医学图像预处理----StatisticalNormalization

基于统计的窗口化操作，根据图像的统计信息，例如图像均值作为窗口中心，正负2.5（这个值并非固定）的方差作为窗口宽度

sitk.NormalizeImageFilter()通过将其均值设置为零并将方差设置为 1 来标准化图像。

sitk.StatisticsImageFilter()计算 Image 的最小值、最大值、方差和平均值。

sitk.IntensityWindowingImageFilter()将线性变换应用于用户定义的间隔内的输入图像的强度级别。 低于此间隔的值被映射到一个常数。 区间内的值被映射到另一个常量。

class StatisticalNormalization(object):
	"""
	Normalize an image by mapping intensity with intensity distribution
	"""

	def __init__(self, sigma, pre_norm=False):
		self.name = 'StatisticalNormalization'
		assert isinstance(sigma, float)
		self.sigma = sigma
		self.pre_norm=pre_norm

	def __call__(self, sample):
		image, label = sample['image'], sample['label']

		for image_channel in range(len(image)):
			if self.pre_norm:
				normalFilter= sitk.NormalizeImageFilter()
				image[image_channel] = normalFilter.Execute(image[image_channel])

			statisticsFilter = sitk.StatisticsImageFilter()
			statisticsFilter.Execute(image[image_channel])

			intensityWindowingFilter = sitk.IntensityWindowingImageFilter()
			intensityWindowingFilter.SetOutputMaximum(255)
			intensityWindowingFilter.SetOutputMinimum(0)
			intensityWindowingFilter.SetWindowMaximum(statisticsFilter.GetMean()+self.sigma*statisticsFilter.GetSigma());
			intensityWindowingFilter.SetWindowMinimum(statisticsFilter.GetMean()-self.sigma*statisticsFilter.GetSigma());

			image[image_channel] = intensityWindowingFilter.Execute(image[image_channel])

		return {'image': image, 'label': label}