Research for ConvU_NExT: Dataset Containing normal dose CT images for training(Head, Lungs, Abdominal and Knee procedures), LDCT images for test, and Water Phantom Test images for augmentation

WaterPhantomImages.zip contains one series sample of the images resulted from the water phantom test. These images were used to augment the normal dose CT image to simulate LDCT while training an encoder-decoder model for denoising.

CTImages.zip contais 3 more folders with samples for: 

Test - native LDCT images used for the final measurements of model's performance

Train - normal dose CT images from multiple procedures (Abdominal, Head, Knee, Lungs), each in it's individual folder

Validation - normal dose CT, where beam hardening effect is present

Images provided in the dataset are in tiff format, and they have float32 data type. The conversion code from dicom to tiff is provided below: 

def get_first_of_dicom_field_as_int(x):
if type(x) == dcm.multival.MultiValue: return int(x[0])
else: return int(x)

def get_windowing(data):
dicom_fields = [data[('0028','1050')].value, #window center
data[('0028','1051')].value, #window width
data[('0028','1052')].value, #intercept
data[('0028','1053')].value] #slope
return [get_first_of_dicom_field_as_int(x) for x in dicom_fields]

def window_image(img, window_center=50, window_width=400, slope=1, intercept=-1024):
img = (img*slope + intercept) #for translation adjustments given in the dicom file.
img = np.array(img).astype('int16')
img_min = window_center - window_width//2 #minimum HU level
img_max = window_center + window_width//2 #maximum HU level
img[imgimg_max] = img_max #set img_max for all HU levels higher than maximum HU level
img = (img - img_min) / (img_max - img_min)
img = (img).astype('float32')
return img

for partial_path in os.listdir(input_path):
path = os.path.join(input_path, partial_path)
data = dcm.read_file(path)
img = data.pixel_array
window_center , window_width, intercept, slope = get_windowing(data)
img = window_image(img, window_center, window_width, slope, intercept)
tiff.imwrite(os.path.join(output_path, f'{i}.tiff'), img)