = img.shape
scale = dimensions[0] // 256
img = cv2.resize(img, (256, 256))
# Skull detection
copy = img.copy()
img = cv2.medianBlur(img, 5)
img = cv2.inRange(img, 20, 255)
_, contours, _ = cv2.findContours(img.copy(), cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
skull = max(contours, key=cv2.contourArea)
M = cv2.moments(skull)
skull_cent = (M['m10'] // M['m00'], M['m01'] // M['m00'])
mask = cv2.drawContours(np.zeros(img.shape), [skull], 0, 1, 20)
mask = np.uint8(mask)
img = cv2.multiply(copy, mask)
# Detect points that are outside the head,
# but close to the bones of the skull
img_diag = (img.shape[0]**2 + img.shape[1]**2)**0.5
img = cv2.linearPolar(img, skull_cent, img_diag,
cv2.WARP_FILL_OUTLIERS)
img = cv2.inRange(img, 120, 255)
_, contours, _ = cv2.findContours(img, cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
bones = [cnt for cnt in contours if cv2.contourArea(cnt) > 30]
img = np.uint8(
cv2.drawContours(np.zeros(copy.shape), bones, -1, 1, -1))
temp = cv2.reduce(img, 0, cv2.REDUCE_MAX)
temp = cv2.findNonZero(temp)
leftmost = temp[temp[:, :, 0].argmin()][0][0]
rightmost = temp[temp[:, :, 0].argmax()][0][0]
mask = np.zeros(copy.shape, np.uint8)
for row in range(img.shape[0]):
for column in range(rightmost, leftmost, -1):
if img[row, column] != 0:
mask[row, column + 2:column + 15] = 1
break
mask = cv2.linearPolar(mask, skull_cent, img_diag,
cv2.WARP_INVERSE_MAP)
img = cv2.multiply(copy, mask)
img = cv2.inRange(img, 160, 255)
# Group nearby points and print output
_, contours, _ = cv2.findContours(img, cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
cents = []
for cnt in contours:
M = cv2.moments(cnt)
if (M['m00'] != 0):
cents.append((M['m10'] // M['m00'], M['m01'] // M['m00']))
else:
cents.append(tuple(cnt[0][0]))
groups = []
for pnt1 in cents:
added = False
for group in groups:
if not added:
for pnt2 in group:
approx_dist = abs(pnt1[0] -
pnt2[0]) + abs(pnt1[1] - pnt2[1])
if approx_dist < 25:
group.append(pnt1)
added = True
break
if not added:
groups.append([pnt1])
fiducials = []
for group in groups:
sum = (0, 0)
for pnt in group:
sum = (sum[0] + pnt[0], sum[1] + pnt[1])
sum = (int(sum[0] / len(group)), int(sum[1] / len(group)))
fiducials.append(sum)
output = copy.copy()
for pnt in fiducials:
cv2.circle(output, pnt, 10, 255, 2)
# Resize the image back to thr original dimensions before grouping
output = cv2.resize(output, dimensions)
# Scale up the coordinates of the fiducials
temp = np.multiply(fiducials, 2)
fiducials = [tuple(x) for x in temp]
#print(str(len(fiducials)), " Fiducials detected:")
#print(fiducials)
if len(fiducials) != 0:
print(file)
ds = dcmlist[pnglist.index(file)]
vector = ds.ImageOrientationPatient
pixelsize = ds.PixelSpacing
voxel = ds.ImagePositionPatient
for k in fiducials:
if abs(vector[0]) == 1:
if abs(vector[4]) == 1:
x = k[0]
y = k[1]
plane = 'xy'
voxelx = voxel[0] + vector[0] * pixelsize[0] * x
voxely = voxel[1] + vector[4] * pixelsize[1] * y
voxelz = voxel[2]
else:
x = k[0]
z = k[1]
plane = 'xz'
voxelx = voxel[0] + vector[0] * pixelsize[0] * x
voxely = voxel[1]
voxelz = voxel[2] + vector[5] * pixelsize[2] * z
else:
y = k[0]
z = k[1]
plane = 'yz'
voxelx = voxel[0]
voxely = voxel[1] + vector[1] * pixelsize[1] * y
voxelz = voxel[2] + vector[5] * pixelsize[0] * z
#print(plane, (voxelx, voxely, voxelz))
fiducoords.append((voxelx, voxely, voxelz))
# iske pehle ka code for all 3 planes, iske baad ka code bas ek baar chlna h last me
# below is the code for elimination of fidu
finalfidu = 0
count = 0
captainswing = []
print(fiducoords)
fiducoords = sorted(fiducoords, key=lambda x: x[0])
correct = []
a = [fiducoords[0]]
for i in fiducoords:
check = True
for j in fiducoords[fiducoords.index(i) + 1:]:
dist1 = abs((i[0] - j[0]))
dist2 = abs(i[1] - j[1])
dist3 = abs((i[2] - j[2]))
if dist1 < 3 and dist2 < 3 and dist3 < 3:
print(fiducoords.index(j))
check = False
break
if check:
correct.append(i)
print(len(fiducoords), fiducoords)
# print(fiducoords)
for (a, b, c) in correct:
print(sqrt(a * a + b * b + c * c))
def convert_to_png(path1, path2):
mritopng.convert_folder(path1, path2)
import mritopng
# Convert a since file
#mritopng.convert_file('E:/setups/MRI/MR/1.2.840.113619.2.5.1762583153.215519.978957063.79/1.2.840.113619.2.5.1762583153.215519.978957063.80.dcm', 'E:/setups/MRI/MRI_out/output.png')
# Convert a whole folder recursively
mritopng.convert_folder('E:/setups/MRI/MRI_sample/', 'E:/setups/MRI/MRI_out1/')